File size: 77,388 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:24:36.350421Z"
},
"title": "Looking for a Role for Word Embeddings in Eye-Tracking Features Prediction: Does Semantic Similarity Help?",
"authors": [
{
"first": "Lavinia",
"middle": [],
"last": "Salicchi",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "The Hong Kong Polytechnic University",
"location": {}
},
"email": "lavinia.salicchi@connect.polyu.hk"
},
{
"first": "Alessandro",
"middle": [],
"last": "Lenci",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Pisa",
"location": {}
},
"email": "alessandro.lenci@unipi.it"
},
{
"first": "Emmanuele",
"middle": [],
"last": "Chersoni",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "The Hong Kong Polytechnic University",
"location": {}
},
"email": "emmanuelechersoni@gmail.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Eye-tracking psycholinguistic studies have suggested that context-word semantic coherence and predictability influence language processing during the reading activity. In this study, we investigated the correlation between the cosine similarities computed with word embedding models (both static and contextualized) and eye-tracking data from two naturalistic reading corpora. We also studied the correlations of surprisal scores computed with three state-of-the-art language models. Our results show strong correlation for the scores computed with BERT and GloVe, suggesting that similarity can play an important role in modeling reading times.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "Eye-tracking psycholinguistic studies have suggested that context-word semantic coherence and predictability influence language processing during the reading activity. In this study, we investigated the correlation between the cosine similarities computed with word embedding models (both static and contextualized) and eye-tracking data from two naturalistic reading corpora. We also studied the correlations of surprisal scores computed with three state-of-the-art language models. Our results show strong correlation for the scores computed with BERT and GloVe, suggesting that similarity can play an important role in modeling reading times.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Eye-tracking data recorded during reading provide invaluable evidence about the factors influencing language comprehension. Research in computational modeling has particularly focused on two factors: i.) the semantic coherence of a word with the rest of the sentence (Ehrlich and Rayner, 1981; Pynte et al., 2008; Mitchell et al., 2010) , measured via semantic similarity metrics and ii.) its predictability from previous context, as measured by surprisal (Hale, 2001; Levy, 2008) . Intuitively, words that have low semantic coherence and low in-context predictability (i.e., high surprisal) induce longer reading times.",
"cite_spans": [
{
"start": 267,
"end": 293,
"text": "(Ehrlich and Rayner, 1981;",
"ref_id": "BIBREF8"
},
{
"start": 294,
"end": 313,
"text": "Pynte et al., 2008;",
"ref_id": "BIBREF25"
},
{
"start": 314,
"end": 336,
"text": "Mitchell et al., 2010)",
"ref_id": "BIBREF22"
},
{
"start": 456,
"end": 468,
"text": "(Hale, 2001;",
"ref_id": "BIBREF11"
},
{
"start": 469,
"end": 480,
"text": "Levy, 2008)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In distributional semantics (Lenci, 2018) , words and their sentence contexts are represented with dense vectors called embeddings and produced by Distributional Semantic Models (DSM). In this paper, we modeled semantic coherence with the cosine similarity between the embeddings of words and their sentence contexts, and then we tested the correlation of the metric with the eye-tracking measures annotated on the GECO and Provo corpora. We analyzed the correlations for the similarity computed with 10 different embedding models (both static and contextualized), as well as for surprisal scores computed with several state-of-the-art neural language models. Among all the features under investigation, the similarity scores obtained with BERT and GloVe obtained the best correlations across features in both the benchmark corpora.",
"cite_spans": [
{
"start": 28,
"end": 41,
"text": "(Lenci, 2018)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Hollenstein et al. 2019proposed a framework to evaluate six state-of-the-art word embedding models (GloVe, Word2Vec, WordNet2Vec, Fast-Text, ELMo, BERT). The evaluation was based on the model capability to reflect semantic representations in the human mind, using cognitive data in different datasets for eye-tracking, EEG, and fMRI. Word embedding models were used to train neural networks on a regression task. While we aim at creating a computational model of the relationship between context processing and the integration of a new word during naturalistic reading, Hollenstein et al. (2019) evaluated embedding models on the prediction of out-of-context word features. The results of their analyses showed that BERT, ELMo, and FastText have the best prediction performances. On the other hand, approaches based on powerful Transformers language models were outperformed by a classifier using linguistic and psychometric features (Bestgen, 2021) in the recent CMCL 2021 Shared Task on Eye-Tracking Data Prediction (Hollenstein et al., 2021) .",
"cite_spans": [
{
"start": 570,
"end": 595,
"text": "Hollenstein et al. (2019)",
"ref_id": "BIBREF14"
},
{
"start": 934,
"end": 949,
"text": "(Bestgen, 2021)",
"ref_id": "BIBREF2"
},
{
"start": 1018,
"end": 1044,
"text": "(Hollenstein et al., 2021)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "A series of contributions explored the role of surprisal in modeling reading times in naturalistic settings, coming to the general conclusion that the predictive power is strongly related to the language model quality, i.e. models with better perplexity perform better (Smith and Levy, 2013; Goodkind and Bicknell, 2018) . Later work explored the most recent neural models, including LSTM (van Schijndel and Linzen, 2018), GRU (Aurnhammer and Frank, 2019) , Transformers (Merkx and Frank, 2020) and GPT-2 (Wilcox et al., 2020) , basically confirming this relationship. 1 Early studies had also found correlations between semantic distance, computed by word embeddings, and eye-tracking features in reading processes (Pynte et al., 2008; Mitchell et al., 2010) . However, the more recent work by Frank (2017) pointed out that, since word embeddings are based on co-occurrences, semantic distance may actually represent word predictability, rather than semantic relatedness, and that those early findings were actually due to a confound between these two concepts. To test this hypothesis, the author used linear regression models with and without surprisal, testing 5 surprisal measures. The results show that the effects of similarity on reading times disappear when surprisal is factored out, thereby proving the existence of a complex interplay between the two factors. Frank's experiments were carried out in a naturalistic reading setting and, to our knowledge, there have been no eye-tracking studies with controlled stimuli investigating a possible separate effect of the two components (for example, by comparing the fixation patterns of words that have low predictability, but different degrees of coherence with the sentence or with the discourse context).",
"cite_spans": [
{
"start": 280,
"end": 291,
"text": "Levy, 2013;",
"ref_id": "BIBREF28"
},
{
"start": 292,
"end": 320,
"text": "Goodkind and Bicknell, 2018)",
"ref_id": "BIBREF10"
},
{
"start": 427,
"end": 455,
"text": "(Aurnhammer and Frank, 2019)",
"ref_id": "BIBREF0"
},
{
"start": 471,
"end": 494,
"text": "(Merkx and Frank, 2020)",
"ref_id": "BIBREF19"
},
{
"start": 499,
"end": 526,
"text": "GPT-2 (Wilcox et al., 2020)",
"ref_id": null
},
{
"start": 569,
"end": 570,
"text": "1",
"ref_id": null
},
{
"start": 716,
"end": 736,
"text": "(Pynte et al., 2008;",
"ref_id": "BIBREF25"
},
{
"start": 737,
"end": 759,
"text": "Mitchell et al., 2010)",
"ref_id": "BIBREF22"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "3 Experimental Setting",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Traditional corpora annotated with eye-tracking data consist of short isolated sentences (or even single words) with particular structures or lexemes, in order to investigate specific syntactic and semantic phenomena. In the present work, we used GECO (Cop et al., 2017) and Provo (Luke and Christianson, 2018), two eye-tracking corpora containing long, complete, and coherent texts. GECO is a monolingual and bilingual (English and Dutch) corpus composed of the entire Agatha Christie's novel The Mysterious Affair at Styles. The corpus is freely downloadable with a related dataset containing eye-tracking data of 33 subjects (19 of them bilingual, 14 English monolingual) reading the full novel text, presented paragraph-by-paragraph on a screen. GECO is composed of 54, 364 tokens. Provo contains 55 short English texts about various topics, with 2.5 sentences and 50 words on average, for a total of 2, 689 tokens, and a vocabu-lary of 1, 197 words. These texts were read by 85 subjects and their eye-tracking measures were collected in an available on-line dataset. GECO and Provo data are particularly interesting because they are recorded during naturalistic reading, instead of short selected stimuli.",
"cite_spans": [
{
"start": 252,
"end": 270,
"text": "(Cop et al., 2017)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Datasets",
"sec_num": "3.1"
},
{
"text": "For every word in the corpora, we extracted its mean total reading time, mean first fixation duration, and mean number of fixations, by averaging over the subjects. The choice of modeling mean eye-tracking measures is justified by the high intersubject consistency of the recorded data. For instance, Cop et al. (2017) report an overall intersubject correlation of 0.9 for the total reading times in GECO. Table 1 shows the embeddings types used in our experiments, consisting of 6 non-contextualized, static DSMs and 4 contextualized DSMs. The former include predict models (SGNS and FastText) (Mikolov et al., 2013; Levy and Goldberg, 2014; Bojanowski et al., 2017) and count models (SVD and GloVe) (Bullinaria and Levy, 2012; Pennington et al., 2014). 2 Four DSMs are window-based and two are syntax-based (synt). Embeddings have 300 dimensions and were trained on a corpus of 3.9 billion tokens ca. (a concatenation of ukWaC and a 2018 dump of Wikipedia). Pre-trained contextualized embeddings include the 512-dimensional vectors produced by the three layers of the ELMo bidirectional LSTM architecture (Peters et al., 2018) , the 1, 024-dimensional vectors produced by the 24-layers BERT-Large Transformer architecture (BERT-Large, Cased) (Devlin et al., 2019) , the 1, 600-dimensional vectors by GPT2-xl (Radford et al.) , and finally, the 200-dimensional vectors produced by the Neural Complexity model by van Schijndel and Linzen (2018).",
"cite_spans": [
{
"start": 301,
"end": 318,
"text": "Cop et al. (2017)",
"ref_id": "BIBREF6"
},
{
"start": 595,
"end": 617,
"text": "(Mikolov et al., 2013;",
"ref_id": "BIBREF21"
},
{
"start": 618,
"end": 642,
"text": "Levy and Goldberg, 2014;",
"ref_id": "BIBREF16"
},
{
"start": 643,
"end": 667,
"text": "Bojanowski et al., 2017)",
"ref_id": "BIBREF3"
},
{
"start": 1107,
"end": 1128,
"text": "(Peters et al., 2018)",
"ref_id": "BIBREF24"
},
{
"start": 1244,
"end": 1265,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF7"
},
{
"start": 1310,
"end": 1326,
"text": "(Radford et al.)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 406,
"end": 413,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Datasets",
"sec_num": "3.1"
},
{
"text": "Our main goals were to investigate the potential contribution of cosine similarity in predicting eyetracking features, to compare different word embedding models, and then to evaluate whether the information represented by cosine similarity is similar to the one represented by surprisal.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "For each target word w in GECO and Provo, we measured the cosine similarity between the embedding of w and the embedding of the context c formed by the previous words in the same sentence. We then computed the Spearman correlation between the cosine and the eye-tracking data for w (total reading time, first fixation duration, and number of fixations). To create context embedding, we used an additive model: the context vector is the sum of all its word embeddings. Given the bidirectional nature of BERT, the input to this model needed a special pre-processing: To prevent that the vectors representing words within the context were computed using the target word itself, we passed to BERT a list of sub-sentences, each of which were composed of context words only. So given the sentence The dog chases the cat: Starting from the second sub-sentence, the cosine similarity was computed between the last word vector and the sum of words vectors belonging to the previous sub-sentence (list element). So, to compute the cosine similarity between cat and the previous context, we selected cat from S[4] and T he + dog + chases + the from S[3].",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "S[0] = [\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "For BERT we used as context also the embedding produced by the model for the special token CLS, which is created using a weighted additive model. As for the simple additive model, BERT was fed with sub-sentences, and for each target word the CLS-context-vector was the one computed at the previous list element. In the previous example, given cat as target word, we used the CLS vector representing all the S[3] elements.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "Given the positive effect of semantic coherence on language processing, we expected that the eyetracking data for w had a negative correlation with its cosine similarity with c: The higher the cosine, the lower the reading time of w measured by eye-tracking.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "We used BERT, GPT2-xl and Neural Complexity to compute word-by-word surprisal. Like with cosine similarity, the input sentences for BERT were organized in sub-sentences, and the last token (i.e., the target word), was replaced with the special tag [MASK] . Finally, we computed the Spearman correlation between the surprisal of w, and the eye-tracking data for the target word. Differently from the cosine, we expected the surprisal to be positively correlated with the word reading time: The less predictable a word is, the slower its processing will be.",
"cite_spans": [
{
"start": 248,
"end": 254,
"text": "[MASK]",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "The analyses have been performed with the following models: 6 values of cosine similarity between non-contextualized vectors, 51 values of cosine similarity between contextualized vectors (48 from 24 layers of BERT in two different ways to compute the context vector, and 3 from ELMo, GPT2-xl and Neural Complexity), 3 values of surprisal from BERT, GPT2-xl, Neural Complexity.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3.3"
},
{
"text": "Looking at the correlations results, it is clear that every model performed better on Provo. One possible explanation for this difference is that GECO eye-tracking data are recorded on participants reading a literary text, while Provo materials are online news articles, science magazines and only partially short text from works of fiction. The consequence is a difference in the syntactic complexity of sentence structure and in the frequency of words. This gap implies that the modeling of GECO contexts is less directly reducible to an additive fashion of processing, and, most importantly, is more likely to find Out Of Vocabulary words in GECO, rather than in Provo. Another aspect that is quite evident are the similar correlation values among different eye-tracking features. This aspect is not surprising: in the original datasets of GECO and Provo, it can be noticed that many words show the same value for the total reading time and the first fixation duration. This happens when i) the word is not read (0 ms for both the features); ii) the word is read only once (total reading time and first fixation duration overlap). Also regarding the similar values of the correlations between similarity and number of fixations and between similarity and total reading times, taking into account the original data gives us an explanation of the results: since the total reading time is computed summing the duration of all the multiple fixations, the higher the number of fixation, the higher the total reading time, leading to a similar tendency in the values of the two features. For these reasons, the total reading time may be considered as a \"bridge\" field, that holds close relations with both first fixation duration and number of fixations, justifying the similar correlation values in our results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Discussion",
"sec_num": "4"
},
{
"text": "Comparing word embedding models, we may notice that correlations can reach very high values, up to \u22120.71 for the total reading time (by BERT CLS layer 22), suggesting that semantic coherence -modeled as cosine similarity between context and target-can be a strong predictor of eye-tracking measures of reading process. GloVe (mean correlation over eye-tracking features on GECO: \u22120.45, on Provo: \u22120.65) and BERT (mean correlation over eye-tracking features on GECO: \u22120.57, on Provo: \u22120.71) score the best results on both corpora, and in the latter case the [CLS] context model brings some advantage over the simple additive one. The lower BERT layers show a steadily decreasing performance (see Figure 1 ). This was expected because, as it was pointed out in the layers analysis by Tenney et al. (2019) , the BERT architecture reproduces the classical NLP pipeline: the lower layers process mainly the syntactic information, while the highest ones give a more precise representation of semantic relations. We also notice a strong variability among the embedding models, which is orthogonal to the contextualized vs. non-contextualized dichotomy. The ELMo contex-tualized vectors perform much worse than BERT ones, probably because they have a lower degree of contextualization, and syntax-based count models are not significantly worse than predict DSMs. Regarding the correlations between the target word surprisal computed with BERT, GPT2-xl and Neural Complexity (NC) and the eye-tracking measures (see Table 3 ), the first striking fact is that the absolute values are generally lower than the scores obtained with the cosine (higher correlations are reached by GPT2-xl con GECO, mean correlation = 0.40, and by NC on Provo, mean correlation = 0.47). This might prompt us to conclude that surprisal is a much weaker predictor than semantic coherence. However, a significant negative correlation between cosine similarity and surprisal (e.g. with BERT it is \u22120.40 on GECO and \u22120.32 on Provo) supports the hypothesis by Frank (2017) that there is a strong overlap between semantic coherence and surprisal. Factoring out the contribution of these two factors on eye-tracking features will be the next step of our research work.",
"cite_spans": [
{
"start": 782,
"end": 802,
"text": "Tenney et al. (2019)",
"ref_id": "BIBREF29"
}
],
"ref_spans": [
{
"start": 695,
"end": 703,
"text": "Figure 1",
"ref_id": "FIGREF1"
},
{
"start": 1506,
"end": 1513,
"text": "Table 3",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Results and Discussion",
"sec_num": "4"
},
{
"text": "In this paper, we have used contextualized and non-contextualized DSMs to compute the cosine between a target word and the previous sentence context. Our results show that cosine similarity is able to achieve very high correlations with the eyetracking metrics of GECO and Provo, especially with the BERT and GloVe models, providing further evidence that semantic coherence is potentially very useful in modeling reading times. Furthermore, we computed word-by-word surprisal using BERT, GPT2-xl, and Neural Complexity.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and ongoing work",
"sec_num": "5"
},
{
"text": "Among the language models, the best results have been achieved by GPT2-xl, confirming the previous findings that Transformers are very good at modeling sentence processing metrics (Wilcox et al., 2020; Hao et al., 2020; Merkx and Frank, 2021) . However, the absolute value of correlation is lower than the one obtained with cosine similarity scores: for example, the mean correlation achieved on Provo with the cosine similarity between vectors produced by BERT is \u22120.71, while the correlation between eye tracking features and the surprisal computed by the same model is 0.24. The comparison between correlations reached by cosine similarity and surprisal may lead us to the conclusion that semantic coherence is a stronger predictor of eye-tracking features than word predictability. However, given the significant degree of correlation between cosine similarity and surprisal, further investigations are needed to disentangle the two factors.",
"cite_spans": [
{
"start": 180,
"end": 201,
"text": "(Wilcox et al., 2020;",
"ref_id": "BIBREF30"
},
{
"start": 202,
"end": 219,
"text": "Hao et al., 2020;",
"ref_id": "BIBREF12"
},
{
"start": 220,
"end": 242,
"text": "Merkx and Frank, 2021)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and ongoing work",
"sec_num": "5"
},
{
"text": "Our next step will be to include Transformersbased surprisal and vector-based cosine similarity in a large-scale regression study to predict eye tracking features, in order to ensure a close comparison with the experimental setting of Frank (2017), and to investigate if semantic similarity models can actually play a distinct role from surprisal in the prediction of reading times. Differently from Frank (2017), we plan to test with several regression models, from a simple linear regression to more advanced regression models (e.g. Gradient Boosting, Multilayer Perceptron etc.), and with different word embedding models, in order to account for the different types of semantic similarity computed by static and contextualized embeddings.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and ongoing work",
"sec_num": "5"
},
{
"text": "Notice however that doubts have been raised on the reliability of perplexity as a metric for comparing large pretrained models, since it does not allow to compare models with different vocabularies(Hao et al., 2020).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "For the distinction between count and predict DSM, we refer toBaroni et al. (2014).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Evaluating Information-theoretic Measures of Word Prediction in Naturalistic Sentence Reading. Neuropsychologia",
"authors": [
{
"first": "Christoph",
"middle": [],
"last": "Aurnhammer",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Stefan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "134",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christoph Aurnhammer and Stefan L Frank. 2019. Evaluating Information-theoretic Measures of Word Prediction in Naturalistic Sentence Reading. Neu- ropsychologia, 134.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Don't Count, Predict! A Systematic Comparison of Context-counting vs. Context-predicting Semantic Vectors",
"authors": [
{
"first": "Marco",
"middle": [],
"last": "Baroni",
"suffix": ""
},
{
"first": "Georgiana",
"middle": [],
"last": "Dinu",
"suffix": ""
},
{
"first": "Germ\u00e1n",
"middle": [],
"last": "Kruszewski",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marco Baroni, Georgiana Dinu, and Germ\u00e1n Kruszewski. 2014. Don't Count, Predict! A Systematic Comparison of Context-counting vs. Context-predicting Semantic Vectors. In Proceed- ings of ACL.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "LAST at CMCL 2021 Shared Task: Predicting Gaze Data During Reading with a Gradient Boosting Decision Tree Approach",
"authors": [
{
"first": "Yves",
"middle": [],
"last": "Bestgen",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the NAACL Workshop on Cognitive Modeling and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yves Bestgen. 2021. LAST at CMCL 2021 Shared Task: Predicting Gaze Data During Reading with a Gradient Boosting Decision Tree Approach. In Pro- ceedings of the NAACL Workshop on Cognitive Mod- eling and Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Enriching Word Vectors with",
"authors": [
{
"first": "Piotr",
"middle": [],
"last": "Bojanowski",
"suffix": ""
},
{
"first": "Edouard",
"middle": [],
"last": "Grave",
"suffix": ""
},
{
"first": "Armand",
"middle": [],
"last": "Joulin",
"suffix": ""
},
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. Enriching Word Vectors with",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Subword Information. Transactions of the Association for Computational Linguistics",
"authors": [],
"year": null,
"venue": "",
"volume": "5",
"issue": "",
"pages": "135--146",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Subword Information. Transactions of the Associa- tion for Computational Linguistics, 5:135-146.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Extracting Semantic Representations from Word Co-Occurrence Statistics",
"authors": [
{
"first": "A",
"middle": [],
"last": "John",
"suffix": ""
},
{
"first": "Joseph P",
"middle": [],
"last": "Bullinaria",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Levy",
"suffix": ""
}
],
"year": 2012,
"venue": "Stop-Lists, Stemming, and SVD. Behavior Research Methods",
"volume": "44",
"issue": "",
"pages": "890--907",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John A Bullinaria and Joseph P Levy. 2012. Ex- tracting Semantic Representations from Word Co- Occurrence Statistics: Stop-Lists, Stemming, and SVD. Behavior Research Methods, 44(3):890-907.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Presenting GECO: An Eye-Tracking Corpus of Monolingual and Bilingual Sentence Reading",
"authors": [
{
"first": "Uschi",
"middle": [],
"last": "Cop",
"suffix": ""
},
{
"first": "Nicolas",
"middle": [],
"last": "Dirix",
"suffix": ""
},
{
"first": "Denis",
"middle": [],
"last": "Drieghe",
"suffix": ""
},
{
"first": "Wouter",
"middle": [],
"last": "Duyck",
"suffix": ""
}
],
"year": 2017,
"venue": "Behavior Reseach Methods",
"volume": "49",
"issue": "2",
"pages": "602--615",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Uschi Cop, Nicolas Dirix, Denis Drieghe, and Wouter Duyck. 2017. Presenting GECO: An Eye-Tracking Corpus of Monolingual and Bilingual Sentence Reading. Behavior Reseach Methods, 49(2):602- 615.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"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": 2019,
"venue": "Proceedings of NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Un- derstanding. In Proceedings of NAACL.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Contextual Effects on Word Perception and Eye Movements During Reading",
"authors": [
{
"first": "Susan",
"middle": [
"E"
],
"last": "Ehrlich",
"suffix": ""
},
{
"first": "Keith",
"middle": [],
"last": "Rayner",
"suffix": ""
}
],
"year": 1981,
"venue": "Journal of Verbal Learning and Verbal Behavior",
"volume": "20",
"issue": "",
"pages": "641--65",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Susan E. Ehrlich and Keith Rayner. 1981. Contex- tual Effects on Word Perception and Eye Movements During Reading. Journal of Verbal Learning and Verbal Behavior, 20:641-65.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Word Embedding Distance Does not Predict Word Reading Time",
"authors": [
{
"first": "L",
"middle": [],
"last": "Stefan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of CogSci",
"volume": "",
"issue": "",
"pages": "385--390",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stefan L Frank. 2017. Word Embedding Distance Does not Predict Word Reading Time. In Proceedings of CogSci, pages 385-390.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Predictive Power of Word Surprisal for Reading Times is a Linear Function of Language Model Quality",
"authors": [
{
"first": "Adam",
"middle": [],
"last": "Goodkind",
"suffix": ""
},
{
"first": "Klinton",
"middle": [],
"last": "Bicknell",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the LSA Workshop on Cognitive Modeling and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Adam Goodkind and Klinton Bicknell. 2018. Predic- tive Power of Word Surprisal for Reading Times is a Linear Function of Language Model Quality. In Pro- ceedings of the LSA Workshop on Cognitive Model- ing and Computational Linguistics.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "A Probabilistic Earley Parser as a Psycholinguistic Model",
"authors": [
{
"first": "John",
"middle": [],
"last": "Hale",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John Hale. 2001. A Probabilistic Earley Parser as a Psycholinguistic Model. In Proceedings of NAACL.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Probabilistic Predictions of People Perusing: Evaluating Metrics of Language Model Performance for Psycholinguistic Modeling",
"authors": [
{
"first": "Yiding",
"middle": [],
"last": "Hao",
"suffix": ""
},
{
"first": "Simon",
"middle": [],
"last": "Mendelsohn",
"suffix": ""
},
{
"first": "Rachel",
"middle": [],
"last": "Sterneck",
"suffix": ""
},
{
"first": "Randi",
"middle": [],
"last": "Martinez",
"suffix": ""
},
{
"first": "Robert",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the EMNLP Workshop on Cognitive Modeling and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yiding Hao, Simon Mendelsohn, Rachel Sterneck, Randi Martinez, and Robert Frank. 2020. Prob- abilistic Predictions of People Perusing: Evaluat- ing Metrics of Language Model Performance for Psycholinguistic Modeling. In Proceedings of the EMNLP Workshop on Cognitive Modeling and Com- putational Linguistics.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "CMCL 2021 Shared Task on Eye-Tracking Prediction",
"authors": [
{
"first": "Nora",
"middle": [],
"last": "Hollenstein",
"suffix": ""
},
{
"first": "Emmanuele",
"middle": [],
"last": "Chersoni",
"suffix": ""
},
{
"first": "Cassandra",
"middle": [
"L"
],
"last": "Jacobs",
"suffix": ""
},
{
"first": "Yohei",
"middle": [],
"last": "Oseki",
"suffix": ""
},
{
"first": "Laurent",
"middle": [],
"last": "Pr\u00e9vot",
"suffix": ""
},
{
"first": "Enrico",
"middle": [],
"last": "Santus",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the NAACL Workshop on Cognitive Modeling and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nora Hollenstein, Emmanuele Chersoni, Cassandra L Jacobs, Yohei Oseki, Laurent Pr\u00e9vot, and Enrico Santus. 2021. CMCL 2021 Shared Task on Eye- Tracking Prediction. In Proceedings of the NAACL Workshop on Cognitive Modeling and Computa- tional Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "CogniVal: A Framework for Cognitive Word Embedding Evaluation",
"authors": [
{
"first": "Nora",
"middle": [],
"last": "Hollenstein",
"suffix": ""
},
{
"first": "Antonio",
"middle": [],
"last": "De La Torre",
"suffix": ""
},
{
"first": "Nicolas",
"middle": [],
"last": "Langer",
"suffix": ""
},
{
"first": "Ce",
"middle": [],
"last": "Zhang",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of CONLL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nora Hollenstein, Antonio de la Torre, Nicolas Langer, and Ce Zhang. 2019. CogniVal: A Framework for Cognitive Word Embedding Evaluation. In Proceed- ings of CONLL.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Distributional Models of Word Meaning",
"authors": [
{
"first": "Alessandro",
"middle": [],
"last": "Lenci",
"suffix": ""
}
],
"year": 2018,
"venue": "Annual Review of Linguistics",
"volume": "4",
"issue": "",
"pages": "151--171",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alessandro Lenci. 2018. Distributional Models of Word Meaning. Annual Review of Linguistics, 4:151-171.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Dependency-Based Word Embeddings",
"authors": [
{
"first": "Omer",
"middle": [],
"last": "Levy",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Goldberg",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Omer Levy and Yoav Goldberg. 2014. Dependency- Based Word Embeddings. In Proceedings of ACL.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Expectation-based Syntactic Comprehension",
"authors": [
{
"first": "Roger",
"middle": [],
"last": "Levy",
"suffix": ""
}
],
"year": 2008,
"venue": "Cognition",
"volume": "106",
"issue": "3",
"pages": "1126--1177",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Roger Levy. 2008. Expectation-based Syntactic Com- prehension. Cognition, 106(3):1126-1177.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "The Provo Corpus: A Large Eye-tracking Corpus with Predictability Norms",
"authors": [
{
"first": "G",
"middle": [],
"last": "Steven",
"suffix": ""
},
{
"first": "Kiel",
"middle": [],
"last": "Luke",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Christianson",
"suffix": ""
}
],
"year": 2018,
"venue": "Behavior Research Methods",
"volume": "50",
"issue": "2",
"pages": "826--833",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Steven G Luke and Kiel Christianson. 2018. The Provo Corpus: A Large Eye-tracking Corpus with Predictability Norms. Behavior Research Methods, 50(2):826-833.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Comparing Transformers and RNNs on Predicting Human Sentence Processing Data",
"authors": [
{
"first": "Danny",
"middle": [],
"last": "Merkx",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Stefan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2005.09471"
]
},
"num": null,
"urls": [],
"raw_text": "Danny Merkx and Stefan L Frank. 2020. Com- paring Transformers and RNNs on Predicting Hu- man Sentence Processing Data. arXiv preprint arXiv:2005.09471.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Human Sentence Processing: Recurrence or Attention?",
"authors": [
{
"first": "Danny",
"middle": [],
"last": "Merkx",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Stefan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the NAACL Workshop on Cognitive Modeling and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Danny Merkx and Stefan L Frank. 2021. Human Sen- tence Processing: Recurrence or Attention? In Pro- ceedings of the NAACL Workshop on Cognitive Mod- eling and Computational Linguistics.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Efficient Estimation of Word Representations in Vector Space",
"authors": [
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "Kai",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Greg",
"middle": [],
"last": "Corrado",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2013,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1301.3781"
]
},
"num": null,
"urls": [],
"raw_text": "Tomas Mikolov, Kai Chen, Greg Corrado, and Jef- frey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arXiv preprint arXiv:1301.3781.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Syntactic and Semantic Factors in Processing Difficulty: An Integrated Measure",
"authors": [
{
"first": "Jeff",
"middle": [],
"last": "Mitchell",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
},
{
"first": "Vera",
"middle": [],
"last": "Demberg",
"suffix": ""
},
{
"first": "Frank",
"middle": [],
"last": "Keller",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jeff Mitchell, Mirella Lapata, Vera Demberg, and Frank Keller. 2010. Syntactic and Semantic Factors in Processing Difficulty: An Integrated Measure. In Proceedings of ACL.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Glove: Global Vectors for Word Representation",
"authors": [
{
"first": "Jeffrey",
"middle": [],
"last": "Pennington",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Socher",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. Glove: Global Vectors for Word Representation. In Proceedings of EMNLP.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Deep Contextualized Word Representations",
"authors": [
{
"first": "Matthew",
"middle": [
"E"
],
"last": "Peters",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Neumann",
"suffix": ""
},
{
"first": "Mohit",
"middle": [],
"last": "Iyyer",
"suffix": ""
},
{
"first": "Matt",
"middle": [],
"last": "Gardner",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Clark",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep Contextualized Word Rep- resentations. In Proceedings of NAACL.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Online Contextual Influences During Reading Normal Text: A Multiple-Regression Analysis. Vision research",
"authors": [
{
"first": "Joel",
"middle": [],
"last": "Pynte",
"suffix": ""
},
{
"first": "Boris",
"middle": [],
"last": "New",
"suffix": ""
},
{
"first": "Alan",
"middle": [],
"last": "Kennedy",
"suffix": ""
}
],
"year": 2008,
"venue": "",
"volume": "48",
"issue": "",
"pages": "2172--2183",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joel Pynte, Boris New, and Alan Kennedy. 2008. On- line Contextual Influences During Reading Normal Text: A Multiple-Regression Analysis. Vision re- search, 48(21):2172-2183.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Language Models are Unsupervised Multitask Learners",
"authors": [
{
"first": "A",
"middle": [],
"last": "Radford",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Child",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Luan",
"suffix": ""
},
{
"first": "Dario",
"middle": [],
"last": "Amodei",
"suffix": ""
},
{
"first": "Ilya",
"middle": [],
"last": "Sutskever",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Radford, Jeffrey Wu, R. Child, David Luan, Dario Amodei, and Ilya Sutskever. Language Models are Unsupervised Multitask Learners. In Open-AI Blog.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "A Neural Model of Adaptation in Reading",
"authors": [
{
"first": "Marten",
"middle": [],
"last": "Van Schijndel",
"suffix": ""
},
{
"first": "Tal",
"middle": [],
"last": "Linzen",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marten van Schijndel and Tal Linzen. 2018. A Neural Model of Adaptation in Reading. In Proceedings of EMNLP.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "The Effect of Word Predictability on Reading Time Is Logarithmic",
"authors": [
{
"first": "J",
"middle": [],
"last": "Nathaniel",
"suffix": ""
},
{
"first": "Roger",
"middle": [],
"last": "Smith",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Levy",
"suffix": ""
}
],
"year": 2013,
"venue": "Cognition",
"volume": "128",
"issue": "3",
"pages": "302--319",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nathaniel J Smith and Roger Levy. 2013. The Effect of Word Predictability on Reading Time Is Logarith- mic. Cognition, 128(3):302-319.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "BERT Rediscovers the Classical NLP Pipeline",
"authors": [
{
"first": "Ian",
"middle": [],
"last": "Tenney",
"suffix": ""
},
{
"first": "Dipanjan",
"middle": [],
"last": "Das",
"suffix": ""
},
{
"first": "Ellie",
"middle": [],
"last": "Pavlick",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1905.05950"
]
},
"num": null,
"urls": [],
"raw_text": "Ian Tenney, Dipanjan Das, and Ellie Pavlick. 2019. BERT Rediscovers the Classical NLP Pipeline. arXiv preprint arXiv:1905.05950.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "On the Predictive Power of Neural Language Models for Human Real-Time Comprehension Behavior",
"authors": [
{
"first": "Ethan",
"middle": [
"Gotlieb"
],
"last": "Wilcox",
"suffix": ""
},
{
"first": "Jon",
"middle": [],
"last": "Gauthier",
"suffix": ""
},
{
"first": "Jennifer",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Peng",
"middle": [],
"last": "Qian",
"suffix": ""
},
{
"first": "Roger",
"middle": [],
"last": "Levy",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ethan Gotlieb Wilcox, Jon Gauthier, Jennifer Hu, Peng Qian, and Roger Levy. 2020. On the Predictive Power of Neural Language Models for Human Real- Time Comprehension Behavior.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"uris": null,
"text": "The\"] S[1] = [\"The dog\"] S[2] = [\"The dog chases\"] S[3] = [\"The dog chases the\"] S[4] = [\"The dog chases the cat\"]",
"num": null
},
"FIGREF1": {
"type_str": "figure",
"uris": null,
"text": "Spearman's correlation of different layers of BERT on GECO.",
"num": null
},
"TABREF0": {
"text": "ModelHyperparameters Non-contextualized DSMs SVD.w2count DSM with 345K window-selected context words, window of width 2, reduced with SVD SVD.synt count DSM with 345K syntactically typed context words reduced with SVD GloVe count DSM with context window of width 2, reduced with log-bilinear regression SGNS.w2Skip-gram with negative sampling, context window of width 2, 15 negative examples SGNS.synt Skip-gram with negative sampling, syntactically-typed context words, 15 negative examples FastText Skip-gram with subword information, context window of width 2, 15 negative examples Contextualized DSMs ELMo Pretrained ELMo embeddings on the 1 Billion Word Benchmark BERT Pretrained BERT-Large embeddings on the concatenation of the Books corpus and Wikipedia",
"html": null,
"type_str": "table",
"num": null,
"content": "<table><tr><td>GPT2-xl</td><td>Pretrained GPT2-xl embeddings on WebText</td></tr><tr><td>Neural Complexity</td><td>Pretrained Neural Complexity embeddings on Wikipedia</td></tr></table>"
},
"TABREF1": {
"text": "List of the embedding models used for the study, together with their hyperparameter settings.",
"html": null,
"type_str": "table",
"num": null,
"content": "<table/>"
},
"TABREF3": {
"text": "Spearman correlations between the target-context cosine and the eye-tracking measures. Numbers in parenthesis indicate models' layers.",
"html": null,
"type_str": "table",
"num": null,
"content": "<table><tr><td>Corpus</td><td>Model</td><td colspan=\"3\">total reading time 1st fixation duration number fixations</td></tr><tr><td/><td>BERT</td><td>0.28</td><td>0.26</td><td>0.28</td></tr><tr><td>GECO</td><td>GPT2-xl</td><td>0.41</td><td>0.39</td><td>0.41</td></tr><tr><td/><td>NC</td><td>0.31</td><td>0.30</td><td>0.32</td></tr><tr><td/><td>BERT</td><td>0.25</td><td>0.24</td><td>0.24</td></tr><tr><td>Provo</td><td>GPT2-xl</td><td>0.44</td><td>0.43</td><td>0.44</td></tr><tr><td/><td>NC</td><td>0.46</td><td>0.48</td><td>0.46</td></tr></table>"
},
"TABREF4": {
"text": "Spearman correlations between surprisal and eye-tracking measures.",
"html": null,
"type_str": "table",
"num": null,
"content": "<table/>"
}
}
}
} |