File size: 83,474 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 | {
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T13:05:05.674104Z"
},
"title": "Supporting terminology extraction with dependency parses",
"authors": [
{
"first": "Ma\u0142gorzata",
"middle": [],
"last": "Marciniak",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Polish Academy of Sciences",
"location": {
"addrLine": "Jana Kazimierza 5",
"postCode": "01-248",
"settlement": "Warsaw",
"country": "Poland"
}
},
"email": ""
},
{
"first": "Piotr",
"middle": [],
"last": "Rychlik",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Polish Academy of Sciences",
"location": {
"addrLine": "Jana Kazimierza 5",
"postCode": "01-248",
"settlement": "Warsaw",
"country": "Poland"
}
},
"email": "rychlik@ipipan.waw.pl"
},
{
"first": "Agnieszka",
"middle": [],
"last": "Mykowiecka",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Polish Academy of Sciences",
"location": {
"addrLine": "Jana Kazimierza 5",
"postCode": "01-248",
"settlement": "Warsaw",
"country": "Poland"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Terminology extraction procedure usually consists of selecting candidates for terms and ordering them according to their importance for the given text or set of texts. Depending on the method used, a list of candidates contains different fractions of grammatically incorrect, semantically odd and irrelevant sequences. The aim of this work was to improve term candidate selection by reducing the number of incorrect sequences using a dependency parser for Polish.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Terminology extraction procedure usually consists of selecting candidates for terms and ordering them according to their importance for the given text or set of texts. Depending on the method used, a list of candidates contains different fractions of grammatically incorrect, semantically odd and irrelevant sequences. The aim of this work was to improve term candidate selection by reducing the number of incorrect sequences using a dependency parser for Polish.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Extracting important domain related phrases is a part of very many NLP tasks such as information extraction, indexing or text classification. Depending on a particular scenario either more precise or more robust solutions are preferable. In our terminology extraction work, the aim is to prepare preliminary lists for building terminology resources or text indexing. As manual checking of the prepared list is expensive, we are interested in a solution in which the top of the ordered candidates list is of the highest quality. One of the problems of all term extraction methods is the fact that some extracted sequences are incorrect. The sequences recognized using statistical methods or shallow grammars can sometimes be semantically odd or even incorrect at the syntactic level. We identify two types of errors. In the first, the shallow patterns cover only part of the phrase, e.g., resolution microscopy. In the second, parts of two independent phrases are merged into a sequence which does not form a coherent phrase, e.g., high resolution microscopy designed. The aim of this work was to improve term candidate selection by reducing the number of incorrect sequences using a dependency parser for Polish. The answer to the question whether using a deep parser improves term identification would have been evident if the parsing were perfect. In such a case, at least all syntactically incorrect phrases (the errors of the second type mentioned above) would have been eliminated. However, errors of the first type are rather hard to identify on syntactic grounds. Dependency analysis classifies all modifiers as adjuncts, some of them are necessary term parts and indicate a particular subtype, e.g., basic income, while others are just modifications which specify frequency, intensity or quality features and do not constitute a part of a term, e.g., bigger income. That is why we propose a hybrid approach, not just dependency parsing. In this paper, we will not discuss the computational aspects of dependency parsing. Although it can significantly slow down the extraction process, it might still be useful in cases where the potential user wants to improve the quality of the output. Besides, not all sentences of the processed text need to be analyzed by a dependency parser, but only those containing examined terms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Terminology extraction (sometimes under the name of keyword/keyphrase extraction) is quite a popular NLP task which is tackled by several tools available both as open access and commercial systems. An overview of biomedical terminology extraction is presented in (Lossio-Ventura et al., 2016) , several keyphrase extraction systems described in the scientific literature were later presented in (Merrouni et al., 2019) . The latter paper mainly describes solutions which were proposed within the area of text mining or artificial intelligence, while quite a lot of other approaches were proposed at more natural language processing and terminology extraction oriented venues, e.g., TermSuite (Cram and Daille, 2016) and Sketch Engine (Kilgarriff et al., 2014) . Competitions in automatic term extractions have been also organised, e.g., at SemEval workshop (Kim et al., 2010) or (Augenstein et al., 2017) . Terminology extraction systems can be divided into two groups. In one group, term extraction is treated as any other extraction task and is usually solved as a classification task using statistical, e.g., CRF, (Zhang, 2008) , (Yu et al., 2012) , or deep learning methods, e.g., (Zhang et al., 2016) , (Meng et al., 2017) . The other approach, also accepted by the extraction tool we use (TermoPL), comes from collocation/phrase recognition work. Most of the term extraction systems which were developed along these lines follow the standard three phase procedure consisting of text preprocessing, potential term selection and term scoring. Text preprocessing depends on the source of texts and the language in which they are written and usually consists of filtering out unnecessary information, tokenization and sometimes POS tagging. As a lot of work was done for English, most approaches for candidate selections are based on selecting just word n-grams on the basis of the simple frequency based statistics, e.g., (Rose et al., 2010) or on the shallow grammars usually written as a set of regular expressions over POS tags, e.g., (Cram and Daille, 2016) . Deep syntactic grammars are hardly used at all. One solution in which dependency grammar is used to extract term candidates is described in Gamallo (2017) . Dependency parses were also analyzed in Liu et al. (2018) . All the above approaches to candidate selection are approximate (for different reasons), i.e. some term candidates are improper while others are omitted. In our work, we used shallow grammars with additional specification of morphological values dependencies. As Polish is an inflectional language, this approach allows a lot of grammatically incorrect phrases to be filtered out while, at the same time, it is not limited to the sequences recognized properly by a deep parser for Polish, which for a specific domain might not have enough coverage.",
"cite_spans": [
{
"start": 263,
"end": 292,
"text": "(Lossio-Ventura et al., 2016)",
"ref_id": "BIBREF9"
},
{
"start": 395,
"end": 418,
"text": "(Merrouni et al., 2019)",
"ref_id": "BIBREF14"
},
{
"start": 692,
"end": 715,
"text": "(Cram and Daille, 2016)",
"ref_id": "BIBREF2"
},
{
"start": 734,
"end": 759,
"text": "(Kilgarriff et al., 2014)",
"ref_id": "BIBREF5"
},
{
"start": 857,
"end": 875,
"text": "(Kim et al., 2010)",
"ref_id": "BIBREF6"
},
{
"start": 879,
"end": 904,
"text": "(Augenstein et al., 2017)",
"ref_id": "BIBREF0"
},
{
"start": 1117,
"end": 1130,
"text": "(Zhang, 2008)",
"ref_id": "BIBREF21"
},
{
"start": 1133,
"end": 1150,
"text": "(Yu et al., 2012)",
"ref_id": "BIBREF19"
},
{
"start": 1185,
"end": 1205,
"text": "(Zhang et al., 2016)",
"ref_id": "BIBREF20"
},
{
"start": 1208,
"end": 1227,
"text": "(Meng et al., 2017)",
"ref_id": "BIBREF13"
},
{
"start": 1925,
"end": 1944,
"text": "(Rose et al., 2010)",
"ref_id": "BIBREF16"
},
{
"start": 2041,
"end": 2064,
"text": "(Cram and Daille, 2016)",
"ref_id": "BIBREF2"
},
{
"start": 2207,
"end": 2221,
"text": "Gamallo (2017)",
"ref_id": "BIBREF4"
},
{
"start": 2264,
"end": 2281,
"text": "Liu et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2."
},
{
"text": "The second step of the process -candidate ranking -is also carried out in very different ways. The frequency of a term or frequency based coefficients play the most prominent role. The most popular is tf-idf, but the C-value (Frantzi et al., 2000) , used in this paper, also became widely used. Unlike many other coefficients, the C-value takes into account not only the longest phrases or sequences of a given length, but also sequences included in other, longer, sequences.",
"cite_spans": [
{
"start": 225,
"end": 247,
"text": "(Frantzi et al., 2000)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2."
},
{
"text": "Although in some approaches the ranking procedure may be very complex, the idea of an additional phase of filtering improperly built pre-selected phrases, as suggested in our paper, is not very popular. There are, however, some solutions with a post filtering phrase, e.g. (Liu et al., 2015) , in which the candidates are compared to different external terminology resources. This approach was not adopted in our work, as it cannot be used to identify new terms and it requires resources adequate for a specific domain. Another postulated modification of the overall processing schema is the final re-ranking procedure adopted in (Gamallo, 2017) .",
"cite_spans": [
{
"start": 273,
"end": 291,
"text": "(Liu et al., 2015)",
"ref_id": "BIBREF7"
},
{
"start": 630,
"end": 645,
"text": "(Gamallo, 2017)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2."
},
{
"text": "As in many other NLP tasks, evaluation of the terminology extraction results is both crucial and hard to perform. Evaluation can either be performed manually or automatically. In the first case, apart from the cost of the evaluation, the main problem is that sometimes it is hard to judge whether a particular term is domain related or comes from general language. Automatic evaluation requires terminological resources (which, even if they exist, are usually not complete), or preparing the gold standard labelled text (which has similar problems to direct manual evaluation). In statistical methods, the automatic evaluation procedure is usually used. In (Merrouni et al., 2019) , the results of several systems show the overall very poor recall (0.12-0.5) and a little higher precision (0.25-0.7) with the F1 measure usually below 0.3. Manual verification usually covers the top few hundred terms which are judged by a domain expert to be domain related terms or not. In this approach, only the precision of the results can be evaluated at reasonable cost. Gamallo (2017) reports precision of 0.93 for the first 800 terms extracted from English biomedical texts using an approach similar to that adopted by us. In (Marciniak and Mykowiecka, 2014) , the then existing version of the Ter-moPL gave precision of 0.85 for 800 top positions of the terms list obtained from medical clinical reports. The recall counted on four reports (a very small dataset) was 0.8. The poorer results obtained for Polish data are mainly caused by the poor quality of text with many errors and missing punctuation marks (both commas and dots).",
"cite_spans": [
{
"start": 657,
"end": 680,
"text": "(Merrouni et al., 2019)",
"ref_id": "BIBREF14"
},
{
"start": 1060,
"end": 1074,
"text": "Gamallo (2017)",
"ref_id": "BIBREF4"
},
{
"start": 1217,
"end": 1249,
"text": "(Marciniak and Mykowiecka, 2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2."
},
{
"text": "The results of the two groups of methods described above cannot be directly compared, but the good quality of the linguistically based methods is the reason why we want to develop this approach to terminology extraction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2."
},
{
"text": "As the baseline method of term selection for our experiments we chose one implemented in the publicly available tool -TermoPL (Marciniak et al., 2016) . The tool operates on the text tagged with POS and morphological features values and uses shallow grammar to select the term candidates. Grammar rules operate on forms, lemmas and morphological tags of the tokens. They thus allow for imposing agreement requirements important for recognizing phrase borders in inflectional languages, such as Polish. TermoPL has a built-in grammar describing basic Polish noun phrases and also allows for defining custom grammars for other types of phrases. The program was originally developed for the Polish language so it is capable of handling the relatively complex structural tagset of Polish (Przepi\u00f3rkowski et al., 2012) . It is also possible to redefine this tagset and process texts in other languages. To eliminate sub-sequences with borders crossing strong collocations, the NPMI (Bouma, 2009) based method of identifying the proper sub-sequences was proposed (Marciniak and Mykowiecka, 2015) . According to this method, subphrase borders are subsequently identified between the tokens with the smallest NPMI coefficient (counted for bigrams on the basis of the whole corpus). So, if a bigram constitutes a strong collocation, the phrase is not being divided in this place, and this usually blocks creation of semantically odd nested phrases. The final list of terms is ordered according to the C-value adapted for taking one word terms into account. The Cvalue is a frequency dependent coefficient but takes into account not only the occurrences of the longest phrase, but also counts occurrences of its sub-sequences.",
"cite_spans": [
{
"start": 126,
"end": 150,
"text": "(Marciniak et al., 2016)",
"ref_id": "BIBREF12"
},
{
"start": 784,
"end": 813,
"text": "(Przepi\u00f3rkowski et al., 2012)",
"ref_id": "BIBREF15"
},
{
"start": 977,
"end": 990,
"text": "(Bouma, 2009)",
"ref_id": "BIBREF1"
},
{
"start": 1057,
"end": 1089,
"text": "(Marciniak and Mykowiecka, 2015)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "TermoPL",
"sec_num": "3.1."
},
{
"text": "In our experiments we use a publicly available Polish dependency parser -COMBO (Rybak and Wr\u00f3blewska, 2018) . COMBO is a neural net based jointly trained tagger, lemmatizer and dependency parser. It assigns labels based on features extracted by a biLSTM network. The system uses both fully connected and dilated convolutional neural architectures. The parser is trained on the Polish Dependency Bank (http://zil.ipipan.waw.pl/PDB). In our work we used the version trained on PDB annotated with a set of relations extended specifically for Polish (http://zil.ipipan.waw.pl/PDB/DepRelTypes).",
"cite_spans": [
{
"start": 79,
"end": 107,
"text": "(Rybak and Wr\u00f3blewska, 2018)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "COMBO",
"sec_num": "3.2."
},
{
"text": "The experiment was conducted on the textual part of an economics articles taken from Polish Wikipedia. It was collected in 2011 as part of the Polish Nekst project (POIG.01.01.02-14-013/10). The data contains 1219 articles that have economics related headings and those linked to them. The data was processed by the Concraft tagger (Waszczuk, 2012) which uses Morfeusz morphological dictionary and a guesser module for unknown words. The processed text has about 460K tokens in around 20,000 sentences. There are about 46,600 different token types of 17,900 different lemmas or known word forms within the text. Figure 1: The built-in grammar represents noun phrases comprised of nominal phrases built from nouns or gerunds optionally modified by adjectival phrases located either before or after them. Nominal phrases can be modified by any number of nominal phrases in the genitive.",
"cite_spans": [
{
"start": 332,
"end": 348,
"text": "(Waszczuk, 2012)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Data Description",
"sec_num": "4."
},
{
"text": "A selection of candidate phrases is performed by a shallow grammar defined over lemmatized and morphologically annotated text. TermoPL recognizes the maximal sequences of tokens which meet the conditions set out in a grammar. The built-in grammar, see Fig. 1 , recognizes noun phrases where the head element can by modified by adjectives appearing before or after it, such as mi\u0119dzynarodowe stosunki gospodarcze 'international economic relations'. All these elements must agree in number, case and gender, which is marked in the rules by the agreement parameter. The noun phrase can be modified by another noun phrase in the genitive, e.g., ubezpieczenie [odpowiedzialno\u015bci cywilnej] gen 'insurance of civil responsibility'. All these elements can be combined, e.g., samodzielny publiczny zak\u0142ad [opieki zdrowotnej] gen 'independent public health care'. The $ character marks a token or a group of tokens which should be replaced by their nominal forms when base forms are generated. It does not affect the type of phrase being recognized. In the economics texts, the built-in grammar collects 61,966 phrases when the NPMI driven selection method is used (without the NPMI it collects 82,930 phrases). The built-in grammar does not cover noun phrases modified by prepositional phrases which quite often create important terms, e.g., sp\u00f3\u0142ka z ograniczon\u0105 odpowiedzialno\u015bci\u0105 'limited liability company'. This decision was made because it was difficult to recognize the role of a prepositional phrase in a sentence. A phrase very similar to the one above, e.g., umowa sp\u00f3\u0142ki z uniwersytetem 'a company agreement with the university' (word for word translation: ('agreement' 'company' 'with' 'university') should not lead to a conclusion that sp\u00f3\u0142ka z uniwersytetem creates a term -both nouns firma 'company' and uniwersytet 'university' are complements of the noun umowa 'agreement'. If the only criterion is a shallow grammar, we are unable to distinguish between such uses. When analyzing the results obtained by the grammar defined in Fig. 1 , we realised that some nominal phrases can have a noun phrase complement in the instrumental case. It applies to phrases such as, e.g., handel [rop\u0105 naftow\u0105] instr 'trading in petroleum', gospodarka nieruchomo\u015bciami instr 'management of real estate' opodatkowanie [podatkiem dochodowym] instr 'taxation of income'. But a similar problem, as for prepositional phrases, occurs for noun complements in the instrumental case, as we don't know if they are complements of a preceding nominal phrase or if they refer to another element in the sentence. For example: rz\u0105d ob\u0142o\u017cy\u0142 [papierosy] [akcyz\u0105] instr (word for word translation: 'government' 'charged' 'cigarettes' 'excise duty') 'the government charged cigarettes with excise duty', where akcyz\u0105 'excise duty' is the complement of ob\u0142o\u017cy\u0142 'charged' and not papierosy 'cigarettes'. Both constructions described above, i.e. prepositional modifiers and noun complements in the instrumental case, are taken into account in the grammar given in Fig. 2 . It collects 72,758 phrases when the NPMI driven selection method is used, which is over 10,000 more than for the built-in grammar (without the NPMI the grammar collects 113,687 phrases). Although the number of new terms is high, there are a couple of new top candidates on our list. The top 100 terms contains three correct phrases with prepositional modifiers sp\u00f3\u0142ka z ograniczon\u0105 odpowiedzialno\u015bci\u0105 'limited liability company', ustawa o rachunkowo\u015bci 'accounting act' and podatek od towar\u00f3w 'tax on goods', and no term with a noun complement in the instrumental case. The first such phrase prawo o publicznym obrocie papierami warto\u015bciowymi 'law on public trading of securities' is in position 637. As we wanted to know how productive the above grammatical constructions are, we have defined two grammars describing them alone. This allows us to check how many phrases might be introduced to the term candidate list by these constructions. The first dedicated grammar (NPPP) defines nominal phrases with a prepositional modifier. It consists of all rules given in Fig. 2 except the first, third and the seventh one. When the NPMI method is used, the grammar selects 22,150 terms. We evaluate all phrases which occurred at least 2 times and have a C-value of at least 3.0, i.e. 686 phrases. The results are given in Tab 1 -66.6% of them are correct phrases (i.e. for these phrases precision is 0.666), 16.5% are phrases where a preposition phrase does not modify the preceding noun phrase, and for 16.9% a reason for not accepting the phrase is different. Many incorrect phrases are incomplete, such as r\u00f3\u017cnica mi\u0119dzy sum\u0105 przychod\u00f3w uzyskanych 'difference between the sum of revenues obtained' which is a part of r\u00f3\u017cnica mi\u0119dzy sum\u0105 przychod\u00f3w uzyskanych z tytu\u0142u . . . a kosztami uzyskania przychod\u00f3w 'difference between the sum of revenues from . . . and tax deductible costs'. The second grammar (NPInst) defines nominal phrases modified by noun phrases in the instrumental case. It consists of all rules given in Fig. 2 except the first and the second one. It selects fewer phrases, namely 1390. As there was only 44 phrases with a C-value of at least 3.0, we evaluated all 110 phrases which occurred at least twice. The results are given in Tab. 3. The example of an incorrect phrase recognised by the grammar is budownictwo kosztorysantem ('architecture' 'estimator') which actually is built up from two phrases and occurred three times in sentences similar to the following: w [budownictwie kosztorysantem] jest rzeczoznawca 'in architecture, an appraiser is an estimator'. Tab. 4 gives the frequency of the evaluated phrases. The statistics show that such constructions are not common. Moreover, we observe that only a small number of nouns and gerunds (acting as nouns) were used to create valid phrases in our data. These are: obr\u00f3t 'trading' (21), zarz\u0105dzanie 'management' (21), handel 'trade' (9), opodatkowanie 'taxation' (5), gospodarka/gospodarowanie'management' (4). ",
"cite_spans": [],
"ref_spans": [
{
"start": 252,
"end": 258,
"text": "Fig. 1",
"ref_id": null
},
{
"start": 2035,
"end": 2041,
"text": "Fig. 1",
"ref_id": null
},
{
"start": 3032,
"end": 3038,
"text": "Fig. 2",
"ref_id": "FIGREF2"
},
{
"start": 4107,
"end": 4113,
"text": "Fig. 2",
"ref_id": "FIGREF2"
},
{
"start": 5060,
"end": 5066,
"text": "Fig. 2",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Phrase identification",
"sec_num": "5."
},
{
"text": "In the postprocessing phase, we match the phrases found by NPPP and NPInst TermoPL grammars to some fragments of the dependency trees generated by COMBO for sentences containing these phrases. We imposed a few simple constraints that must be satisfied by matched tree fragments. The first one concerns prepositional phrases. If the preposition in the phrase being examined is associated with a part of the sentence that is not included in the phrase, then this phrase is certainly not a valid term. In other words, it means that no link in the dependency tree is allowed to connect something from outside the matched fragment with the preposition that lies inside this fragment. Examples of a good and a bad prepositional phrase are shown in Fig. 3 and 4 , respectively. The first of these phrases sp\u00f3\u0142ka z ograniczon\u0105 odpowiedzialno\u015bci\u0105 'limited liability company' is a good example of an economics term. The second one przedsi\u0119biorstwo pod w\u0142asn\u0105 firm\u0105 is a nonsense phrase that has a word for word translation 'enterprise under own company', which is equally nonsensical. It turns out that there are phrases that in some sentences are good candidates for terms, and in others not. A string podatek od dochodu which has the word for word translation 'tax from income' can be a noun modifier, e.g., [podatek od dochodu] nale\u017cy zap\u0142aci\u0107 w terminie do . . . 'income tax must be paid by . . . ', or it can be a valency constraint in the following sentence Wyliczaj\u0105c kwot\u0119 do za- . 'When calculating the amount to be paid, tax must be deducted from the income.' In the first example (see Fig. 5 ), the term podatek od dochodu is accepted by the constraint we mentioned above. In the second example, the same constraint rejects this phrase as a term (see Fig. 6 ).",
"cite_spans": [],
"ref_spans": [
{
"start": 742,
"end": 754,
"text": "Fig. 3 and 4",
"ref_id": "FIGREF3"
},
{
"start": 1586,
"end": 1592,
"text": "Fig. 5",
"ref_id": null
},
{
"start": 1752,
"end": 1758,
"text": "Fig. 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "Filtering phrases with COMBO",
"sec_num": "6."
},
{
"text": "Podatek od dochodu tax from income adjunct comp The second constraint we impose on dependency graphs concerns the consistency of its matched fragment. A fragment of the graph corresponding to the examined phrase is consistent if, passing from the node considered as the head of the phrase, we pass through all its nodes. Fig. 7 presents an inconsistent graph for the phrase koszty dojazd\u00f3w\u015brodkami (with word for word transtation 'travel costs by means'), which is syntactically correct, but without sense. However, when we consider the broader context, the phrase pokrycie koszt\u00f3w dojazd\u00f3w\u015brodkami komunikacji miejscowej 'coverage of travel costs by local transport', we obtain a phrase that makes sense and has a consistent graph. The graph for the phrase podatek od dochodu depicted in Fig. 6 is also inconsistent with this constraint (although it would anyway be rejected by the first rule described above). Finally, we eliminate graphs that correspond to some types of truncated phrases. They are depicted in Fig. 8-10 . Fig. 8 shows an example in which a named entity phrase should not be divided. The phrase Ustawa o Funduszu Kolejowym 'Act on the Railway Fund' may not be shortened to the phrase Ustawa o Funduszu 'Act on the Fund', although it is still acceptable at the syntactic level. The other two examples show situations in which an adjective or participle, modifying an object or a complement, should not be cut from the phrase on its right side, as they are usually necessary components of terms. The phrase podatek dochodowy od os\u00f3b fizycznych 'personal income tax' (see Fig. 9 ) cannot be shortened to podatek dochodowy od os\u00f3b. Similarly, the phrase opodatkowanie podatkiem dochodowym 'taxation on income' (see Figure 10 ) cannot be shortened to opodatkowanie podatkiem. Sometimes, truncated phrases can be identified by their inconsistent graphs as shown in Fig. 7 . We can now use all the above constraints to filter phases. If a phrase is supported by more than 50% of its dependency trees (which means that these trees satisfy all constraints), it is considered as a good term candidate. Otherwise, it is rejected.",
"cite_spans": [],
"ref_spans": [
{
"start": 321,
"end": 327,
"text": "Fig. 7",
"ref_id": "FIGREF6"
},
{
"start": 789,
"end": 795,
"text": "Fig. 6",
"ref_id": null
},
{
"start": 1014,
"end": 1023,
"text": "Fig. 8-10",
"ref_id": null
},
{
"start": 1026,
"end": 1032,
"text": "Fig. 8",
"ref_id": null
},
{
"start": 1589,
"end": 1595,
"text": "Fig. 9",
"ref_id": null
},
{
"start": 1731,
"end": 1740,
"text": "Figure 10",
"ref_id": null
},
{
"start": 1879,
"end": 1885,
"text": "Fig. 7",
"ref_id": "FIGREF6"
}
],
"eq_spans": [],
"section": "SUBST PREP SUBST",
"sec_num": null
},
{
"text": "We compare the manual evaluation of all phrases obtained by two separate grammars with the results of filtering described in Sec. 6. The filtering gives the binary information: correct/incorrect phrase so we assume that the result is proper if an incorrect phrase is manually classified as incorrect modification or incorrect 'other reason'. Tables 5-6 give the evaluation of phrases with prepositional modifiers and phrases with instrumental modifiers respectively, classified by the dependency parser. The results depicted there show that the proposed approach is not precise enough. For phrases with prepositional phrases, 74% of correct phrases are correctly classified as valid terms, but there are about twenty percent of the proper terms which are discarded. There are even more incorrect sequences which are classified as correct (about quarter). For instrumental modifications, there are far fewer incorrect sequences accepted as good, while the percentage of the correct terms which are classified as bad is even higher than for prepositional modifiers. The answer to the question whether these results are due to our classification strategy not being good enough or to the insufficient quality of the parser needs further research.",
"cite_spans": [],
"ref_spans": [
{
"start": 342,
"end": 352,
"text": "Tables 5-6",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "Evaluation of the method",
"sec_num": "7."
},
{
"text": "Manual eval. COMBO correct incorrect correct 365 87 incorrect 131 103 ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of the method",
"sec_num": "7."
},
{
"text": "In this section, we analyze results of TermoPL using the extended grammar given in Fig. 2 . A set of phrases for which the C-value is at least 3.0 are called hereinafter top3.0. For the plain method of term selection (without NPMI), the top3.0 consists of 5,935 terms. Phrases with prepositional modifiers are 11.9% of the top3.0 set. 7.6% of them are correct phrases and 4.3% are incorrect ones.",
"cite_spans": [],
"ref_spans": [
{
"start": 83,
"end": 89,
"text": "Fig. 2",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "8."
},
{
"text": "Then, we test if the NPMI method can prevent us from introducing incorrect phrases with prepositional modifiers into the top3.0 set. For some phrases, the NPMI method reduces their C-value which means they are pushed to the end of the list. Moreover, some phrases may not even appear on the term list. The top3.0 set for TermoPL with NPMI consists of 5,078 phrases. The statistics are given in Tab. 7, where the 'out top 3.0' column indicates the number of phrases whose C-value fell below the 3.0 level, and the 'out' column indicates the number of phrases which disappeared from the list. This method introduced 487 prepositional phrases into top3.0, which is 9.5%. 7.7% of them are correct phrases and 1.8% are incorrect ones. Table 9 : Phrases with a preposition modifier filtered by the dependency parser from the plain TermoPL results.",
"cite_spans": [],
"ref_spans": [
{
"start": 730,
"end": 737,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "8."
},
{
"text": "As we expected, application of the NPMI method in candidate phrase recognition reduces the number of incorrect phrases in the top3.0. In our experiment, they drop from 4.3% to 1.8% of all the top3.0. It slightly declines the share of phrases with prepositional modifiers on the top3.0 list from 11.9% to 9.5%. Moreover, it seems that this method works better for phrases which are incorrect because of 'other reasons' (e.g. truncated ones), as from the top3.0, it eliminates 71 of 130 such phrases (i.e. 54.6%) while for incorrect modifiers it eliminates 47 of 114 phrases which is 41.2%. For nominal phrases with instrumental modifiers only 37 correct and 2 incorrect (because of 'other reasons') were on the top3.0 list generated without NPMI. Statistics are given in Tab. 3. It gives 0.66% of all top3.0 phrases, where 0.62% are correct new phrases. When the NPMI method is used, the top3.0 list contains 30 correct and 2 incorrect phrases with instrumental modifiers, which gives 0.63% of all top3.0 phrases including 0.59% correct new phrases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "8."
},
{
"text": "To assess the usefulness of the dependency parsing we checked how many phrases with prepositional modifiers were accepted or deleted from the top3.0 of the TermoPL results generated without NPMI. The results for preposition modifiers are given in Tab. 9. So, filtering prepositional phrases by dependency grammar results in removing 187 phrases, where 101 of them were incorrect (i.e., their removal was justified).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "8."
},
{
"text": "The purpose of this work was to test whether dependency parsing can be useful in filtering out incorrectly constructed phrases in automatic terminology extraction. We tested this approach on phrases containing prepositional modifiers and nominal modifiers in the instrumental case. We realised that noun phrases with prepositional modifiers are important in the terminology extraction task, as they constitute about 10% of the top term phrases. The phrases with instrumental case modifiers are much less important as they create only 0.65% of the top phrases. However, it is worth noting that there are only two incorrect such phrases among the top3.0. These constructions are much rarer and the most frequent phrases usually form correct terms. There are about 6% of correct and 2% of incorrect preposition phrases on the top3.0 list generated without applying NPMI and filtered with the dependency parser. These results seem slightly worse than the results obtained by the NPMI method alone. It occurs that dependency parsing filters out an additional 43 incorrect phrases from the top3.0 list when the NPMI method is applied. Unfortunately, it also filters out 85 correct phrases. This observation requires further investigation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "9."
},
{
"text": "As the quality and efficiency of the dependency parsing is constantly improving, we hope that these methods will better support the selection of term candidates. We also plan to check how the proposed filtering methods will work on terms with other syntactic structures.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "9."
}
],
"back_matter": [
{
"text": "Work financed as part of the investment in the CLARIN-PL research infrastructure funded by the Polish Ministry of Science and Higher Education.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "SemEval 2017 task10: ScienceIEextracting keyphrases and relations from scientific publications",
"authors": [
{
"first": "I",
"middle": [],
"last": "Augenstein",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Das",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Riedel",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Vikraman",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mc-Callum",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 11th International Workshop on Semantic Evaluation",
"volume": "",
"issue": "",
"pages": "546--555",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Augenstein, I., Das, M., Riedel, S., Vikraman, L., and Mc- Callum, A. (2017). SemEval 2017 task10: ScienceIE- extracting keyphrases and relations from scientific publi- cations. In Proceedings of the 11th International Work- shop on Semantic Evaluation, pages 546-555. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Normalized (pointwise) mutual information in collocation",
"authors": [
{
"first": "G",
"middle": [],
"last": "Bouma",
"suffix": ""
}
],
"year": 2009,
"venue": "From Form to Meaning: Processing Texts Automatically, Proceedings of the Biennial GSCL Conference",
"volume": "",
"issue": "",
"pages": "31--40",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bouma, G. (2009). Normalized (pointwise) mutual infor- mation in collocation. In Christian Chiarcos, et al., edi- tors, From Form to Meaning: Processing Texts Automat- ically, Proceedings of the Biennial GSCL Conference, pages 31-40. Tubingen: Gunter Narr Verlag.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "TermSuite: Terminology extraction with term variant detection",
"authors": [
{
"first": "D",
"middle": [],
"last": "Cram",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Daille",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics-System Demonstrations",
"volume": "",
"issue": "",
"pages": "13--18",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cram, D. and Daille, B. (2016). TermSuite: Terminology extraction with term variant detection. In Proceedings of the 54th Annual Meeting of the Association for Compu- tational Linguistics-System Demonstrations, pages 13- 18. Association for Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Automatic recognition of multi-word terms: the C-value/NCvalue method",
"authors": [
{
"first": "K",
"middle": [],
"last": "Frantzi",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Ananiadou",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Mima",
"suffix": ""
}
],
"year": 2000,
"venue": "Int. Journal on Digital Libraries",
"volume": "3",
"issue": "",
"pages": "115--130",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Frantzi, K., Ananiadou, S., and Mima, H. (2000). Auto- matic recognition of multi-word terms: the C-value/NC- value method. Int. Journal on Digital Libraries, 3:115- 130.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Citius at SemEval-2017 task 2: Cross-lingual similarity from comparable corpora and dependency-based contexts",
"authors": [
{
"first": "P",
"middle": [],
"last": "Gamallo",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 11th International Workshop on Semantic Evaluation",
"volume": "",
"issue": "",
"pages": "226--229",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gamallo, P. (2017). Citius at SemEval-2017 task 2: Cross-lingual similarity from comparable corpora and dependency-based contexts. In Proceedings of the 11th International Workshop on Semantic Evaluation, pages 226-229. Association for Computational Linguistics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "The Sketch Engine: ten years on. Lexicography",
"authors": [
{
"first": "A",
"middle": [],
"last": "Kilgarriff",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Baisa",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Bu\u0161ta",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Jakub\u00ed\u010dek",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Kov\u00e1\u0159",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Michelfeit",
"suffix": ""
},
{
"first": "Pavel",
"middle": [],
"last": "Rychl\u00fd",
"suffix": ""
},
{
"first": "V",
"middle": [
"S"
],
"last": "",
"suffix": ""
}
],
"year": 2014,
"venue": "",
"volume": "1",
"issue": "",
"pages": "7--36",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kilgarriff, A., Baisa, V., Bu\u0161ta, J., Jakub\u00ed\u010dek, M., Kov\u00e1\u0159, V., Michelfeit, J., and Pavel Rychl\u00fd, V. S. (2014). The Sketch Engine: ten years on. Lexicography, 1:7-36.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Semeval-2010 task 5: Automatic keyphrase extraction from scientific articles",
"authors": [
{
"first": "S",
"middle": [
"N"
],
"last": "Kim",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Medelyan",
"suffix": ""
},
{
"first": "M.-Y",
"middle": [],
"last": "Kan",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Baldwin",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the 5th International Workshop on Semantic Evaluation",
"volume": "",
"issue": "",
"pages": "21--26",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim, S. N., Medelyan, O., Kan, M.-Y., and Baldwin, T. (2010). Semeval-2010 task 5: Automatic keyphrase ex- traction from scientific articles. In Proceedings of the 5th International Workshop on Semantic Evaluation, pages 21-26. Association for Computational Linguistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "A genetic algorithm enabled ensemble for unsupervised medical term extraction from clinical letters",
"authors": [
{
"first": "W",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "B",
"middle": [
"C"
],
"last": "Chung",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Morlet",
"suffix": ""
}
],
"year": 2015,
"venue": "Health Information Science and Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Liu, W., Chung, B. C., Wang, R., Ng, J., and Morlet, N. (2015). A genetic algorithm enabled ensemble for un- supervised medical term extraction from clinical letters. Health Information Science and Systems.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "A novel parsing-based automatic domain terminology extraction method",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Quan",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Wen",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "He",
"suffix": ""
}
],
"year": 2018,
"venue": "Lecture Notes in Computer Science",
"volume": "10862",
"issue": "",
"pages": "796--802",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Liu, Y., Zhang, T., Quan, P., Wen, Y., Wu, K., and He, H. (2018). A novel parsing-based automatic domain termi- nology extraction method. In Shi Y. et al., editor, Com- putational Science -ICCS 2018. Lecture Notes in Com- puter Science, vol 10862. Springer, Cham, pages 796- 802.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Biomedical term extraction: overview and a new methodology",
"authors": [
{
"first": "J",
"middle": [
"A"
],
"last": "Lossio-Ventura",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Jonquet",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Roche",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Teisseire",
"suffix": ""
}
],
"year": 2016,
"venue": "Information Retrieval Journal",
"volume": "",
"issue": "",
"pages": "1573--7659",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lossio-Ventura, J. A., Jonquet, C., Roche, M., and Teis- seire, M. (2016). Biomedical term extraction: overview and a new methodology. Information Retrieval Journal, pages 1573-7659.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Terminology extraction from medical texts in Polish",
"authors": [
{
"first": "M",
"middle": [],
"last": "Marciniak",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mykowiecka",
"suffix": ""
}
],
"year": 2014,
"venue": "Journal of biomedical semantics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marciniak, M. and Mykowiecka, A. (2014). Terminol- ogy extraction from medical texts in Polish. Journal of biomedical semantics, 24.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Nested term recognition driven by word connection strength",
"authors": [
{
"first": "M",
"middle": [],
"last": "Marciniak",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mykowiecka",
"suffix": ""
}
],
"year": 2015,
"venue": "Terminology",
"volume": "2",
"issue": "",
"pages": "180--204",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marciniak, M. and Mykowiecka, A. (2015). Nested term recognition driven by word connection strength. Termi- nology, 2:180-204.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "TermoPL -a flexible tool for terminology extraction",
"authors": [
{
"first": "M",
"middle": [],
"last": "Marciniak",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mykowiecka",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Rychlik",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation, LREC 2016",
"volume": "",
"issue": "",
"pages": "2278--2284",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marciniak, M., Mykowiecka, A., and Rychlik, P. (2016). TermoPL -a flexible tool for terminology extraction. In Nicoletta Calzolari, et al., editors, Proceedings of the Tenth International Conference on Language Resources and Evaluation, LREC 2016, pages 2278-2284, Por- toro\u017e, Slovenia. ELRA, European Language Resources Association (ELRA).",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Deep keyphrase generation",
"authors": [
{
"first": "R",
"middle": [],
"last": "Meng",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Han",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Brusilovsky",
"suffix": ""
},
{
"first": "Chi",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "582--592",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Meng, R., Zhao, S., Han, S., He, D., Brusilovsky, P., and Chi, Y. (2017). Deep keyphrase generation. In Pro- ceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 582-592, Vancouver, Canada, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Automatic keyphrase extraction: a survey and trends",
"authors": [
{
"first": "Z",
"middle": [
"A"
],
"last": "Merrouni",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Frikh",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Ouhbi",
"suffix": ""
}
],
"year": 2019,
"venue": "Journal of Intelligent Information Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Merrouni, Z. A., Frikh, B., and Ouhbi, B. (2019). Auto- matic keyphrase extraction: a survey and trends. Journal of Intelligent Information Systems.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Narodowy Korpus J\u0119zyka Polskiego",
"authors": [
{
"first": "Adam",
"middle": [],
"last": "Przepi\u00f3rkowski",
"suffix": ""
}
],
"year": 2012,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Adam Przepi\u00f3rkowski, et al., editors. (2012). Naro- dowy Korpus J\u0119zyka Polskiego. Wydawnictwo Naukowe PWN.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Automatic keyword extraction from individual documents",
"authors": [
{
"first": "S",
"middle": [],
"last": "Rose",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Engel",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Cramer",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Cowley",
"suffix": ""
}
],
"year": 2010,
"venue": "Text Mining: Applications and Theory",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rose, S., Engel, D., Cramer, N., and Cowley, W. (2010). Automatic keyword extraction from individual docu- ments. Text Mining: Applications and Theory, pages 1 -20, 03.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Semi-supervised neural system for tagging, parsing and lematization",
"authors": [
{
"first": "P",
"middle": [],
"last": "Rybak",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Wr\u00f3blewska",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies",
"volume": "",
"issue": "",
"pages": "45--54",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rybak, P. and Wr\u00f3blewska, A. (2018). Semi-supervised neural system for tagging, parsing and lematization. In Proceedings of the CoNLL 2018 Shared Task: Multilin- gual Parsing from Raw Text to Universal Dependencies, page 45-54. Association for Computational Linguistics.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Harnessing the CRF complexity with domain-specific constraints. The case of morphosyntactic tagging of a highly inflected language",
"authors": [
{
"first": "J",
"middle": [],
"last": "Waszczuk",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 24th International Conference on Computational Linguistics (COLING 2012)",
"volume": "",
"issue": "",
"pages": "2789--2804",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Waszczuk, J. (2012). Harnessing the CRF complexity with domain-specific constraints. The case of morphosyntac- tic tagging of a highly inflected language. In Proceed- ings of the 24th International Conference on Computa- tional Linguistics (COLING 2012), pages 2789-2804.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Key-phrase extraction based on a combination of CRF model with document structure",
"authors": [
{
"first": "F",
"middle": [],
"last": "Yu",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Xuan",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Zheng",
"suffix": ""
}
],
"year": 2012,
"venue": "Eighth International Conference on Computational Intelligence and Security",
"volume": "",
"issue": "",
"pages": "406--410",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yu, F., Xuan, H., and Zheng, D. (2012). Key-phrase ex- traction based on a combination of CRF model with document structure. In Eighth International Conference on Computational Intelligence and Security, pages 406- 410.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Keyphrase extraction using deep recurrent neural networks on twitter",
"authors": [
{
"first": "Q",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Gong",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Huang",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "836--845",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhang, Q., Wang, Y., Gong, Y., and Huang, X. (2016). Keyphrase extraction using deep recurrent neural net- works on twitter. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 836-845, Austin, Texas, November. Association for Computational Linguistics.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Automatic keyword extraction from documents using conditional random fields",
"authors": [
{
"first": "C",
"middle": [],
"last": "Zhang",
"suffix": ""
}
],
"year": 2008,
"venue": "Journal of Computational Information Systems",
"volume": "4",
"issue": "3",
"pages": "1169--1180",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhang, C. (2008). Automatic keyword extraction from documents using conditional random fields. Journal of Computational Information Systems, 4(3):1169-1180.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"text": "NPP : $NAP NGEN * ; NAP [agreement] : AP * N AP * ; NGEN [case = gen] : NAP ; AP : ADJ | PPAS | ADJA DASH ADJ ; N [pos = subst, ger ]; ADJ [pos = adj ]; ADJA[pos = adja]; PPAS [pos = ppas]; DASH [form = \"-\"];",
"num": null,
"uris": null
},
"FIGREF1": {
"type_str": "figure",
"text": "NP : NPPINST | PPP | NPAPGEN ; PPP : NPAPGEN PREP NAP + ; NPPINST : NPAPGEN NINST NGEN * ; NPAPGEN : $NAP NGEN * ; NAP [agreement] : AP * N AP * ; NGEN [case = gen] : NAP ; NINST [case = inst] : NAP ; AP : ADJ | PPAS | ADJA DASH ADJ ; N [pos = subst, ger ]; ADJ [pos = adj ]; ADJA[pos = adja]; PPAS [pos = ppas]; DASH [form = \"-\"]; PREP [pos = prep];",
"num": null,
"uris": null
},
"FIGREF2": {
"type_str": "figure",
"text": "The final grammar with added modifiers being noun phrases in the instrumental case and prepositional phrases.",
"num": null,
"uris": null
},
"FIGREF3": {
"type_str": "figure",
"text": "Dependency graph corresponding to correct prepositional phrase sp\u00f3\u0142ka z ograniczon\u0105 odpowiedzialno\u015bci\u0105.",
"num": null,
"uris": null
},
"FIGREF4": {
"type_str": "figure",
"text": "Dependency graph corresponding to the incorrect prepositional phrase przedsi\u0119biorstwo pod w\u0142asn\u0105 firm\u0105. p\u0142aty nale\u017cy odj\u0105\u0107 [podatek] [od dochodu]",
"num": null,
"uris": null
},
"FIGREF5": {
"type_str": "figure",
"text": "Accepted term podatek od dochodu. Rejected term podatek od dochodu.",
"num": null,
"uris": null
},
"FIGREF6": {
"type_str": "figure",
"text": "Graph inconsistency for the phrase koszty dojazd\u00f3w\u015brodkami.",
"num": null,
"uris": null
},
"FIGREF7": {
"type_str": "figure",
"text": "Truncated named entity (ne) phrase. Truncated phrase podatek dochodowy od os\u00f3b. Rejected term opodatkowanie podatkiem.",
"num": null,
"uris": null
},
"TABREF1": {
"type_str": "table",
"content": "<table/>",
"text": "",
"num": null,
"html": null
},
"TABREF3": {
"type_str": "table",
"content": "<table><tr><td colspan=\"3\">: Manual evaluation of selected phrases with an in-</td></tr><tr><td>strumental modifier.</td><td/><td/></tr><tr><td/><td colspan=\"2\">number of phrases</td></tr><tr><td colspan=\"3\">frequency absolute relative</td></tr><tr><td>10-16</td><td>4</td><td>0.04</td></tr><tr><td>5-9</td><td>11</td><td>0.10</td></tr><tr><td>3-4</td><td>20</td><td>0.18</td></tr><tr><td>2</td><td>75</td><td>0.68</td></tr></table>",
"text": "",
"num": null,
"html": null
},
"TABREF4": {
"type_str": "table",
"content": "<table/>",
"text": "Number of top phrases with an instrumental modifier in different frequency groups.",
"num": null,
"html": null
},
"TABREF6": {
"type_str": "table",
"content": "<table><tr><td colspan=\"3\">: Comparison of the manual evaluation of the</td></tr><tr><td colspan=\"3\">phrases with a preposition modifier with the dependency</td></tr><tr><td colspan=\"3\">parser filtering. The results achieved for classification</td></tr><tr><td colspan=\"3\">a phrase as correct by the parser: precision=0.74, re-</td></tr><tr><td>call=0.81.</td><td/><td/></tr><tr><td>Manual eval.</td><td colspan=\"2\">COMBO correct incorrect</td></tr><tr><td>correct</td><td>50</td><td>34</td></tr><tr><td>incorrect</td><td>4</td><td>22</td></tr></table>",
"text": "",
"num": null,
"html": null
},
"TABREF7": {
"type_str": "table",
"content": "<table><tr><td>type</td><td colspan=\"3\">in top3.0 out top3.0 out</td></tr><tr><td>correct term</td><td>391</td><td>27</td><td>34</td></tr><tr><td>incorr modif.</td><td>67</td><td>15</td><td>32</td></tr><tr><td>incorr. -other</td><td>59</td><td>33</td><td>38</td></tr><tr><td>total</td><td>487</td><td colspan=\"2\">75 104</td></tr></table>",
"text": "Evaluation of the phrases with an instrument modifier filtered by the dependency parser. The results achieved for classification a phrase as a correct one by the parser: precision=0.93, recall=0.60.",
"num": null,
"html": null
},
"TABREF8": {
"type_str": "table",
"content": "<table/>",
"text": "Phrases with a preposition modifier -with NPMI.",
"num": null,
"html": null
},
"TABREF10": {
"type_str": "table",
"content": "<table><tr><td>type</td><td>in top3.0</td></tr><tr><td>correct-accepted</td><td>365</td></tr><tr><td>incorrect-accepted</td><td>131</td></tr><tr><td>correct-deleted</td><td>86</td></tr><tr><td>incorrect-deleted</td><td>101</td></tr></table>",
"text": "Distribution of the correct phrases with prepositional modifiers in top3.0 of TermoPL with NPMI.",
"num": null,
"html": null
}
}
}
} |