File size: 73,617 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:12:34.692947Z"
},
"title": "Are BERTs Sensitive to Native Interference in L2 Production?",
"authors": [
{
"first": "Zixin",
"middle": [],
"last": "Tang",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Prasenjit",
"middle": [],
"last": "Mitra",
"suffix": "",
"affiliation": {},
"email": "pmitra@psu.edu"
},
{
"first": "David",
"middle": [],
"last": "Reitter",
"suffix": "",
"affiliation": {},
"email": "reitter@google.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "With the essays part from The International Corpus Network of Asian Learners of English (ICNALE) and the TOEFL11 corpus, we finetuned neural language models based on BERT to predict English learners' native languages. Results showed neural models can learn to represent and detect such native language impacts, but multilingually trained models have no advantage in doing so.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "With the essays part from The International Corpus Network of Asian Learners of English (ICNALE) and the TOEFL11 corpus, we finetuned neural language models based on BERT to predict English learners' native languages. Results showed neural models can learn to represent and detect such native language impacts, but multilingually trained models have no advantage in doing so.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "With globalization progress, more and more people start acquiring more than one language. Therefore, computational models and theories to better understand multilingualism among machines and humans start receiving more and more attention. Even though language models (LMs) succeeded in a wide range of tasks in natural language understanding and processing (i.e. Enguehard et al., 2017; Linzen and Leonard, 2018; Mueller et al., 2020) , how to do such models process and represent languages and knowledge remain unclear. We also have few ideas about these models' abilities when facing various types of language input, such as input from non-native speakers or speakers with lower proficiency. In this study, we implemented state-of-the-art neural LMs to predict non-native (L2) speakers with different language backgrounds through native language identification tasks. Furthermore, we investigated whether cross-lingual training can help identify a language learner's writing patterns, which his native languages (L1s) can cause interference. This study wants to expand current knowledge on representations and crosslingual components in transformer-based LMs.",
"cite_spans": [
{
"start": 363,
"end": 386,
"text": "Enguehard et al., 2017;",
"ref_id": "BIBREF6"
},
{
"start": 387,
"end": 412,
"text": "Linzen and Leonard, 2018;",
"ref_id": "BIBREF16"
},
{
"start": 413,
"end": 434,
"text": "Mueller et al., 2020)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "With developments in computer science, data sciences, and cognitive and language sciences, artificial language models (LMs) based on language representations and embeddings have received attention in multiple areas. Current deep neural LMs based on various architectures can capture syntactic information through natural language input and utilize language knowledge to complete different tasks, such as identifying specific sentence structures (Enguehard et al., 2017; Kelly et al., 2020; Marvin and Linzen, 2018) and predicting grammar acceptability (Warstadt et al., 2019) . Beyond training on a single language, multilingual neural language models could process cross-lingual tasks such as language identification and translation. Studies found that language representations in these types of models are transferable or overlapped across languages to create universal representations of grammatical structures (Chi et al., 2020) . Models can also develop language-neutral components for better word alignment and improve performance on several cross-linguistic tasks (Libovick\u1ef3 et al., 2019) . However, studies indicated that these LMs have lower performance in crosslinguistic tasks such as machine translations (Libovick\u1ef3 et al., 2019) . Comparing to monolingual models, multilingual LMs have relatively lower performance and unbalanced performance in syntacticrelated tasks between English and other languages (Mueller et al., 2020) .",
"cite_spans": [
{
"start": 445,
"end": 469,
"text": "(Enguehard et al., 2017;",
"ref_id": "BIBREF6"
},
{
"start": 470,
"end": 489,
"text": "Kelly et al., 2020;",
"ref_id": "BIBREF12"
},
{
"start": 490,
"end": 514,
"text": "Marvin and Linzen, 2018)",
"ref_id": "BIBREF19"
},
{
"start": 552,
"end": 575,
"text": "(Warstadt et al., 2019)",
"ref_id": "BIBREF23"
},
{
"start": 914,
"end": 932,
"text": "(Chi et al., 2020)",
"ref_id": "BIBREF3"
},
{
"start": 1071,
"end": 1095,
"text": "(Libovick\u1ef3 et al., 2019)",
"ref_id": "BIBREF15"
},
{
"start": 1217,
"end": 1241,
"text": "(Libovick\u1ef3 et al., 2019)",
"ref_id": "BIBREF15"
},
{
"start": 1417,
"end": 1439,
"text": "(Mueller et al., 2020)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Even though transformer-based LMs showed impressive abilities in multiple tasks, few studies tried to analyze whether and how multilingual LMs identify and distinguish writers' native interference in their L2 production caused by their L1s. One task for detecting such interference across speakers is the native language identification task, which predicts a speaker's L1 using his language production in L2. By sharing representations and transferring knowledge across languages (Putnam et al., 2018; Hartsuiker et al., 2004) , multilingual speakers can have different structures, preferences, and variations for language comprehension and production than monolingual speakers, result in cross-linguistic behaviors such as cross-lingual priming effects (Bernolet et al., 2007; Hartsuiker et al., 2016 Hartsuiker et al., , 2004 Gries and Kootstra, 2017) and syntactic preferences in language processing (Hsiao and Gibson, 2003) . Previously, statistical-based language models (N-gram models) dominated the tasks (Malmasi et al., 2017) , as statistical models are more sensitive in detecting syntactic and morpho-syntactic patterns. Embedding models, on the other hand, are more sensitive in capturing semantic and lexical information and less sensitive in classifying syntactic patterns and differences (Vajjala and Banerjee, 2017).",
"cite_spans": [
{
"start": 480,
"end": 501,
"text": "(Putnam et al., 2018;",
"ref_id": "BIBREF21"
},
{
"start": 502,
"end": 526,
"text": "Hartsuiker et al., 2004)",
"ref_id": "BIBREF9"
},
{
"start": 754,
"end": 777,
"text": "(Bernolet et al., 2007;",
"ref_id": "BIBREF1"
},
{
"start": 778,
"end": 801,
"text": "Hartsuiker et al., 2016",
"ref_id": "BIBREF8"
},
{
"start": 802,
"end": 827,
"text": "Hartsuiker et al., , 2004",
"ref_id": "BIBREF9"
},
{
"start": 828,
"end": 853,
"text": "Gries and Kootstra, 2017)",
"ref_id": "BIBREF7"
},
{
"start": 903,
"end": 927,
"text": "(Hsiao and Gibson, 2003)",
"ref_id": "BIBREF10"
},
{
"start": 1012,
"end": 1034,
"text": "(Malmasi et al., 2017)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Given the high performance in syntactic-related tasks (Devlin et al., 2018; Mueller et al., 2020) , comparing to previous embedding-based models, transformer-based models may perform better in detecting native language interference. Such advantages can help models better identify specific patterns in L2 learners. Furthermore, LMs with multilingual pre-training may be benefited in identifying different non-native speakers using additional knowledge across languages. However, existed results contradicted such assumptions, indicating that multilingual LMs generally performed worse than monolingual models in the same tasks (Mueller et al., 2020) . To better understand transformerbased LMs, multilingual pre-training experiences, and language representations in LMs, our study aims to use native language identification task to answer the following research questions:",
"cite_spans": [
{
"start": 54,
"end": 75,
"text": "(Devlin et al., 2018;",
"ref_id": "BIBREF5"
},
{
"start": 76,
"end": 97,
"text": "Mueller et al., 2020)",
"ref_id": "BIBREF20"
},
{
"start": 627,
"end": 649,
"text": "(Mueller et al., 2020)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "1. Can transformer-based LMs detect native interference among L2 learners with different L1s? 2. Can multilingual-LMs transfer knowledge from other languages to gain advantages in such native language prediction tasks?",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "3 Methods",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "We used two different corpora in this study: the TOEFL11 corpus (Blanchard et al., 2013) used in previous native language identification shared task (Malmasi et al., 2017) , and the written essays from The International Corpus Network of Asian Learners of English (ICNALE) 1 (Ishikawa, 2013). The TOEFL11 corpus contains TOEFL essays written by English learners with 11 native languages, while ICNALE corpus contains short essays written by 1 The ICNALE corpus: http://language.sakura.ne.jp/icnale/ English learners from 10 regions and English native speakers. We used document-level input to capture general writing patterns and styles among writers. We excluded essays with less than 50 words for future processes. Both corpora have held-out testing data sets: for TOEFL11, the development set serves as the held-out testing set; for ICNALE, we randomly selected 20% essays to serve as the testing set. Table 1 is the detailed description for each corpus. Two potential factors could interfere with the final result: speakers' proficiency in the target language (English) and the semantic/lexical information inside essays. To minimize effects from the proficiency, we further developed two types of subsets to examine whether and how proficiency influences transformer-based LMs in native language identification tasks during the fine-tuning process:",
"cite_spans": [
{
"start": 64,
"end": 88,
"text": "(Blanchard et al., 2013)",
"ref_id": "BIBREF2"
},
{
"start": 149,
"end": 171,
"text": "(Malmasi et al., 2017)",
"ref_id": "BIBREF18"
},
{
"start": 441,
"end": 442,
"text": "1",
"ref_id": null
}
],
"ref_spans": [
{
"start": 905,
"end": 912,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Corpora and data pre-processing",
"sec_num": "3.1"
},
{
"text": "\u2022 Proficiency subsets (B1 and Mid set): serve as proficiency control during the fine-tuning process. These subsets included essays written by intermediate level (B1) English learners in ICNALE and speakers with median TOEFL scores (Mid) from TOEFL11. \u2022 Balanced subset (BA set): serve as data input control during the fine-tuning process to provide better comparisons to proficiency subsets. These subsets had the same number of essays as the proficiency subsets, formed by randomly selected essays from the original fine-tuning set. We also developed another version of testing sets by randomly shuffling the original texts in the essays. Through the shuffling process, we minimized the impact of syntactic information, leaving most of the lexical and semantic information available inside the testing sets. This shuffling test set can examine whether LMs rely on lexical and semantic information to process native language identification tasks.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Corpora and data pre-processing",
"sec_num": "3.1"
},
{
"text": "For models fine-tuned on ICNALE data, essays were converted into measures of syntax complexity using syntax complexity analyzer developed by Lu (2010) . The baseline model is a multiple logistic regression model developed with outcome measures. Since there is no previous identification task involves ICNALE, we did not include any other models for comparison.",
"cite_spans": [
{
"start": 141,
"end": 150,
"text": "Lu (2010)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Baseline and models for comparison",
"sec_num": "3.2"
},
{
"text": "For models fine-tuned on TOEFL11 data, we used the baseline unigram model mentioned in Malmasi et al. (2017) . We also included some models mentioned in Malmasi et al. (2017) ",
"cite_spans": [
{
"start": 87,
"end": 108,
"text": "Malmasi et al. (2017)",
"ref_id": "BIBREF18"
},
{
"start": 153,
"end": 174,
"text": "Malmasi et al. (2017)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Baseline and models for comparison",
"sec_num": "3.2"
},
{
"text": "For this study, we fine-tuned pre-trained BERT models to predict a speaker's L1 background through different sets of corpora mentioned above. Introduced in 2018, the BERT model is still considered a top pre-trained model in current natural language processing studies. One advantage of using BERT comparing to other non-pre-trained models is its rich knowledge in lexical information, syntactic representations, and semantic networks, which comes from the large scale of training data (Devlin et al., 2018) . For this study, two different BERT models will be fine-tuned with multiple datasets: the 12-layer English BERT_base model (BERTbase) and the 12-layer multilingual BERT_base model (mBERT-base).",
"cite_spans": [
{
"start": 485,
"end": 506,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model fine-tuning",
"sec_num": "3.3"
},
{
"text": "We fine-tuned all models with an additional linear classification layer through the Transformer package. The classification layer is directly finetuned on the final [CLS] embedding with the default Adams optimizer, a learning rate of 2e-5, and an early-stop of 8 epoch. The performance of each model was reported with F-scores using the heldout testing set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model fine-tuning",
"sec_num": "3.3"
},
{
"text": "Overall, all models with different architectures and fine-tuning data sets captured some features among speakers with different L1 backgrounds, as accuracy is better than random guessing and baseline models. Table 2 and Table 3 showed the results of models trained on ICNALE and TOEFL11 corpus. Performance. The models fine-tuned with TOEFL11 data did not beat the state-of-the-art native language identifier using statistical models. For models fine-tuned with ICNALE data, controlling learners' proficiency did not critically impact models in identification. This showed that speaker proficiency did not strongly interfere with language background identification, indicating the robustness of native language interference across learners' L2 written production. Shuffling word orders did impact all models severely in prediction (Table 4) , which showed that BERT models cannot identify and detect native language interference solely with lexical information in the essays. In summary, transformer-based LMs can capture and involve representations of word orders, general writing patterns, and sentence structures to detect English learners' native languages, and rely less on lexical or semantic information.",
"cite_spans": [],
"ref_spans": [
{
"start": 208,
"end": 227,
"text": "Table 2 and Table 3",
"ref_id": "TABREF4"
},
{
"start": 831,
"end": 840,
"text": "(Table 4)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4"
},
{
"text": "Multilingual pre-training. When using the Corpus Full Shuffled ICNALE 0.86 (0.87) 0.24 (0.29) TOEFL11 0.72 (0.68) 0.46 (0.26) Table 4 : Performance of BERT (mBERT) models on original and shuffle text. All models were fine-tuned without proficiency control. same data sets, models fine-tuned on mBERT had slightly lower or similar results comparing to models fine-tuned on BERT. Additional knowledge across multiple languages did not help in predicting speakers' syntax background in their L1s.",
"cite_spans": [],
"ref_spans": [
{
"start": 126,
"end": 133,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4"
},
{
"text": "While transformer-based LMs reached impressive results in language processing and generation, we still know little about how LMs acquire, process, and store language representations. Our study used native language identification task to investigate 1) whether transformer-based models (BERT & mBERT) can detect native interference in written production; 2) whether multilingual pre-training help in identifying native interference. Overall, the results were negative, indicating 1) compared to statistical models, transformer-based models are less capable of capturing native impacts and interference, and 2) additional knowledge across multiple languages in mBERT did not help predict such impacts across multilingual speakers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "General Discussion",
"sec_num": "5"
},
{
"text": "Compared to state-of-the-art native language identifiers relying on statistical methods (such as Ngrams) with F-scores above 0.85, the performance of embedding models is still worse. Previous studies (Vajjala and Banerjee, 2017; Jing et al., 2020) argued that such worse predictions from embedding models indicate embedding models might have a stronger ability in capturing semantic information; while statistical-based language models might have better performance in capturing morphosyntactic and syntactic information.",
"cite_spans": [
{
"start": 200,
"end": 228,
"text": "(Vajjala and Banerjee, 2017;",
"ref_id": "BIBREF22"
},
{
"start": 229,
"end": 247,
"text": "Jing et al., 2020)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Statistical vs. embedding models",
"sec_num": "5.1"
},
{
"text": "Another explanation for the current lower performance for our embedding models relates to the knowledge structures in BERTs. Previously, studies indicated that for BERTs, the mid-layers have the best performance in syntactic-related tasks (Kelly et al., 2020) . Such results may indicate that, unlike previous statistical models that reach the highest performance in output layers, embeddingbased models like BERTs may store different types of knowledge across different layers: the deeper the layers are, the more abstract concepts they will store. Therefore, mid-layer embeddings may contain and utilize more syntactic and morphosyntactic information than final-layer embeddings, leading to a better performance in language interference identification. The simplicity of our classification layer may also lead to lower performance across our BERT-based models. Future studies can further investigate how embeddings from different layers of BERTs respond to these similar tasks, and how different classification algorithms change the overall performance.",
"cite_spans": [
{
"start": 239,
"end": 259,
"text": "(Kelly et al., 2020)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Statistical vs. embedding models",
"sec_num": "5.1"
},
{
"text": "Another key result from this study, as mentioned previously, is that cross-lingual pre-training did not help in predicting English learner's L1 background. Based on previous findings of languageneutral components within the models (Libovick\u1ef3 et al., 2019) and existences of universal linguistics (Chi et al., 2020), we expected that models trained with multiple languages should be more sensitive to syntactic preferences or specific patterns in English interfered with by L2 syntactic knowledge. Unfortunately, similar to the study by Mueller et al. (2020), we did not find advantages of cross-lingual pre-training for native language identification tasks. We had two potential explanations for such results: the training data size and the representation structures in mBERT.",
"cite_spans": [
{
"start": 231,
"end": 255,
"text": "(Libovick\u1ef3 et al., 2019)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingualism",
"sec_num": "5.2"
},
{
"text": "Through the pre-training process, mBERT uses a smaller training English corpus for the mBERT compared to the English BERT: English BERT has an addition Book corpus as training data (Devlin et al., 2018) 2 . The smaller English training set may lead to less knowledge and representations for English in mBERT, making it a less proficient \"speaker\" than English BERT. Previous research also found similar deficits related to language resources and training data size. Mueller et al. (2020) found that mBERT models perform best in Englishrelated tasks; Wu and Dredze (2020) found that lower-resource languages had significantly lower performance than higher-resource languages when handling the same tasks.",
"cite_spans": [
{
"start": 181,
"end": 202,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF5"
},
{
"start": 466,
"end": 487,
"text": "Mueller et al. (2020)",
"ref_id": "BIBREF20"
},
{
"start": 550,
"end": 570,
"text": "Wu and Dredze (2020)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingualism",
"sec_num": "5.2"
},
{
"text": "The second explanation for mBERT's lower performance belongs to the structures of knowledge within the models. Even though existing evidence showed the possibilities of forming components and representations across languages within embedding models (Chi et al., 2020; Libovick\u1ef3 et al., 2019; Artetxe et al., 2019) . However, Libovick\u1ef3 et al. (2019) pointed out, these cross-lingual components are still not sufficient enough for higherlevel tasks such as automatic machine translation. Our study provided further evidence on such insufficiency, indicating that while identifying native languages across English learners, knowledge from other languages and the language-general component may not involve, even when interference might appear in writers' original writing.",
"cite_spans": [
{
"start": 249,
"end": 267,
"text": "(Chi et al., 2020;",
"ref_id": "BIBREF3"
},
{
"start": 268,
"end": 291,
"text": "Libovick\u1ef3 et al., 2019;",
"ref_id": "BIBREF15"
},
{
"start": 292,
"end": 313,
"text": "Artetxe et al., 2019)",
"ref_id": "BIBREF0"
},
{
"start": 325,
"end": 348,
"text": "Libovick\u1ef3 et al. (2019)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingualism",
"sec_num": "5.2"
},
{
"text": "To build a better multilingual model, we need further investigations on the multilingual pre-training processes, the cross-linguistic components, and how they interact across languages within multilingual LMs. Even though our models did not reach our expectations, future studies can help better understand why it happened. Specifically, we can focus on three areas: layer-wise embeddings within embedding models, size for pre-training data and models, and cross-lingual representations. For embeddings within models, future studies can use probing tasks for layer-wise embeddings to help us better understand how contextual embedding models process, represent, and utilize syntactic information and knowledge. Second, future studies can compare mBERT with smaller monolingual BERT models to balance out the impact on BERT's richer knowledge in high-resource languages. Lastly, we need more studies and methods to measure the cross-lingual components in multilingual models, which help us better understand multilingualism and cross-lingual knowledge in state-of-the-art LMs. In short, more studies, models, and theories are needed to develop LMs with better abilities in multilingual and cross-linguistic tasks to serve multilingual communities better in practical applications.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Future studies",
"sec_num": "5.3"
},
{
"text": "While transformer-based language models and multilingual language models start dominating the natural language process area, little is known about the knowledge structures within such models. In this study, we explored Bert-like models' performance in identifying a speaker's native language. Our models did not outperform the state-of-the-art native language identifier based on statistical-based models. Multilingual pre-training did not improve the performance in predicting native language, indicating the insufficiencies for detecting writing styles and structural patterns among different multilingual speakers in current LMs. Further explorations are needed to investigate how multilingual LMs store, process, communicate, and transfer representations and knowledge across languages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summary",
"sec_num": "6"
},
{
"text": "https://github.com/googleresearch/bert/blob/master/multilingual.md",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This study was supported by the National Science Foundation (PAC 1734304). We thank Dr. Michael T. Putnam, Dr. M. Alex Kelly, and Saranya Venkatraman for their comments and suggestions on the original manuscript. We also thank the anonymous reviewers for providing helpful comments on the earlier draft.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "On the cross-lingual transferability of monolingual representations",
"authors": [
{
"first": "Mikel",
"middle": [],
"last": "Artetxe",
"suffix": ""
},
{
"first": "Sebastian",
"middle": [],
"last": "Ruder",
"suffix": ""
},
{
"first": "Dani",
"middle": [],
"last": "Yogatama",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1910.11856"
]
},
"num": null,
"urls": [],
"raw_text": "Mikel Artetxe, Sebastian Ruder, and Dani Yo- gatama. 2019. On the cross-lingual transferabil- ity of monolingual representations. arXiv preprint arXiv:1910.11856.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Shared syntactic representations in bilinguals: Evidence for the role of word-order repetition",
"authors": [
{
"first": "Sarah",
"middle": [],
"last": "Bernolet",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Robert",
"suffix": ""
},
{
"first": "Martin J",
"middle": [],
"last": "Hartsuiker",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Pickering",
"suffix": ""
}
],
"year": 2007,
"venue": "Journal of Experimental Psychology: Learning",
"volume": "33",
"issue": "5",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sarah Bernolet, Robert J Hartsuiker, and Martin J Pick- ering. 2007. Shared syntactic representations in bilinguals: Evidence for the role of word-order repe- tition. Journal of Experimental Psychology: Learn- ing, Memory, and Cognition, 33(5):931.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Toefl11: A corpus of non-native english",
"authors": [
{
"first": "Daniel",
"middle": [],
"last": "Blanchard",
"suffix": ""
},
{
"first": "Joel",
"middle": [],
"last": "Tetreault",
"suffix": ""
},
{
"first": "Derrick",
"middle": [],
"last": "Higgins",
"suffix": ""
},
{
"first": "Aoife",
"middle": [],
"last": "Cahill",
"suffix": ""
},
{
"first": "Martin",
"middle": [],
"last": "Chodorow",
"suffix": ""
}
],
"year": 2013,
"venue": "ETS Research Report Series",
"volume": "2013",
"issue": "2",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Daniel Blanchard, Joel Tetreault, Derrick Higgins, Aoife Cahill, and Martin Chodorow. 2013. Toefl11: A corpus of non-native english. ETS Research Re- port Series, 2013(2):i-15.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Finding universal grammatical relations in multilingual bert",
"authors": [
{
"first": "A",
"middle": [],
"last": "Ethan",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Chi",
"suffix": ""
},
{
"first": "Christopher D",
"middle": [],
"last": "Hewitt",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2005.04511"
]
},
"num": null,
"urls": [],
"raw_text": "Ethan A Chi, John Hewitt, and Christopher D Manning. 2020. Finding universal grammatical relations in multilingual bert. arXiv preprint arXiv:2005.04511.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Stacked sentence-document classifier approach for improving native language identification",
"authors": [
{
"first": "Andrea",
"middle": [],
"last": "Cimino",
"suffix": ""
},
{
"first": "Felice",
"middle": [],
"last": "Dell'orletta",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications",
"volume": "",
"issue": "",
"pages": "430--437",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andrea Cimino and Felice Dell'Orletta. 2017. Stacked sentence-document classifier approach for improv- ing native language identification. In Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications, pages 430-437.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1810.04805"
]
},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understand- ing. arXiv preprint arXiv:1810.04805.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Exploring the syntactic abilities of rnns with multi-task learning",
"authors": [
{
"first": "Emile",
"middle": [],
"last": "Enguehard",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Goldberg",
"suffix": ""
},
{
"first": "Tal",
"middle": [],
"last": "Linzen",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1706.03542"
]
},
"num": null,
"urls": [],
"raw_text": "Emile Enguehard, Yoav Goldberg, and Tal Linzen. 2017. Exploring the syntactic abilities of rnns with multi-task learning. arXiv preprint arXiv:1706.03542.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Structural priming within and across languages: A corpusbased perspective",
"authors": [
{
"first": "Th",
"middle": [],
"last": "Stefan",
"suffix": ""
},
{
"first": "Gerrit",
"middle": [],
"last": "Gries",
"suffix": ""
}
],
"year": 2017,
"venue": "Bilingualism: Language and Cognition",
"volume": "20",
"issue": "2",
"pages": "235--250",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stefan Th Gries and Gerrit Jan Kootstra. 2017. Struc- tural priming within and across languages: A corpus- based perspective. Bilingualism: Language and Cognition, 20(2):235-250.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Crosslinguistic structural priming in multilinguals: Further evidence for shared syntax",
"authors": [
{
"first": "J",
"middle": [],
"last": "Robert",
"suffix": ""
},
{
"first": "Saskia",
"middle": [],
"last": "Hartsuiker",
"suffix": ""
},
{
"first": "Maaike",
"middle": [],
"last": "Beerts",
"suffix": ""
},
{
"first": "Timothy",
"middle": [],
"last": "Loncke",
"suffix": ""
},
{
"first": "Sarah",
"middle": [],
"last": "Desmet",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Bernolet",
"suffix": ""
}
],
"year": 2016,
"venue": "Journal of Memory and Language",
"volume": "90",
"issue": "",
"pages": "14--30",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robert J Hartsuiker, Saskia Beerts, Maaike Loncke, Timothy Desmet, and Sarah Bernolet. 2016. Cross- linguistic structural priming in multilinguals: Fur- ther evidence for shared syntax. Journal of Memory and Language, 90:14-30.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Is syntax separate or shared between languages? cross-linguistic syntactic priming in spanish-english bilinguals",
"authors": [
{
"first": "J",
"middle": [],
"last": "Robert",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Hartsuiker",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Martin",
"suffix": ""
},
{
"first": "Eline",
"middle": [],
"last": "Pickering",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Veltkamp",
"suffix": ""
}
],
"year": 2004,
"venue": "Psychological science",
"volume": "15",
"issue": "6",
"pages": "409--414",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robert J Hartsuiker, Martin J Pickering, and Eline Veltkamp. 2004. Is syntax separate or shared be- tween languages? cross-linguistic syntactic prim- ing in spanish-english bilinguals. Psychological sci- ence, 15(6):409-414.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Processing relative clauses in chinese",
"authors": [
{
"first": "Franny",
"middle": [],
"last": "Hsiao",
"suffix": ""
},
{
"first": "Edward",
"middle": [],
"last": "Gibson",
"suffix": ""
}
],
"year": 2003,
"venue": "Cognition",
"volume": "90",
"issue": "1",
"pages": "3--27",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Franny Hsiao and Edward Gibson. 2003. Processing relative clauses in chinese. Cognition, 90(1):3-27.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "The icnale and sophisticated contrastive interlanguage analysis of asian learners of english",
"authors": [
{
"first": "",
"middle": [],
"last": "Shin'ichiro Ishikawa",
"suffix": ""
}
],
"year": 2013,
"venue": "Learner corpus studies in Asia and the world",
"volume": "1",
"issue": "",
"pages": "91--118",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shin'ichiro Ishikawa. 2013. The icnale and sophis- ticated contrastive interlanguage analysis of asian learners of english. Learner corpus studies in Asia and the world, 1:91-118.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Do we need neural models to explain human judgments of acceptability?",
"authors": [
{
"first": "Wang",
"middle": [],
"last": "Jing",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Matthew",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Kelly",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Reitter",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 42nd",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wang Jing, Matthew A Kelly, and David Reitter. 2020. Do we need neural models to explain human judg- ments of acceptability? In Proceedings of the 42nd",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Annual Conference of the Cognitive Science Society",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "1289--1295",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Annual Conference of the Cognitive Science Society, pages 1289-1295.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Jes\u00fas Calvillo, and David Reitter. 2020. Which sentence embeddings and which layers encode syntactic structure?",
"authors": [
{
"first": "Alex",
"middle": [],
"last": "Kelly",
"suffix": ""
},
{
"first": "Yang",
"middle": [],
"last": "Xu",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of the 42nd Annual Conference of the Cognitive Science Society",
"volume": "",
"issue": "",
"pages": "2375--2381",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M Alex Kelly, Yang Xu, Jes\u00fas Calvillo, and David Re- itter. 2020. Which sentence embeddings and which layers encode syntactic structure? In Proceedings of the 42nd Annual Conference of the Cognitive Sci- ence Society, pages 2375-2381.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "How language-neutral is multilingual bert? arXiv preprint",
"authors": [
{
"first": "Jind\u0159ich",
"middle": [],
"last": "Libovick\u1ef3",
"suffix": ""
},
{
"first": "Rudolf",
"middle": [],
"last": "Rosa",
"suffix": ""
},
{
"first": "Alexander",
"middle": [],
"last": "Fraser",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1911.03310"
]
},
"num": null,
"urls": [],
"raw_text": "Jind\u0159ich Libovick\u1ef3, Rudolf Rosa, and Alexander Fraser. 2019. How language-neutral is multilingual bert? arXiv preprint arXiv:1911.03310.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Distinct patterns of syntactic agreement errors in recurrent networks and humans",
"authors": [
{
"first": "Tal",
"middle": [],
"last": "Linzen",
"suffix": ""
},
{
"first": "Brian",
"middle": [],
"last": "Leonard",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1807.06882"
]
},
"num": null,
"urls": [],
"raw_text": "Tal Linzen and Brian Leonard. 2018. Distinct patterns of syntactic agreement errors in recurrent networks and humans. arXiv preprint arXiv:1807.06882.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Automatic analysis of syntactic complexity in second language writing",
"authors": [
{
"first": "Xiaofei",
"middle": [],
"last": "Lu",
"suffix": ""
}
],
"year": 2010,
"venue": "International journal of corpus linguistics",
"volume": "15",
"issue": "4",
"pages": "474--496",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiaofei Lu. 2010. Automatic analysis of syntactic com- plexity in second language writing. International journal of corpus linguistics, 15(4):474-496.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "A report on the 2017 native language identification shared task",
"authors": [
{
"first": "Shervin",
"middle": [],
"last": "Malmasi",
"suffix": ""
},
{
"first": "Keelan",
"middle": [],
"last": "Evanini",
"suffix": ""
},
{
"first": "Aoife",
"middle": [],
"last": "Cahill",
"suffix": ""
},
{
"first": "Joel",
"middle": [],
"last": "Tetreault",
"suffix": ""
},
{
"first": "Robert",
"middle": [],
"last": "Pugh",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Hamill",
"suffix": ""
},
{
"first": "Diane",
"middle": [],
"last": "Napolitano",
"suffix": ""
},
{
"first": "Yao",
"middle": [],
"last": "Qian",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications",
"volume": "",
"issue": "",
"pages": "62--75",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shervin Malmasi, Keelan Evanini, Aoife Cahill, Joel Tetreault, Robert Pugh, Christopher Hamill, Diane Napolitano, and Yao Qian. 2017. A report on the 2017 native language identification shared task. In Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications, pages 62-75.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Targeted syntactic evaluation of language models",
"authors": [
{
"first": "Rebecca",
"middle": [],
"last": "Marvin",
"suffix": ""
},
{
"first": "Tal",
"middle": [],
"last": "Linzen",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1808.09031"
]
},
"num": null,
"urls": [],
"raw_text": "Rebecca Marvin and Tal Linzen. 2018. Targeted syn- tactic evaluation of language models. arXiv preprint arXiv:1808.09031.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Cross-linguistic syntactic evaluation of word prediction models",
"authors": [
{
"first": "Aaron",
"middle": [],
"last": "Mueller",
"suffix": ""
},
{
"first": "Garrett",
"middle": [],
"last": "Nicolai",
"suffix": ""
},
{
"first": "Panayiota",
"middle": [],
"last": "Petrou-Zeniou",
"suffix": ""
},
{
"first": "Natalia",
"middle": [],
"last": "Talmina",
"suffix": ""
},
{
"first": "Tal",
"middle": [],
"last": "Linzen",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2005.00187"
]
},
"num": null,
"urls": [],
"raw_text": "Aaron Mueller, Garrett Nicolai, Panayiota Petrou- Zeniou, Natalia Talmina, and Tal Linzen. 2020. Cross-linguistic syntactic evaluation of word predic- tion models. arXiv preprint arXiv:2005.00187.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Integrated, not isolated: Defining typological proximity in an integrated multilingual architecture",
"authors": [
{
"first": "T",
"middle": [],
"last": "Michael",
"suffix": ""
},
{
"first": "Matthew",
"middle": [],
"last": "Putnam",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Carlson",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Reitter",
"suffix": ""
}
],
"year": 2018,
"venue": "Frontiers in psychology",
"volume": "8",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Michael T Putnam, Matthew Carlson, and David Reit- ter. 2018. Integrated, not isolated: Defining typolog- ical proximity in an integrated multilingual architec- ture. Frontiers in psychology, 8:2212.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "A study of n-gram and embedding representations for native language identification",
"authors": [
{
"first": "Sowmya",
"middle": [],
"last": "Vajjala",
"suffix": ""
},
{
"first": "Sagnik",
"middle": [],
"last": "Banerjee",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 12th Workshop on Innovative Use of NLP for Building Educational Applications",
"volume": "",
"issue": "",
"pages": "240--248",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sowmya Vajjala and Sagnik Banerjee. 2017. A study of n-gram and embedding representations for native language identification. In Proceedings of the 12th Workshop on Innovative Use of NLP for Building Ed- ucational Applications, pages 240-248.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "vestigating bert's knowledge of language: Five analysis methods with npis",
"authors": [
{
"first": "Alex",
"middle": [],
"last": "Warstadt",
"suffix": ""
},
{
"first": "Yu",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "Ioana",
"middle": [],
"last": "Grosu",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Peng",
"suffix": ""
},
{
"first": "Hagen",
"middle": [],
"last": "Blix",
"suffix": ""
},
{
"first": "Yining",
"middle": [],
"last": "Nie",
"suffix": ""
},
{
"first": "Anna",
"middle": [],
"last": "Alsop",
"suffix": ""
},
{
"first": "Shikha",
"middle": [],
"last": "Bordia",
"suffix": ""
},
{
"first": "Haokun",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Alicia",
"middle": [],
"last": "Parrish",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1909.02597"
]
},
"num": null,
"urls": [],
"raw_text": "Alex Warstadt, Yu Cao, Ioana Grosu, Wei Peng, Ha- gen Blix, Yining Nie, Anna Alsop, Shikha Bor- dia, Haokun Liu, Alicia Parrish, et al. 2019. In- vestigating bert's knowledge of language: Five analysis methods with npis. arXiv preprint arXiv:1909.02597.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Are all languages created equal in multilingual bert?",
"authors": [
{
"first": "Shijie",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Dredze",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 5th Workshop on Representation Learning for NLP",
"volume": "",
"issue": "",
"pages": "120--130",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shijie Wu and Mark Dredze. 2020. Are all languages created equal in multilingual bert? In Proceedings of the 5th Workshop on Representation Learning for NLP, pages 120-130.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"html": null,
"content": "<table/>",
"text": "Corpus Description. Bold numbers are the size of testing sets.",
"type_str": "table",
"num": null
},
"TABREF4": {
"html": null,
"content": "<table><tr><td colspan=\"2\">: Models weighted F-scores for Native Lan-</td></tr><tr><td colspan=\"2\">guage Identification: ICNALE</td></tr><tr><td/><td>Full Mid</td></tr><tr><td>BERT-base</td><td>0.72 0.64</td></tr><tr><td colspan=\"2\">mBERT-base 0.68 0.62</td></tr><tr><td>Doc2Vec</td><td>0.71 -</td></tr><tr><td colspan=\"2\">Stacked SVM 0.88 -</td></tr><tr><td>Random</td><td>0.09 0.09</td></tr><tr><td>Baseline</td><td>0.71 -</td></tr></table>",
"text": "",
"type_str": "table",
"num": null
},
"TABREF5": {
"html": null,
"content": "<table/>",
"text": "",
"type_str": "table",
"num": null
}
}
}
} |