File size: 90,115 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 | {
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:43:44.745086Z"
},
"title": "Starting a new treebank? Go SUD! Theoretical and practical benefits of the Surface-Syntactic distributional approach",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": "",
"affiliation": {},
"email": "gerdes@lisn.fr"
},
{
"first": "Bruno",
"middle": [],
"last": "Guillaume",
"suffix": "",
"affiliation": {},
"email": "bruno.guillaume@inria.fr"
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": "",
"affiliation": {},
"email": "sylvain@kahane.fr"
},
{
"first": "Guy",
"middle": [],
"last": "Perrier",
"suffix": "",
"affiliation": {},
"email": "guy.perrier@loria.fr"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "The paper brings to the fore some advantages to first develop a new treebank in Surface-Syntactic Universal Dependencies (SUD) annotation scheme, even if the goal is to obtain a UD treebank. Theoretical benefits of SUD are presented, as well as UD-compatible SUD innovations. The twoway UD \u21d4 SUD conversion is explained, as well as the possibility to customize the conversion for a given language. The paper concludes by a practical guide for the development of a SUD treebank. This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http: //creativecommons.org/licenses/by/4.0/. 1 UD is initially based on Stanford dependencies, which was itself the conversion into a dependency tree of the outputs of a phrase-structure-based parser. In consequence, UD dependency relations combine both functional and categorical information, for instance with the nsubj vs csubj distinction between nominal and clausal subjects, the obj vs ccomp distinction between nominal and clausal objects, or the amod vs nmod vs advmod distinction between adjectival, nominal, and adverbial modifiers, as well as the obl vs nmod distinction between adpositional phrases depending on a verb or a noun. Moreover, UD is very semantically-oriented, favoring relations between content words, leaning towards a sort of interlingua representation. The part of speech tags, stemming from Google's universal POS (Petrov et al., 2012) and the Interset interlingua tagset (Zeman, 2008), were added independently, resulting in some redundancy. 2 Let us recall that in UD function words depend on content words. As a consequence, adpositions are dependents of the noun with which they form a phrase. This is in complete contradiction with typological studies that show that the adpositionnoun relation tends to have similar properties than the verb-object relation. In particular, VO languages have prepositions while OV languages have postpositions (Dryer, 1992).",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "The paper brings to the fore some advantages to first develop a new treebank in Surface-Syntactic Universal Dependencies (SUD) annotation scheme, even if the goal is to obtain a UD treebank. Theoretical benefits of SUD are presented, as well as UD-compatible SUD innovations. The twoway UD \u21d4 SUD conversion is explained, as well as the possibility to customize the conversion for a given language. The paper concludes by a practical guide for the development of a SUD treebank. This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http: //creativecommons.org/licenses/by/4.0/. 1 UD is initially based on Stanford dependencies, which was itself the conversion into a dependency tree of the outputs of a phrase-structure-based parser. In consequence, UD dependency relations combine both functional and categorical information, for instance with the nsubj vs csubj distinction between nominal and clausal subjects, the obj vs ccomp distinction between nominal and clausal objects, or the amod vs nmod vs advmod distinction between adjectival, nominal, and adverbial modifiers, as well as the obl vs nmod distinction between adpositional phrases depending on a verb or a noun. Moreover, UD is very semantically-oriented, favoring relations between content words, leaning towards a sort of interlingua representation. The part of speech tags, stemming from Google's universal POS (Petrov et al., 2012) and the Interset interlingua tagset (Zeman, 2008), were added independently, resulting in some redundancy. 2 Let us recall that in UD function words depend on content words. As a consequence, adpositions are dependents of the noun with which they form a phrase. This is in complete contradiction with typological studies that show that the adpositionnoun relation tends to have similar properties than the verb-object relation. In particular, VO languages have prepositions while OV languages have postpositions (Dryer, 1992).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "SUD, Surface-Syntactic Universal Dependencies, is a syntactic annotation scheme, which is a convertible variant of Universal Dependencies (UD). UD is a very successful treebank development project that is now an indispensable standard of data-based syntax (de Marneffe et al., 2021) . To benefit from UD's wealth of expertise, tools, and cross-language comparability, any annotation scheme must eventually be convertible into UD. Nevertheless, the UD annotation scheme was initially developed in the context of NLP applications, rather than pure linguistic considerations and some initial choices are problematic. 1 SUD is based on a different theoretical framework that has many advantages for treebank development as we will show in this paper.",
"cite_spans": [
{
"start": 256,
"end": 282,
"text": "(de Marneffe et al., 2021)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "SUD has already been presented in two papers by (Gerdes et al., 2018; Gerdes et al., 2019) . While SUD's theoretical foundations remain unchanged, this paper proposes one change of SUD's philosophy. At first, SUD was thought of as a pure variant of UD with a complete equivalence between SUD and UD. Initially, SUD was more interested in the UD \u21d2 SUD conversion because for some studies, especially on word order typology, a more surfacic annotation was required. 2 This paper reports on a growing interest in SUD \u21d2 UD conversions and the development of treebanks in SUD in order to obtain both SUD and UD variants of the treebank. The UD \u21d2 SUD conversion grammar is still maintained and has even been improved with the possibility to more easily customize the conversion for a given language. Recent views on SUD abandons the idea of having an equivalence between the two annotation schemes, and this paper postulate that SUD is a richer annotation scheme than UD. In other words, no information is lost in UD \u21d2 SUD and a double conversion UD \u21d2 SUD \u21d2 UD should give the initial treebank, eventually with additional features. 3 But a SUD \u21d2 UD conversion generally causes a loss of information and SUD treebanks obtained from a UD conversion are underspecified for some features considered as relevant for the SUD annotation scheme, such as the internal structure of nuclei or of MWEs. As a simple example consider the verbal chain in the sentence I would have left. SUD annotates the hierarchical relation between the three verbs (would \u2192 have \u2192 left), UD sees a flat structure in these three verbs with the lexical verb (left) at its head. Therefore, the hierarchical relation between would and have is not encoded in UD, and requires language specific heuristics to obtain the correct SUD structure. Theoretical benefits of SUD are presented in Section 2 and completed in Section 3 by UD-compatible SUD innovations.",
"cite_spans": [
{
"start": 48,
"end": 69,
"text": "(Gerdes et al., 2018;",
"ref_id": "BIBREF6"
},
{
"start": 70,
"end": 90,
"text": "Gerdes et al., 2019)",
"ref_id": "BIBREF7"
},
{
"start": 1126,
"end": 1127,
"text": "3",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Due to the fact the SUD is richer than UD, we encourage developers of treebank to start with a SUD annotation, which allows them to obtain a high-quality UD treebank, while keeping information that is flattened out in UD. Moreover if a treebank already exists in a third format, it can be easier to convert it into SUD and only then into UD rather than to aim UD directly because of the unconventional lexicalword-centric approach of UD. We may further assume that SUD's additional richness does not slow down the overall annotation process as it also removes some redundancies of UD. The UD \u21d2 SUD and SUD \u21d2 UD conversions are presented in Sections 4 and 5, as well as the possibility to customize the conversions for a given language. Section 6 sketches a practical guide for the development of a SUD treebank.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We discuss four benefits of SUD compared to UD: a definition of dependency based on distributional criteria, an encoding of the internal structure of nuclei, a definition of syntactic relations based on commutation positional paradigms, and a more symmetrical analysis of coordination. These properties are core elements that cannot be integrated in UD, which is based on different fundamentals. Other benefits of the current SUD annotation that could be adopted in UD are presented in Section 3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Theoretical benefit of SUD",
"sec_num": "2"
},
{
"text": "UD favors relations between content words, while function words are treated as dependents of content words. While it may seem at first view that it is easy to establish the difference between function and lexical words for a new language, it turns out to be a hard task to delimit the content word -function word opposition that is compatible with a coherent non-catastrophic annotation. 4 Moreover, supposing that the opposition is semantic or language independent can lead to erasing typologically important structural differences, for example when languages differ precisely in the structure of function words. Relegating all function words as done by UD makes us loose some syntactic information as we will see in the next section.",
"cite_spans": [
{
"start": 388,
"end": 389,
"text": "4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "SUD favors a definition of the dependency structure based on a more traditional definition of head: The head of a unit U is the element A that controls the distribution of U. By distribution, we mean what Mel'\u010duk's (1988) calls the passive valency, that is, the set of possible syntactic governors for U, or, similarly, the set of syntactic positions that U can occupy. Even if the notion of governor is based on the notion of distribution, we avoid the circularity, because in most cases the question of the head is not controversial, especially for the governor of a sentence.",
"cite_spans": [
{
"start": 205,
"end": 221,
"text": "Mel'\u010duk's (1988)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "As soon as we can determine units and a head for each unit, we have a dependency structure (Gerdes and Kahane, 2013) : B depends on A as soon as A is the head of the unit that A and B form together.",
"cite_spans": [
{
"start": 91,
"end": 116,
"text": "(Gerdes and Kahane, 2013)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "This definition of the head is based on formal criteria that we want to recall here because they have often been misstated. Let us consider a unit U = AB. The simplest case is when A or B can stand alone.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "In this case the distribution of A or B can be considered and compared with the whole unit U. 5 It gives us two criteria.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "Positive distributional criterion with deletion. If U = AB, A can stand alone (i.e., B can be deleted), and U and A have the same distribution, then A is a head of U.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "Negative distributional criterion with deletion. If U = AB, B can stand alone, and U and B do not have the same distribution, then A is a head of U.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "The second criteria can be applied to examples such as U = John ran or U' = with John, where B = John. Clearly B does not have the same distribution as the clause U or the phrase U' and then the verb is the head of U and the adposition the head of U'. In the same way, a combination auxiliary-verb such as U = is expected has the auxiliary as head, because the past participle has a different distribution: It can be the dependent of a noun (that's the guy expected at noon), while is expected can be the dependent of a verb (he knows he is expected). 6 It is not needed to delete an element to decide which element is the head, a commutation with another element is sufficient:",
"cite_spans": [
{
"start": 552,
"end": 553,
"text": "6",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "Distributional criterion without deletion. If U = AB, A can commute with an A', and U and U' = A'B does not have the same distribution, then A is a head of U. 7 In other words, if B depends on A, then B must not modify the distribution of A and a commutation on B does not change the distribution of the unit it forms with A.",
"cite_spans": [
{
"start": 159,
"end": 160,
"text": "7",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "For instance, U = with John and U' = by John have different distributions. In other words, the commutation of with and by change the distribution, which implies that the preposition is the head. The same criteria can be used with the determiner-noun combination: Some nouns such as day (she stayed two days) or time (I will do that (the) next time) have a very special distribution, being able to work as an adverbial phrase, whatever the determiner is. This is a good argument to take the noun as the head, even if there are also arguments to take the determiner as the head.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of dependency based on distributional criteria",
"sec_num": "2.1"
},
{
"text": "In a recent paper on UD, de Marneffe et al. (2021) justify treating function words as dependents as follows: \"Sometimes linguistic head functions are divided between a structural center (an auxiliary or function word) and a semantic center (a lexical or content word), such as for periphrastic verb tenses like has arrived. This is what Tesni\u00e8re (2015 [1959] , ch. 23) refers to as a dissociated nucleus. In such cases, UD chooses the lexical or content word as the head, and makes function words dependents of the head in the dependency tree structure, while recognizing that they do form a nucleus together with the content word.\" Nevertheless in case of the presence of multiple function words, Tesni\u00e8re considers that there is an embedding of nuclei, while UD only considers a flat structure with all function words depending on the same content word and the internal structure of the nucleus is completely lost. For instance, in the sentence of Fig. 1 , in Mesoamerica is clearly a nucleus that is put in a comparison with in the Americas and then embedded in than in Mesoamerica. 8 The UD analysis does not have a phrase in Mesoamerica.",
"cite_spans": [
{
"start": 346,
"end": 358,
"text": "(2015 [1959]",
"ref_id": null
}
],
"ref_spans": [
{
"start": 950,
"end": 956,
"text": "Fig. 1",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Internal structure of nuclei",
"sec_num": "2.2"
},
{
"text": "In the UD \u21d2 SUD conversion, we use heuristics that are described in Section 4, depending on the order of the function words and their function. In particular, the closer a function word is to the content word, the earlier they combine. The SUD structure of the same sentence is given in Fig. 1 (lower part). But this heuristic does not work in some cases. For instance, Wolof has a multitude of auxiliaries that are used to focus the subject, a complement, or the verb itself, which will occupy the first place in the clause (Robert, 1991; Bond\u00e9elle and Kahane, 2021) . The auxiliary na, used to focus a verb, can also focus an auxiliary, as in Fig. 2 where the past imperfective auxiliary doon is focalized by na, which is the head of the nucleus doon na VERB. Here, na is the closest function word to the content word, but it combines last. Another problematic case is when there are function words on both sides of the content word. This can be illustrated by the auxiliaries in German, as in sentence (1). German is a V2 language, where the finite verbal form always occupies the second position of a declarative sentence, whether it is a content verb or an auxiliary. In (1), the verb has two auxiliaries, wird 'will' on the left and haben 'have' on the right. The auxiliary on the left, which is in the second position in the sentence and has a finite form, is the root of the syntactic structure, which cannot be guessed from the flat UD structure alone.",
"cite_spans": [
{
"start": 525,
"end": 539,
"text": "(Robert, 1991;",
"ref_id": "BIBREF13"
},
{
"start": 540,
"end": 567,
"text": "Bond\u00e9elle and Kahane, 2021)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 287,
"end": 293,
"text": "Fig. 1",
"ref_id": "FIGREF1"
},
{
"start": 645,
"end": 651,
"text": "Fig. 2",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Internal structure of nuclei",
"sec_num": "2.2"
},
{
"text": "In SUD, two dependents that belong to the same positional paradigm have the same syntactic relations, in accordance with Mel'\u010duk's (1988) or Van den Eynde & Mertens' (2003) definitions, while UD takes also into account the POS of the governor and/or the dependent (see Note 1 about the definition of relations in UD). One advantage of the SUD definition is the possibility to compare the valency of two occurrences of the same lemma and to extract a syntactic lexicon more easily.",
"cite_spans": [
{
"start": 121,
"end": 137,
"text": "Mel'\u010duk's (1988)",
"ref_id": "BIBREF11"
},
{
"start": 149,
"end": 172,
"text": "Eynde & Mertens' (2003)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Definition of syntactic relation based on positional paradigms",
"sec_num": "2.3"
},
{
"text": "As UD, SUD uses the notation rel:subrel for a sub-relation of a given relation. Syntactic relations are part of a hierarchy and comp:obj or comp:obl must be understood as sub-relations of a more generic comp relation. Modifiers (mod) and complements (comp) are distinguished, but a super-relation udep (underspecified dependency) can be used if we do not want to make this distinction. We use it for noun dependents and it is used in non-native SUD treebanks for the conversion of the UD obl relation, which gives the udep relation in SUD. 9 Figures 4, 5, and 6 give UD and SUD annotations of verb dependents which are respectively modifier, argument and underspecified. Annotations in Figures 4 and 5 are SUD-native and contain a distinction between complements and modifiers, which is kept in the conversion with the UD relations obl:arg, iobj, and obl:mod. Conversely, the sentence in Figure 6 comes from UD ENGLISH-GUM, where the distinction between complements and modifiers is not present for preposition phrases and the conversion to SUD gives us a udep relation. Additional features on relations are clearly separated from the relation itself, especially when it is semantic information. We use for this the delimiter @. For instance, the semantic value of an auxiliary (tense, passive, causative) can be indicated on the comp:aux relation: comp:aux@tense, comp:aux@pass, comp:aux@caus. Subjects all have the function subj, but expletive or passive subjects can be marked by an additional feature: subj@expl, subj@pass. 10 In spoken corpora, the feature @scrap has been used for incomplete units. This feature is particularly useful for error mining: for instance, a relation between a verb and a determiner (in an incomplete sentence such as I see the. . . ) should not be allowed without a @scrap.",
"cite_spans": [],
"ref_spans": [
{
"start": 888,
"end": 896,
"text": "Figure 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "Definition of syntactic relation based on positional paradigms",
"sec_num": "2.3"
},
{
"text": "In UD, the dependent shared by all the conjuncts are attached to the head of the coordination, the leftmost conjunct. In the example of Fig. 7 , from the UD ENGLISH-EWT corpus, there are two modifiers of the coordination food and coffee: a left modifier Good and a right modifier with a nice atmosphere. Since the right modifier is after the second conjunct, the UD annotation has only one interpretation: It cannot be the modifier of the first conjunct alone but only of the coordination as a whole. However, for the left modifier, the UD annotation does not indicate whether it is a modifier of food only or of food and coffee. This is an unfortunate asymmetry.",
"cite_spans": [],
"ref_spans": [
{
"start": 136,
"end": 142,
"text": "Fig. 7",
"ref_id": "FIGREF7"
}
],
"eq_spans": [],
"section": "A more symmetrical analysis of coordination",
"sec_num": "2.4"
},
{
"text": "In SUD, as in UD, the head of the coordination is the head of the leftmost conjunct, but for the dependents, the annotation is perfectly symmetrical. They are attached to the nearest conjunct: the left to the leftmost conjunct and the right to the rightmost conjunct. In order to indicate which dependents are shared, we introduce the feature Shared with values Yes and No. Conversions of UD treebanks, only give a partial instantiation of the Shared feature. In the native SUD FRENCH-GSD, Shared=Yes features have been systematically introduced. Note also the considerably shorter overall dependency lengths of the SUD annotation scheme, which is not only cognitively more plausible but also facilitates manual annotation and correction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "A more symmetrical analysis of coordination",
"sec_num": "2.4"
},
{
"text": "This section presents features of the SUD annotation scheme that could, and we believe should, be integrated into the UD annotation guidelines. For now, the SUD \u21d2 UD conversion will encode these SUD features as optional additional information in the MISC column.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UD-compatible SUD innovations",
"sec_num": "3"
},
{
"text": "Multi-Word expressions (MWE) cover a wide heterogeneous field of constructions such as use of foreign words that have no internal structure in the host language (Burkina Faso, Hong Kong, ad hoc), or completely regular structures in named entities (the Embassy of Ecuador in London, the United States).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Internal structure of Multi-Word Expressions",
"sec_num": "3.1"
},
{
"text": "Interesting from a syntactic point of view is another set of phenomena: constructions that have a regular internal structure but that intervene as a whole at an unexpected point in the sentence. For example, in order (to VERB) is analyzed as an MWE in English treebanks, as shown in Fig. 8 (upper part) from UD ENGLISH-GUM, with a fixed relation between in and order.",
"cite_spans": [],
"ref_spans": [
{
"start": 283,
"end": 302,
"text": "Fig. 8 (upper part)",
"ref_id": "FIGREF9"
}
],
"eq_spans": [],
"section": "Internal structure of Multi-Word Expressions",
"sec_num": "3.1"
},
{
"text": "Even if in order is semantically frozen it is nevertheless a syntactically regular preposition-noun combination. In native SUD, the sentence is analyzed with the standard comp:obj relation between in and order (the noun is the object of the adposition) and the idiomaticity is encoded by additional features Idiom=Yes on the head and InIdiom=Yes on the other elements. Moreover, we consider that in order as a whole works as an adverb, which is encoded in SUD by the feature ExtPos=ADV (for external POS). 11 Of course, this SUD analysis translates into a different UD analysis, because adverbs are analyzed as content words. 12 Arguably, the UD analysis would have been different if the internal structure of the MWE had been taken into account.",
"cite_spans": [
{
"start": 626,
"end": 628,
"text": "12",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Internal structure of Multi-Word Expressions",
"sec_num": "3.1"
},
{
"text": "It was identified in UD that, in several places, syntactic units do not exactly correspond to orthographic units given in the raw text. 13 For instance, in French the orthographic unit au is a contraction of two syntactic units: the preposition\u00e0 and the determiner le (such amalgams are called Multi-Word Tokens or MWT). With a focus on syntax, it is natural to consider syntactic units as the basic units of annotation; this is what is done both in UD and in SUD. However, it is necessary to keep all the information and to also encode the orthographic unit when it differs from the units of the structure. The UD guidelines 14 introduce the CoNLL-U format with a dedicated mechanism with a new type of line describing a range of tokens (2-3 in the example below) to store the contracted form.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Textform and wordform",
"sec_num": "3.2"
},
{
"text": "The main drawback of this solution is that the syntactic dependency structure, being based on the syntactic units, does not refer to the orthographic units which are then not easily accessible for tools working on the syntactic structure. Having access to these orthographic units is useful for parsing.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "2-3 au _ _ 2\u00e0\u00e0 ADP 3 le le DET",
"sec_num": null
},
{
"text": "There are other cases where an orthographic unit is different from a canonical token. For instance, for several languages, uppercase letters are used at the beginning of a sentence, in specific usages for naming institutions (the White House), in titles (What the Moon Brings [GUM fiction moon-1]), or for emphasis (YES!). It is useful to encode the canonical form in these cases, as it allows for an improved data analysis, performing linguistic queries on canonical forms. 15 We propose a new way to encode the orthographic information in these two cases (MWTs and noncanonical forms) with two new features: textform, which always contains orthographic data and wordform which always contains a canonical lexical form (see Table 1 The main advantage is that, using features, all information is available in the units used in the syntactic structure and it makes it possible to use these features in any tool (for querying the treebank, for conversion. . . ).",
"cite_spans": [
{
"start": 475,
"end": 477,
"text": "15",
"ref_id": null
}
],
"ref_spans": [
{
"start": 725,
"end": 732,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "2-3 au _ _ 2\u00e0\u00e0 ADP 3 le le DET",
"sec_num": null
},
{
"text": "It might seem appealing to use these features for encoding typos as well. But, there may be conflicts, as shown for the phrase [fr] Le maison: Le must be corrected in La (the gender of maison is feminine) but also be normalised into le. So, we decided to use the feature CorrectForm (already used in other UD treebanks) in case of typos, to express the way it should be written.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "2-3 au _ _ 2\u00e0\u00e0 ADP 3 le le DET",
"sec_num": null
},
{
"text": "In order to avoid having an overly verbose CoNLL file, we propose in practice, to explicitly record textform and wordform only when they are different from the feature form (column 2 in CoNLL). In Table 1 , square brackets are used to show feature values which are not stored in the CoNLL file.",
"cite_spans": [],
"ref_spans": [
{
"start": 197,
"end": 204,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "2-3 au _ _ 2\u00e0\u00e0 ADP 3 le le DET",
"sec_num": null
},
{
"text": "Our approach of the conversion between different syntactic annotations is based on graph rewriting. Each annotation is seen as a graph and the conversion of an annotation into another annotation is performed by applying a sequence of local graph rewriting rules. For this, we use the GREW tool 16 . In Grew, a Grew Rewriting System (GRS) is a set of rewriting rules organized into strategies such that these rules can be ordered, iterated and grouped into packages. 17 Since SUD is richer than UD, a universal UD \u21d2 SUD GRS can only approximate the correct SUD annotation due to the lack of information in the UD annotation, and the adaptation of the GRS to each language is crucial.",
"cite_spans": [
{
"start": 294,
"end": 296,
"text": "16",
"ref_id": null
},
{
"start": 466,
"end": 468,
"text": "17",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The conversion UD \u21d2 SUD",
"sec_num": "4"
},
{
"text": "The universal UD \u21d2 SUD system has five main tasks to perform:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The universal conversion UD \u21d2 SUD",
"sec_num": "4.1"
},
{
"text": "1. Replacing UD dependency labels with SUD dependency labels.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The universal conversion UD \u21d2 SUD",
"sec_num": "4.1"
},
{
"text": "adpositional phrases, subordinate clauses, and verb-auxiliary pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Reversing some dependencies between function words and lexical words to change the heads of",
"sec_num": "2."
},
{
"text": "3. Shifting the source of some dependencies as the result of reversing dependencies.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Reversing some dependencies between function words and lexical words to change the heads of",
"sec_num": "2."
},
{
"text": "4. Attaching the right dependents of coordinations to the rightmost conjunct, whereas in UD they are attached to the leftmost conjunct, the coordination head (see Section 2.4).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Reversing some dependencies between function words and lexical words to change the heads of",
"sec_num": "2."
},
{
"text": "Transforming bouquets of coordinated elements into sequences, marking embedded coordinations with the emb extension added to conj relations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "These tasks are not independent of each other and although they can most often be carried out in any order, their forms depend on this order and sometimes one order is more relevant than another. The universal UD \u21d2 SUD GRS contains 89 rules grouped into 20 packages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "As said above, a conversion of an UD annotation into a SUD annotation is necessarily approximate. The lack of information is particularly problematic in four cases:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "1. when several function words depend on the same lexical word in UD (see Section 2.2), 2. when a UD dependency from a lexical word to a function word has to be reversed, some of its dependents has to be transferred to the function word but there is usually no indication on which dependents have to be transferred, 3. to decide whether left dependents of a coordination head are dependent of the whole coordination or of the head alone, 4. when idioms have an internal structure, which is not represented in UD and cannot be recovered in the conversion.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "For the first problem, we assume that the further a function word is from the content word, the higher it is in the dependency structure, but there are cases that cannot be solved by such an heuristic, as shown with auxiliaries in Wolof and German (Section 2.2), and our conversion necessarily produces errors without a language-by-language customization.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "For the second problem, we have implemented some rules for specific cases: for instance, the subject moves to the auxiliary, while the complements stay on the lexical verb. For modifiers, it is more complex and we resort to word order, preserving the projectivity as much as possible, but only a language-specific and lexicon-based conversion could ensure a perfect structure.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "For the third problem, we use heuristics to decide. For example, if the leftmost conjunct of a coordination has a subject to its left and the other conjuncts have no subject, we consider that the subject is shared by all conjuncts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "For the fourth problem, UD flat structures of idioms are converted into SUD flat structures.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "5.",
"sec_num": null
},
{
"text": "We have presented default solutions that minimize errors in the UD \u21d2 SUD conversion. By customizing the GRS for specific languages, we can further reduce the errors. For the case of several function words depending on the same lexical word, our architecture allows us to attribute a feature level to dependencies being to reverse with a value that gives its priority in the reversing process. For instance in French, cop dependencies are assigned a bigger priority than aux dependencies, which means that in case of competition cop dependencies must be reversed before aux dependencies. Such a rule is needed when the predicate has been extracted as in Fig. 9 . For the moment, the UD \u21d2 SUD conversion has been customized for French and Wolof. For French, a lexicon of modifiers that must move to the auxiliary has been developed. For Wolof, the level mechanism is used to take into account the case described in Section 2.2.",
"cite_spans": [],
"ref_spans": [
{
"start": 653,
"end": 659,
"text": "Fig. 9",
"ref_id": "FIGREF11"
}
],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "The universal SUD \u21d2 UD GRS must perform the same tasks as the universal UD \u21d2 SUD GRS (see Section 4.1), but in the opposite direction, and the rule order is not the same. It currently contains 94 rules grouped into 20 packages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "In UD, the label of a dependency takes into account not only the syntactic function realized by the dependency but possibly the POS of the governor and the POS of the dependent. For example, the SUD mod dependency is converted into a UD advmod, amod, nmod, obl or advcl dependency, and knowledge of governor's and dependent's POS does not always identify the dependency label. In specific contexts, some words are not used in their usual syntactic function and this use depends on the language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "For example, a SUD mod dependency from a verb to a noun is by default a UD obl dependency, but there are exceptions. Examples (2) from UD-ENGLISH-GUM illustrate respectively the two cases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "(2) (a) Many times prideful people have a serious 'my-way's-the-only-way' attitude.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "(b) An undistinguished student and an unskilled cricketer, he did represent the school.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "In SUD, the dependencies have \u2192 times (2a) and represent \u2192 student (2b) are both mod dependencies. The first one becomes an obl dependency in UD, whereas the second one becomes an advcl dependency because the noun phrase an undistinguished student and an unskilled cricketer is considered as a clause with an ellipsis equivalent to being an undistinguished student and an unskilled cricketer.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "Since there is no universal criterion to distinguish the two cases, we have designed a SUD \u21d2 UD conversion rule, which transforms mod relations into advcl relations if the governor is a verb and the dependent is a non-temporal nominal preceding the verb, but such a rule only works for certain languages, French and English in particular. Since the rule requires distinguishing temporal nominals, we chose to link the conversion rule to a lexicon. Another solution would have been to mark temporal nominals in the corpus (as it is done in some treebanks with the tmod extension).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "Another difficulty in the SUD \u21d2 UD conversion is that the definition of some UD relations takes into account semantic properties. In particular, the relation between a verb and an argument clause is denoted xcomp if the subject of the object clause is controlled by the main verb. Otherwise, the relation is denoted ccomp. Consider the following examples extracted from the FRENCH-GSD corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "(3) (a) les mesures visant\u00e0 d\u00e9velopper l'accord 'measures (aiming) to develop the agreement' (b) Le tourisme commence\u00e0 se d\u00e9velopper. 'Tourism is starting to develop.'",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "The UD annotation of (3a) includes a visant \u2212[ccomp]\u2192 d\u00e9velopper dependency, whereas the UD annotation of (3b) includes a commence \u2212[xcomp]\u2192 d\u00e9velopper dependency. In SUD, both dependencies are denoted comp:obl according to the fact that the definition of syntactic relations is based on positional paradigms (see Section 2.1). To choose between xcomp and ccomp in the SUD \u21d2 UD conversion of these relations, a way is to use a lexicon of control verbs and a conversion rule, which uses this lexicon. A major drawback is that you it should be done for each language separately. To avoid this drawback, another way is to mark the relations of the control verbs to the concerned argument with a special feature. That is what is done with the extension @x in the SUD annotation. The method we just described for improving the UD annotation resulting from the conversion can be used to take into account the idiosyncrasies of some languages. The diverse interests behind treebank development regularly lead to some idiosyncratic enrichment of the annotation. UD responds to this need with the option of adding language (or treebank) specific subrelations and features, and SUD naturally follows this approach. If and only if the SUD treebank developers have added new subrelations or features and want them to be taken into account when translating to UD, they must add these idiosyncratic rules to the universal SUD \u21d2 UD GRS.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "For the time being, the SUD \u21d2 UD conversion has been customized for French (by inserting two rule packages in the universal GRS), Naija, and Beja. For Beja, which is a strongly head-final language, coordinations have been analyzed in SUD by head-final conj relations (see (Kanayama et al., 2018) for a similar analysis in Japanese and Korean). As conj relations must always be head-initial in UD, we have added an ad hoc conversion to a dep:conj relation, but it is possible to customize the conversion in another way, for instance, by reversing the direction of conj relations.",
"cite_spans": [
{
"start": 272,
"end": 295,
"text": "(Kanayama et al., 2018)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "On the train part of SUD FRENCH-GSD, the language-specific customization fixes 1.2% of the 400,220 dependencies in the UD \u21d2 SUD direction and 0.4% in the other direction (i.e threes times less, which is not surprising). The low percentage shows that idiosyncratic customization can be ignored at first when starting a SUD treebank as the universal SUD \u21d2 UD conversion amply does the trick.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "The lack of gold annotation in UD and SUD does not allow a direct evaluation of our SUD \u21d2 UD and UD \u21d2 SUD conversion tools, but we have done an indirect evaluation, using double conversion. The SUD \u21d2 UD conversion followed by the UD \u21d2 SUD conversion on the SUD FRENCH-GSD corpus gives 6231 different dependencies out of 400,220 dependencies, i.e. 1.56% of the total, between the resulting annotation and the initial annotation. The UD \u21d2 SUD conversion followed by the SUD \u21d2 UD conversion on the UD FRENCH-GSD corpus gives 90 different dependencies out of 400,220 dependencies, i.e. 0.02% of the total, between the resulting annotation and the initial annotation. This highlights that SUD is richer than UD. A closer look at the differences in the first double conversion shows that 82% are due to the flattening of idiomatic structures in UD, the rest coming from the ambiguity of UD in the dependencies on coordinations and nuclei.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "6 A practical guide for the development of a SUD treebank Several tools are already available for helping the start of a new treebank in SUD.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "GREW-MATCH (Guillaume, 2021) is an on-line graph query tool which is dedicated to linguistic structures and in particular dependency graphs. It can be used during annotation in order to have a transversal view on already annotated data which helps to take consistent decisions on new annotations. During the maintenance of the corpora, it also helps to ensure global consistency and to do error-mining. GREW-MATCH can be easily coupled with the two UD \u21d4 SUD conversion systems and gives access to the parallel view of both annotation schemes: you can search in SUD and see also the UD corresponding structure and the reverse.",
"cite_spans": [
{
"start": 11,
"end": 28,
"text": "(Guillaume, 2021)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "The whole annotation process can be managed through the ARBORATORGREW 18 annotation platform (Guibon et al., 2020) : user handling, access control, manual edition of the data. . . GREW-MATCH requests are also available through the ARBORATORGREW platform and detected inconsistencies can be corrected directly. In ARBORATORGREW, the user have also access to some specific tools:",
"cite_spans": [
{
"start": 93,
"end": 114,
"text": "(Guibon et al., 2020)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "\u2022 A lexicon-based view of the treebank for detecting inconsistencies in the annotation of the different occurrences of a form or a lemma",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "\u2022 Automatic graph transformation for the correction of regular errors or for applying changes in the annotation decisions (in the sentence-based as well as in the lexicon-based view of the treebank)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "A validation page for SUD treebank is available through GREW-MATCH. It checks that structures are well-formed and helps keeping consistent decisions during the annotation process. Through the conversion to UD, the validation of the UD data adds another layer of verification. Comparing the output of the double conversion SUD \u21d2 UD \u21d2 SUD with the original data is an additional way to obtain valuable feedback on the annotated data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "It should be noted that in the particular case where a UD treebank already exists, the universal conversion should be tested to verify that the internal structure of the nuclei matches the expected structure. If this is not the case, the conversion may need to be customized as explained in Section 4.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Customization of the UD \u21d2 SUD conversion",
"sec_num": "4.2"
},
{
"text": "SUD is not just a richer and easier annotation scheme than UD that can automatically be converted to UD. Importantly, SUD's distributional criteria facilitate and homogenize the annotation choices, resulting in treebanks that enable typological measures across languages. Also, a rich set of tools is available that allow for a kick-start in annotation of raw or partially annotated data. Several SUD treebanks exist that can serve as examples, with more in the pipeline. Go SUD!",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
},
{
"text": "The lossless conversion might require language-specific rules, see Section 4.4 We use catastrophe here in a strictly mathematical sense of Thom's catastrophe theory(Saunders, 1980), i.e. a brutal structural change in a continuum. In the case of annotation, this boils down to very similar constructions ending up with very different syntactic structures, see(Gerdes and Kahane, 2016) for details.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "When comparing the distribution of two units, we mainly use our intuition. For tricky cases, we also observe the actual distribution in our corpora, but nothing is completely currently formalized.6 is expected can also be the dependent of a noun, but only if it combine with a relativizer (the guy that is expected at noon) and, in this case, it is the relativizer that is head of the relative clause, because the relativizer change the distribution of is expected.7 When saying that A' can commute with A, we are only considering the commutation in the context of B. In other words, this means that A'B is a valid combination and that A and A' exclude each other in this context (i.e. AA'B is not valid).8 Note that the analysis of comparative complements is erroneous in English UD treebanks: than in Mesoamerica should depend on more and not on obvious, because more than in Mesoamerica is a valid sub-unit of the sentence and not *obvious than in Mesoamerica.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "UD uses the obl relation for all adpositional phrases depending on a verb, but for clauses depending on a verb, a distinction is made between complements (ccomp or xcomp) and modifiers (advcl for adverbial clauses).10 Contrary to UD, SUD does not have an expl relation for expletives. We consider that it in it is impossible to do that, is above all a normal subject and is analysed as subj@expl.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "English has adverbs taking a to VERB complement, such as up, next, about, or prior, but there are no subordinating conjunctions with this valency.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://grew.fr/ 17 All GRS described in this section are available on https://github.com/surfacesyntacticud/tools/ tree/master/converter",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The conversion SUD \u21d2 UD Since SUD is richer than UD, we should have no difficulty in designing a universal GRS that converts any SUD annotation of a corpus in any language into an UD annotation. This is globally true but conversion sometimes requires adaptation to the specificity of the language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://arborator.github.io",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "8 as a subordinating conjunction (SCONJ, as in all occurrences of in order in UD ENGLISH-GUM, version 2.8), which is surprising to say the least. 13 Here, orthographic means the actually observed letters in input text",
"authors": [],
"year": null,
"venue": "has been analyzed in the UD analysis of Fig",
"volume": "14",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "In order to keep a function word status for in order, in has been analyzed in the UD analysis of Fig. 8 as a subordinating conjunction (SCONJ, as in all occurrences of in order in UD ENGLISH-GUM, version 2.8), which is surprising to say the least. 13 Here, orthographic means the actually observed letters in input text. 14 https://universaldependencies.org/format.html#words-tokens-and-empty-nodes 15 Note that this canonical form may not be trivial to recover. In French, diacritics are optional on upper-case letters, and an A as the first word can be either the preposition\u00e0 (ex:\u00e0 qui tu penses ? 'who are you thinking of?') or a verbal form a (a-t-il choisi ? 'has he chosen?').",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Les particules verbales du wolof et leur combinatoire syntaxique et topologique. Bulletin de la Soci\u00e9t\u00e9 de Linguistique de Paris",
"authors": [
{
"first": "Olivier",
"middle": [],
"last": "Bond\u00e9elle",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "115",
"issue": "",
"pages": "391--465",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Olivier Bond\u00e9elle and Sylvain Kahane. 2021. Les particules verbales du wolof et leur combinatoire syntaxique et topologique. Bulletin de la Soci\u00e9t\u00e9 de Linguistique de Paris, 115(1):391-465, January.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Universal Dependencies",
"authors": [
{
"first": "Marie-Catherine",
"middle": [],
"last": "De Marneffe",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
},
{
"first": "Joakim",
"middle": [],
"last": "Nivre",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Zeman",
"suffix": ""
}
],
"year": 2021,
"venue": "Computational Linguistics",
"volume": "47",
"issue": "2",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marie-Catherine de Marneffe, Christopher D. Manning, Joakim Nivre, and Daniel Zeman. 2021. Universal De- pendencies. Computational Linguistics, 47(2):255-308, 07.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "The greenbergian word order correlations. Language",
"authors": [
{
"first": "Matthew",
"middle": [
"S"
],
"last": "Dryer",
"suffix": ""
}
],
"year": 1992,
"venue": "",
"volume": "68",
"issue": "",
"pages": "81--138",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew S. Dryer. 1992. The greenbergian word order correlations. Language, 68(1):81-138.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Defining dependencies (and constituents)",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": ""
}
],
"year": 2013,
"venue": "Frontiers in Artificial Intelligence and Applications",
"volume": "258",
"issue": "",
"pages": "1--25",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim Gerdes and Sylvain Kahane. 2013. Defining dependencies (and constituents). Frontiers in Artificial Intelli- gence and Applications, 258:1-25.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Dependency annotation choices: Assessing theoretical and practical issues of universal dependencies",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": ""
}
],
"year": 2016,
"venue": "LAW X (2016) The 10th Linguistic Annotation Workshop",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim Gerdes and Sylvain Kahane. 2016. Dependency annotation choices: Assessing theoretical and practical issues of universal dependencies. In LAW X (2016) The 10th Linguistic Annotation Workshop: 131.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "SUD or surface-syntactic Universal Dependencies: An annotation scheme near-isomorphic to UD",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": ""
},
{
"first": "Bruno",
"middle": [],
"last": "Guillaume",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": ""
},
{
"first": "Guy",
"middle": [],
"last": "Perrier",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the Second Workshop on Universal Dependencies (UDW 2018)",
"volume": "",
"issue": "",
"pages": "66--74",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim Gerdes, Bruno Guillaume, Sylvain Kahane, and Guy Perrier. 2018. SUD or surface-syntactic Universal Dependencies: An annotation scheme near-isomorphic to UD. In Proceedings of the Second Workshop on Uni- versal Dependencies (UDW 2018), pages 66-74, Brussels, Belgium, November. Association for Computational Linguistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Improving surface-syntactic Universal Dependencies (SUD): MWEs and deep syntactic features",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": ""
},
{
"first": "Bruno",
"middle": [],
"last": "Guillaume",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Kahane",
"suffix": ""
},
{
"first": "Guy",
"middle": [],
"last": "Perrier",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 18th International Workshop on Treebanks and Linguistic Theories (TLT, SyntaxFest 2019)",
"volume": "",
"issue": "",
"pages": "126--132",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim Gerdes, Bruno Guillaume, Sylvain Kahane, and Guy Perrier. 2019. Improving surface-syntactic Universal Dependencies (SUD): MWEs and deep syntactic features. In Proceedings of the 18th International Workshop on Treebanks and Linguistic Theories (TLT, SyntaxFest 2019), pages 126-132, Paris, France, August. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "When Collaborative Treebank Curation Meets Graph Grammars",
"authors": [
{
"first": "Ga\u00ebl",
"middle": [],
"last": "Guibon",
"suffix": ""
},
{
"first": "Marine",
"middle": [],
"last": "Courtin",
"suffix": ""
},
{
"first": "Kim",
"middle": [],
"last": "Gerdes",
"suffix": ""
},
{
"first": "Bruno",
"middle": [],
"last": "Guillaume",
"suffix": ""
}
],
"year": 2020,
"venue": "LREC 2020 -12th Language Resources and Evaluation Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ga\u00ebl Guibon, Marine Courtin, Kim Gerdes, and Bruno Guillaume. 2020. When Collaborative Treebank Curation Meets Graph Grammars. In LREC 2020 -12th Language Resources and Evaluation Conference, Marseille, France, May.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Graph Matching and Graph Rewriting: GREW tools for corpus exploration, maintenance and conversion",
"authors": [
{
"first": "Bruno",
"middle": [],
"last": "Guillaume",
"suffix": ""
}
],
"year": 2021,
"venue": "EACL 2021 -16th conference of the European Chapter of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bruno Guillaume. 2021. Graph Matching and Graph Rewriting: GREW tools for corpus exploration, maintenance and conversion. In EACL 2021 -16th conference of the European Chapter of the Association for Computational Linguistics, Kiev/Online, Ukraine, April.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Coordinate structures in Universal Dependencies for head-final languages",
"authors": [
{
"first": "Hiroshi",
"middle": [],
"last": "Kanayama",
"suffix": ""
},
{
"first": "Na-Rae",
"middle": [],
"last": "Han",
"suffix": ""
},
{
"first": "Masayuki",
"middle": [],
"last": "Asahara",
"suffix": ""
},
{
"first": "Jena",
"middle": [
"D"
],
"last": "Hwang",
"suffix": ""
},
{
"first": "Yusuke",
"middle": [],
"last": "Miyao",
"suffix": ""
},
{
"first": "Jinho",
"middle": [
"D"
],
"last": "Choi",
"suffix": ""
},
{
"first": "Yuji",
"middle": [],
"last": "Matsumoto",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the Second Workshop on Universal Dependencies (UDW 2018)",
"volume": "",
"issue": "",
"pages": "75--84",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hiroshi Kanayama, Na-Rae Han, Masayuki Asahara, Jena D. Hwang, Yusuke Miyao, Jinho D. Choi, and Yuji Matsumoto. 2018. Coordinate structures in Universal Dependencies for head-final languages. In Proceedings of the Second Workshop on Universal Dependencies (UDW 2018), pages 75-84, Brussels, Belgium, November. Association for Computational Linguistics.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Dependency Syntax: Theory and Practice",
"authors": [
{
"first": "Igor",
"middle": [],
"last": "Mel",
"suffix": ""
},
{
"first": "'",
"middle": [],
"last": "\u010cuk",
"suffix": ""
}
],
"year": 1988,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Igor Mel'\u010duk. 1988. Dependency Syntax: Theory and Practice. Albany, N.Y.: The SUNY Press.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "A universal part-of-speech tagset",
"authors": [
{
"first": "Slav",
"middle": [],
"last": "Petrov",
"suffix": ""
},
{
"first": "Dipanjan",
"middle": [],
"last": "Das",
"suffix": ""
},
{
"first": "Ryan",
"middle": [],
"last": "Mcdonald",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)",
"volume": "",
"issue": "",
"pages": "2089--2096",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. A universal part-of-speech tagset. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12), pages 2089-2096, Istanbul, Turkey, May. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Approche\u00e9nonciative du syst\u00e8me verbal: le cas du wolof",
"authors": [
{
"first": "St\u00e9phane",
"middle": [],
"last": "Robert",
"suffix": ""
}
],
"year": 1991,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "St\u00e9phane Robert. 1991. Approche\u00e9nonciative du syst\u00e8me verbal: le cas du wolof. CNRS Editions.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "An introduction to catastrophe theory",
"authors": [
{
"first": "Saunders",
"middle": [],
"last": "Peter Timothy",
"suffix": ""
}
],
"year": 1980,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Peter Timothy Saunders. 1980. An introduction to catastrophe theory. Cambridge University Press.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "La valence: l'approche pronominale et son application au lexique verbal",
"authors": [
{
"first": "Karel",
"middle": [],
"last": "Van Den Eynde",
"suffix": ""
},
{
"first": "Piet",
"middle": [],
"last": "Mertens",
"suffix": ""
}
],
"year": 2003,
"venue": "Journal of French language studies",
"volume": "13",
"issue": "1",
"pages": "63--104",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Karel Van den Eynde and Piet Mertens. 2003. La valence: l'approche pronominale et son application au lexique verbal. Journal of French language studies, 13(1):63-104.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Reusable tagset conversion using tagset drivers",
"authors": [
{
"first": "Daniel",
"middle": [],
"last": "Zeman",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Daniel Zeman. 2008. Reusable tagset conversion using tagset drivers. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08), Marrakech, Morocco, May. European Lan- guage Resources Association (ELRA).",
"links": null
}
},
"ref_entries": {
"FIGREF1": {
"text": "UD and SUD analysis of Sentence Nowhere in the Americas was this more obvious than in Mesoamerica.",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF2": {
"text": "SUD analysis of the Wolof sentence Firnde loolu doon na fee\u00f1i 'This sign was to be revealed.'",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF3": {
"text": "seven employees will have received their notice by then.'",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF4": {
"text": "UD and SUD analysis of Sentence (1)",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF5": {
"text": "UD and SUD analysis of Allez-y en confiance ! 'Go there with confidence!'",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF6": {
"text": "UD and SUD analysis of De qui se moque-t-on ? 'Who are we kidding?' UD and SUD analysis of Look at that.",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF7": {
"text": "UD and SUD annotation of Good food and coffee with a nice atmosphere",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF9": {
"text": "UD and SUD annotation of Multi-Word Expressions",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF11": {
"text": "UD (left) and SUD (right) trees for ce que nous avons\u00e9t\u00e9 'what we have been'",
"uris": null,
"type_str": "figure",
"num": null
},
"TABREF0": {
"text": "for examples). Examples on the usage of features textform, wordform and CorrectForm.",
"html": null,
"content": "<table><tr><td/><td>form</td><td colspan=\"2\">lemma textform</td><td colspan=\"2\">wordform CorrectForm</td></tr><tr><td colspan=\"3\">[fr] au\u00e0\u00e0 le le</td><td>au</td><td>[\u00e0] [le]</td><td/></tr><tr><td>[en] wanna</td><td>want to</td><td>want to</td><td>wanna</td><td>[want] [to]</td><td/></tr><tr><td>[en] The</td><td>The</td><td>the</td><td>[The]</td><td>the</td><td/></tr><tr><td>[fr] Le maison</td><td>Le</td><td>le</td><td>[Le]</td><td>le</td><td>La</td></tr><tr><td>[en] egg plant</td><td>egg plant</td><td>egg plant</td><td>[egg] [plant]</td><td>eggplant</td><td/></tr><tr><td colspan=\"3\">[en] NEEEVERR NEEEVERR never</td><td colspan=\"2\">[NEEEVERR] neeeverr</td><td>never</td></tr></table>",
"type_str": "table",
"num": null
}
}
}
} |