File size: 57,381 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 | {
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T15:48:51.561481Z"
},
"title": "CTLR@WiC-TSV: Target Sense Verification using Marked Inputs and Pre-trained Models",
"authors": [
{
"first": "Jose",
"middle": [
"G"
],
"last": "Moreno",
"suffix": "",
"affiliation": {
"laboratory": "UMR 5505 CNRS F-31000",
"institution": "University of Toulouse IRIT",
"location": {
"settlement": "Toulouse",
"country": "France"
}
},
"email": "jose.moreno@irit.fr"
},
{
"first": "Elvys",
"middle": [
"Linhares"
],
"last": "Pontes",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University",
"location": {
"addrLine": "of La Rochelle L3i F-17000, La Rochelle",
"country": "France"
}
},
"email": "pontes@univ-lr.fr"
},
{
"first": "Ga\u00ebl",
"middle": [],
"last": "Dias",
"suffix": "",
"affiliation": {
"laboratory": "UMR 6072 CNRS F-14000",
"institution": "University of Caen GREYC",
"location": {
"settlement": "Caen",
"country": "France"
}
},
"email": "gael.dias@unicaen.fr"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper describes the CTRL participation in the Target Sense Verification of the Words in Context challenge (WiC-TSV) at SemDeep-6. Our strategy is based on a simplistic annotation scheme of the target words to later be classified by well-known pre-trained neural models. In particular, the marker allows to include position information to help models to correctly identify the word to disambiguate. Results on the challenge show that our strategy outperforms other participants (+11, 4 Accuracy points) and strong baselines (+1, 7 Accuracy points).",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper describes the CTRL participation in the Target Sense Verification of the Words in Context challenge (WiC-TSV) at SemDeep-6. Our strategy is based on a simplistic annotation scheme of the target words to later be classified by well-known pre-trained neural models. In particular, the marker allows to include position information to help models to correctly identify the word to disambiguate. Results on the challenge show that our strategy outperforms other participants (+11, 4 Accuracy points) and strong baselines (+1, 7 Accuracy points).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "This paper describes the CTLR 1 participation at the Word in Context challenge on the Target Sense Verification (WiC-TSV) task at SemDeep-6. In this challenge, given a target word w within its context participants are asked to solve a binary task organised in three sub-tasks:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Sub-task 1 consists in predicting if the target word matches with a given definition,",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Sub-task 2 consists in predicting if the target word matches with a given set of hypernyms, and",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Sub-task 3 consists in predicting if the target word matches with a given couple definition and set of hypernyms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our system is based on a masked neural language model with position information for Word Sense Disambiguation (WSD). Neural language models are recent and powerful resources useful for multiple Natural Language Processing (NLP) tasks (Devlin et al., 2018) . However, little effort has been made to perform tasks, where positions represent meaningful information. Regarding this line of research, Baldini Soares et al. (2019) include markers into the learning inputs for the task of relation classification and Boualili et al. (2020) into an information retrieval model. In both cases, the tokens allow the model to carefully identify the targets and to make an informed prediction. Besides these works, we are not aware of any other text-based tasks that have been tackled with this kind of information included into the models. To cover this gap, we propose to use markers to deal with target sense verification task.",
"cite_spans": [
{
"start": 222,
"end": 227,
"text": "(NLP)",
"ref_id": null
},
{
"start": 234,
"end": 255,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF3"
},
{
"start": 510,
"end": 532,
"text": "Boualili et al. (2020)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The remainder of this paper presents a brief background knowledge in Section 2. Details of our strategy, including input modification and prediction mixing is presented in Section 3. Then, unofficial and official results are presented in Section 4. Finally, conclusions are drawn in Section 5.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "NLP research has recently been boosted by new ways to use neural networks. Two main groups of neural networks can be distinguished 2 on NLP based on the training model and feature modification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "\u2022 First, classical neural networks usually use pre-trained embeddings as input and models learn their own weights during training time. Those weights are calculated directly on the target task and integration of new features or resources is intuitive. As an example, please refer to the Figure 1 the positional features (PF in the figure) that enrich the word embeddings (WF in the figure) to better represent the target words in the sentence. In this first group, models tend to use few parameters because embeddings are not fine-tuned. This characteristic does not dramatically impact the model performances.",
"cite_spans": [],
"ref_spans": [
{
"start": 287,
"end": 295,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "\u2022 The second group of models deals with neural language models 3 such as BERT (Devlin et al., 2018) . The main difference, w.r.t. the first group, is that the weights of the models are not calculated during the training step of the target task. Instead, they are pre-calculated in an elegant but expensive fashion by using generic tasks that deal with strong initialised models. Then, these models are fine-tuned to adapt their weights to the target task. 4 (Peters et al., 2019) or SenseBERT (Levine et al., 2019) .",
"cite_spans": [
{
"start": 78,
"end": 99,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF3"
},
{
"start": 456,
"end": 457,
"text": "4",
"ref_id": null
},
{
"start": 458,
"end": 479,
"text": "(Peters et al., 2019)",
"ref_id": "BIBREF6"
},
{
"start": 493,
"end": 514,
"text": "(Levine et al., 2019)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "We propose an alternative to mix the best of both worlds by including extra tokens into the input in order to improve prediction without re-training it. To do so, we base our strategy on the introduction of signals to the neural language models as depicted in Figure 1 In this work, we mark the target word by modifying the sentence in order to improve performance of BERT for the task of target sense verification.",
"cite_spans": [],
"ref_spans": [
{
"start": 260,
"end": 268,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "Given a first sentence with a known target word, a second sentence with a definition, and a set of hypernyms, the target sense verification task consists in defining whether or not the target word in the first sentence corresponds to the definition or/and the set of hypernyms. Note that two sub-problems may be set if only the second sentence or the hypernyms are used. These sub-problems are presented as sub-tasks in the WiC-TSV challenge.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem definition",
"sec_num": "3.1"
},
{
"text": "We implemented a target sense verification system as a simplified version 5 of the architecture proposed by Baldini Soares et al. 2019, namely BERT EM . It is based on BERT (Devlin et al., 2018) , where an extra layer is added to make the classification of the sentence representation, i.e. classification is performed using as input the [CLS] token. As reported by Baldini Soares et al. 2019, an important component is the use of mark symbols to identify the entities to classify. In our case, we mark the target word in its context to let the system know where to focus on.",
"cite_spans": [
{
"start": 173,
"end": 194,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF3"
},
{
"start": 338,
"end": 343,
"text": "[CLS]",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "CTLR method",
"sec_num": "3.2"
},
{
"text": "Learning the similarities between a couple of sentences (sub-task 1) can easily be addressed with BERT-based models by concatenating the two inputs one after the other one as presented in Equation 1, where S 1 and S 2 are two sentences given as inputs, t 1 i (i = 1..n) are the tokens in S 1 , and t 2 j (j = 1..m) are the tokens in S 2 . In this case, the model must learn to discriminate the correct definition and also to which of the words in S 1 the definition relates to.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "input(S 1 , S 2 ) = [CLS] t 1 1 t 1 2 ... t 1 n [SEP] t 2 1 t 2 2 ... t 2 m",
"eq_num": "(1)"
}
],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "To avoid the extra effort by the model to evidence the target word, we propose to introduce this information into the learning input. Thus, we mark the target word in S t by using a special token before and after the target word 6 . The input used when two sentences are compared is presented in Equation 2. S t is the first sentence with the target word t i , S d is the definition sentence, and t k",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "x are their respective tokens.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "input sp1 (S t , S d ) = [CLS] t t 1 t t 2 ... $ t t i $ ... t t n [SEP] t d 1 t d 2 ... t d m",
"eq_num": "(2)"
}
],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "In the case of hypernyms (sub-task 2), the input on the left side is kept as in Equation 2, but the right side includes the tagging of each hypernym as presented in Equation 3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "input sp2 (S t , S h ) = [CLS] t t 1 t t 2 ... $ t t i $ ... t t n [SEP] s h 1 $ s h 2 $ ... $ s h l",
"eq_num": "(3)"
}
],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "3.4 Verifying the senses",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "We trained two separated models, one for each subproblem using the architecture defined in Section 3.2. The output predictions of both models are 6 We used '$' but any other special token may be used.",
"cite_spans": [
{
"start": 146,
"end": 147,
"text": "6",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "used to solve the two-tasks problem. So, our overall prediction for the main problem is calculated by combining both prediction scores. First, we normalise the scores by applying a sof tmax function to each model output, and then we select the prediction with the maximum probability as shown in Equation 5.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "pred(x) = \uf8f1 \uf8f4 \uf8f2 \uf8f4 \uf8f3 1, if m sp1 1 (x) + m sp2 1 (x) > m sp1 0 (x) + m sp2 0 (x). 0, otherwise. (4)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "where",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "m spk i = exp(p spk i ) j={0,1} exp(p spk j )",
"eq_num": "(5)"
}
],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "and p spk i is the prediction value for the model k for the class i (m spk i ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pointing-out the target words",
"sec_num": "3.3"
},
{
"text": "The data set was manually created by the task organisers and some basic statistics are presented in Table 1 . Detailed information can be found in the task description paper (Breit et al., 2020) . No extra-annotated data was used for training. ",
"cite_spans": [
{
"start": 174,
"end": 194,
"text": "(Breit et al., 2020)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [
{
"start": 100,
"end": 107,
"text": "Table 1",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Data Sets",
"sec_num": "4.1"
},
{
"text": "We implemented BERT EM of Baldini Soares et al. (2019) using the huggingface library (Wolf et al., 2019) , and trained two models with each training set. We selected the model with best performance on the development set. Parameters were fixed as follows: 20 was used as maximum epochs, Cross Entropy as loss function, Adam as optimiser, bertbase-uncased 7 as pre-trained model, and other parameters were assigned following the library recommendations (Wolf et al., 2019) . The final layer is composed of two neurons (negative or positive). 1 (resp. 2, 3, and 4) includes all sentences for which the target word appears in the first (resp. second, third, and fourth) quarter of the sentence.",
"cite_spans": [
{
"start": 85,
"end": 104,
"text": "(Wolf et al., 2019)",
"ref_id": "BIBREF8"
},
{
"start": 452,
"end": 471,
"text": "(Wolf et al., 2019)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 541,
"end": 562,
"text": "1 (resp. 2, 3, and 4)",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Implementation details",
"sec_num": "4.2"
},
{
"text": "As the test labels are not publicly available, our following analysis is performed exclusively on the development set. Results on the test set were calculated by the task organisers. We analyse confusion matrices depending on the position of the target word in the sentence as our strategy is based on marking the target word. These matrices are presented in Figure 2 . The confusion matrix labelled as position group 1 shows our results when the target word is in the first 25% positions of the S t sentence. Other matrices show the results of the remaining parts of the sentence (second, third, and fourth 25%, for respectively group 2, 3, and 4).",
"cite_spans": [],
"ref_spans": [
{
"start": 359,
"end": 367,
"text": "Figure 2",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "Confusion matrices show that the easiest cases are when the target word is located in the first 25%. Other parts are harder mainly because the system considers positive examples as negatives (high false negative rate). However, the system behaves correctly for negative examples independently of the position of the target word. To better understand this wrong classification of the positive examples, we calculated the true label distribution depending on the normalised prediction score as in Figure 3 . Note that positive examples are mainly located on the right side but a bulk of them are located around the middle of the figure. It means that models m sp1 and m sp2 where in conflict and average results were slightly better for the negative class. In the development set, it seems important to correctly define a threshold strategy to better define which examples are marked as positive.",
"cite_spans": [],
"ref_spans": [
{
"start": 495,
"end": 504,
"text": "Figure 3",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "In our experiments, we implicitly used 0.5 as threshold 8 to define either the example belongs to the 'T' or 'F' class. When comparing Figures 3 and 4 , we can clearly see that small changes in the threshold parameter would affect our results with 8 Because of the condition m sp1 a larger impact in recall than in precision. This is mainly given to the fact that our two models contradict for some examples. We also considered the class distribution depending on a normalised distance between the target token and the beginning of the sentence. From Figure 5 , we observe that both classes are less frequent at the beginning of the sentence with negative examples slightly less frequent than positive ones. It is interesting to remark that negative examples uniformly distribute after the first bin. On the contrary, the positive examples have a more unpredictable distribution indicating that a strategy based on only positions may fail. However, our strategy that combines markers to indicate the target word and a ,1 75,8 70,7 73,2 87,5 82,4 90,3 86,2 85,9 86,7 85,8 86,3 83,3 78,4 88,5 83,1 szte begab 66,9 61,6 92,5 73,9 70,2 66,5 89,6 76,4 55,1 48,9 96,8 65,0 65,4 60,5 95,3 74,0 70,2 61,3 97,4 75,2 szte2 begab 66,3 61,1 92,8 73,7 69,9 66,2 90,2 76,3 53,7 48,1 96,8 64,3 64,4 59,8 95,3 73,5 69,6 60,8 97,4 74,9 BERT -76,6 74,1 82,8 78,2 73,5 76,1 74,2 75,1 79,2 67,8 98,2 80,2 79,8 75,8 89,6 82,1 82,1 73,0 97,9 83,6 FastText -53,4 52,8 79,4 63,4 57,1 58,0 74,0 65,0 43,1 43,1 100,0 60,2 51,1 51,5 90,3 65,6 54,0 50,5 67,1 57,3 Baseline (true) 50,8 50,8 100,0 67,3 53,8 53,8 100,0 70,0 43,1 43,1 100,0 60,2 51,7 51,7 100,0 68,2 46,4 46, Finally, the main results calculated by the organisers are presented in Table 2 . The global column presents the results for the global task, including definitions and hypernyms. Our submission is identified as run2-CTLR. In the global results, our strategy outperforms participants and baselines in terms of Accuracy, Precision, and F1. Best Recall performance is unsurprisingly obtained by the baseline (true) that corresponds to a system that predicts all examples as positives. Two strong baselines are included, FastText and BERT. Both baselines were calculated by the organisers with more details in (Breit et al., 2020) . It is interesting to remark that the baseline BERT is very similar to our model but without the marked information. However, our model focuses more on improving Precision than Recall resulting with a clear improvement in terms of Accuracy but less important in terms of F1.",
"cite_spans": [
{
"start": 2349,
"end": 2369,
"text": "(Breit et al., 2020)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [
{
"start": 135,
"end": 151,
"text": "Figures 3 and 4",
"ref_id": "FIGREF3"
},
{
"start": 547,
"end": 560,
"text": "From Figure 5",
"ref_id": "FIGREF5"
},
{
"start": 1019,
"end": 1742,
"text": ",1 75,8 70,7 73,2 87,5 82,4 90,3 86,2 85,9 86,7 85,8 86,3 83,3 78,4 88,5 83,1 szte begab 66,9 61,6 92,5 73,9 70,2 66,5 89,6 76,4 55,1 48,9 96,8 65,0 65,4 60,5 95,3 74,0 70,2 61,3 97,4 75,2 szte2 begab 66,3 61,1 92,8 73,7 69,9 66,2 90,2 76,3 53,7 48,1 96,8 64,3 64,4 59,8 95,3 73,5 69,6 60,8 97,4 74,9 BERT -76,6 74,1 82,8 78,2 73,5 76,1 74,2 75,1 79,2 67,8 98,2 80,2 79,8 75,8 89,6 82,1 82,1 73,0 97,9 83,6 FastText -53,4 52,8 79,4 63,4 57,1 58,0 74,0 65,0 43,1 43,1 100,0 60,2 51,1 51,5 90,3 65,6 54,0 50,5 67,1 57,3 Baseline (true) 50,8 50,8 100,0 67,3 53,8 53,8 100,0 70,0 43,1 43,1 100,0 60,2 51,7 51,7 100,0 68,2 46,4 46,",
"ref_id": "TABREF2"
},
{
"start": 1815,
"end": 1822,
"text": "Table 2",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "1 (x) + m sp2 1 (x) > m sp1 0 (x) + m sp2 0 (x).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "Organisers also provide results grouped by different types of examples. They included four types with three of them from domains that were not included in the training set 9 . From Table 2 , we can also conclude that our system is able to adapt 9 More details in (Breit et al., 2020) .",
"cite_spans": [
{
"start": 245,
"end": 246,
"text": "9",
"ref_id": null
},
{
"start": 263,
"end": 283,
"text": "(Breit et al., 2020)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [
{
"start": 181,
"end": 188,
"text": "Table 2",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "to out-of-domain topics as it is clearly shown for the Cocktails type in terms of F1, and also for the Medical entities type to a less extent. However, our system fails to provide better results than the standard BERT in terms of F1 for the Computer Science type. But, in terms of Accuracy, our strategy outperforms for a large margin the out-of-domain types (8.3, 6.1, and 1.2 improvements in absolute points for Cocktails, Medical entities, and Computer Science respectively). Surprisingly, it fails on both, F1 and Accuracy, for WordNet/Wiktionary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "4.3"
},
{
"text": "This paper describes our participation in the WiC-TSV task. We proposed a simple but effective strategy for target sense verification. Our system is based on BERT and introduces markers around the target words to better drive the learned model. Our results are strong over an unseen collection used to verify senses. Indeed, our method (Acc=78, 3) outperforms other participants (second best participant, Acc=66, 9) and strong baselines (BERT, Acc=76, 6) when compared in terms of Accuracy, the official metric. This margin is even larger when the results are compared for the out-of-domain examples of the test collection. Thus, the results suggest that the extra information provided to the BERT model through the markers clearly boost performance.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "As future work, we plan to complete the evaluation of our system with the WiC dataset (Pilehvar and Camacho-Collados, 2019) as well as the integration of the model into a recent multi-lingual entity linking system (Linhares Pontes et al., 2020) by marking the anchor texts.",
"cite_spans": [
{
"start": 214,
"end": 244,
"text": "(Linhares Pontes et al., 2020)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "University of Caen Normandie, University of Toulouse, and University of La Rochelle team.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We are aware that our classification is arguable. Although this is not an established classification in the field, it seems important for us to make a difference between them as this work tries to introduce well-established concepts from the first group into the second one.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Some subcategories may exist.4 We can image a combination of both, but models that use BERT as embeddings and do not fine-tune BERT weights may be classified in the first group.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We used the EntityMarkers[CLS] version.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/google-research/bert",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work has been partly supported by the European Union's Horizon 2020 research and innovation programme under grant 825153 (EMBED-DIA).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Matching the Blanks: Distributional Similarity for Relation Learning",
"authors": [
{
"first": "",
"middle": [],
"last": "Livio Baldini",
"suffix": ""
},
{
"first": "Nicholas",
"middle": [],
"last": "Soares",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Fitzgerald",
"suffix": ""
},
{
"first": "Tom",
"middle": [],
"last": "Ling",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Kwiatkowski",
"suffix": ""
}
],
"year": 2019,
"venue": "ACL",
"volume": "",
"issue": "",
"pages": "2895--2905",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Livio Baldini Soares, Nicholas FitzGerald, Jeffrey Ling, and Tom Kwiatkowski. 2019. Matching the Blanks: Distributional Similarity for Relation Learn- ing. In ACL. 2895-2905.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "MarkedBERT: Integrating Traditional IR Cues in Pre-Trained Language Models for Passage Retrieval",
"authors": [
{
"first": "Lila",
"middle": [],
"last": "Boualili",
"suffix": ""
},
{
"first": "Jose",
"middle": [
"G"
],
"last": "Moreno",
"suffix": ""
},
{
"first": "Mohand",
"middle": [],
"last": "Boughanem",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '20)",
"volume": "",
"issue": "",
"pages": "1977--1980",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lila Boualili, Jose G. Moreno, and Mohand Boughanem. 2020. MarkedBERT: Integrating Traditional IR Cues in Pre-Trained Language Models for Passage Retrieval. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '20). Association for Computing Machinery, 1977-1980.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Mohammad Taher Pilehvar, and Jose Camacho-Collados. 2020. WiC-TSV: An Evaluation Benchmark for Target Sense Verification of Words in Context",
"authors": [
{
"first": "Anna",
"middle": [],
"last": "Breit",
"suffix": ""
},
{
"first": "Artem",
"middle": [],
"last": "Revenko",
"suffix": ""
},
{
"first": "Kiamehr",
"middle": [],
"last": "Rezaee",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2004.15016"
]
},
"num": null,
"urls": [],
"raw_text": "Anna Breit, Artem Revenko, Kiamehr Rezaee, Moham- mad Taher Pilehvar, and Jose Camacho-Collados. 2020. WiC-TSV: An Evaluation Benchmark for Target Sense Verification of Words in Context. arXiv:2004.15016 (2020).",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"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 Un- derstanding. arXiv:1810.04805 (2018).",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Sensebert: Driving some Sense into Bert",
"authors": [
{
"first": "Yoav",
"middle": [],
"last": "Levine",
"suffix": ""
},
{
"first": "Barak",
"middle": [],
"last": "Lenz",
"suffix": ""
},
{
"first": "Or",
"middle": [],
"last": "Dagan",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Padnos",
"suffix": ""
},
{
"first": "Or",
"middle": [],
"last": "Sharir",
"suffix": ""
},
{
"first": "Shai",
"middle": [],
"last": "Shalev-Shwartz",
"suffix": ""
},
{
"first": "Amnon",
"middle": [],
"last": "Shashua",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Shoham",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1908.05646"
]
},
"num": null,
"urls": [],
"raw_text": "Yoav Levine, Barak Lenz, Or Dagan, Dan Padnos, Or Sharir, Shai Shalev-Shwartz, Amnon Shashua, and Yoav Shoham. 2019. Sensebert: Driving some Sense into Bert. arXiv:1908.05646 (2019).",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Linking Named Entities across Languages Using Multilingual Word Embeddings",
"authors": [
{
"first": "Jose",
"middle": [
"G"
],
"last": "Elvys Linhares Pontes",
"suffix": ""
},
{
"first": "Antoine",
"middle": [],
"last": "Moreno",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Doucet",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020 (JCDL '20)",
"volume": "",
"issue": "",
"pages": "329--332",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Elvys Linhares Pontes, Jose G. Moreno, and Antoine Doucet. 2020. Linking Named Entities across Lan- guages Using Multilingual Word Embeddings. In Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020 (JCDL '20). Association for Computing Machinery, 329-332.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Knowledge Enhanced Contextual Word Representations",
"authors": [
{
"first": "Matthew",
"middle": [
"E"
],
"last": "Peters",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Neumann",
"suffix": ""
},
{
"first": "Robert",
"middle": [
"L"
],
"last": "Logan",
"suffix": ""
},
{
"first": "Roy",
"middle": [],
"last": "Schwartz",
"suffix": ""
},
{
"first": "Vidur",
"middle": [],
"last": "Joshi",
"suffix": ""
},
{
"first": "Sameer",
"middle": [],
"last": "Singh",
"suffix": ""
},
{
"first": "Noah",
"middle": [
"A"
],
"last": "Smith",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "43--54",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew E. Peters, Mark Neumann, Robert L. Lo- gan, Roy Schwartz, Vidur Joshi, Sameer Singh, and Noah A. Smith. 2019. Knowledge Enhanced Con- textual Word Representations. In EMNLP. 43-54.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "WiC: the Word-in-Context Dataset for Evaluating Context-Sensitive Meaning Representations",
"authors": [
{
"first": "Mohammad",
"middle": [],
"last": "Taher Pilehvar",
"suffix": ""
},
{
"first": "Jose",
"middle": [],
"last": "Camacho-Collados",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "1267--1273",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mohammad Taher Pilehvar and Jose Camacho- Collados. 2019. WiC: the Word-in-Context Dataset for Evaluating Context-Sensitive Meaning Represen- tations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies, Volume 1 (Long and Short Papers). 1267- 1273.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Morgan Funtowicz, and Jamie Brew. 2019. HuggingFace's Transformers: State-of-the-art Natural Language Processing",
"authors": [
{
"first": "Thomas",
"middle": [],
"last": "Wolf",
"suffix": ""
},
{
"first": "Lysandre",
"middle": [],
"last": "Debut",
"suffix": ""
},
{
"first": "Victor",
"middle": [],
"last": "Sanh",
"suffix": ""
},
{
"first": "Julien",
"middle": [],
"last": "Chaumond",
"suffix": ""
},
{
"first": "Clement",
"middle": [],
"last": "Delangue",
"suffix": ""
},
{
"first": "Anthony",
"middle": [],
"last": "Moi",
"suffix": ""
},
{
"first": "Pierric",
"middle": [],
"last": "Cistac",
"suffix": ""
},
{
"first": "Tim",
"middle": [],
"last": "Rault",
"suffix": ""
},
{
"first": "R\u00e9mi",
"middle": [],
"last": "Louf",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1910.03771"
]
},
"num": null,
"urls": [],
"raw_text": "Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier- ric Cistac, Tim Rault, R\u00e9mi Louf, Morgan Funtow- icz, and Jamie Brew. 2019. HuggingFace's Trans- formers: State-of-the-art Natural Language Process- ing. arXiv:1910.03771 (2019).",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Relation classification via convolutional deep neural network",
"authors": [
{
"first": "Daojian",
"middle": [],
"last": "Zeng",
"suffix": ""
},
{
"first": "Kang",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Siwei",
"middle": [],
"last": "Lai",
"suffix": ""
},
{
"first": "Guangyou",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "Jun",
"middle": [],
"last": "Zhao",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers",
"volume": "",
"issue": "",
"pages": "2335--2344",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Daojian Zeng, Kang Liu, Siwei Lai, Guangyou Zhou, and Jun Zhao. 2014. Relation classification via convolutional deep neural network. In Proceedings of COLING 2014, the 25th International Confer- ence on Computational Linguistics: Technical Pa- pers. 2335-2344.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Representation examples for the relation classification problem proposed by Zeng et al. (2014) (a) and Baldini Soares et al. (2019) (b and c)."
},
"FIGREF1": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "(c) and done by Baldini Soares et al. (2019). Note that in this case the input is modified by introducing extra tokens ([E1], [/E1], [E2], and [/E2] are added based on target words (Baldini Soares et al., 2019)) that help the system to point out the target words."
},
"FIGREF2": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Confusion matrices for different position groups. Group"
},
"FIGREF3": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Histograms of predicted values in the dev set."
},
"FIGREF4": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Precision/Recall curve for the development set for different threshold values."
},
"FIGREF5": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Position distribution based on the target token distances."
},
"TABREF2": {
"type_str": "table",
"html": null,
"num": null,
"content": "<table/>",
"text": ""
},
"TABREF5": {
"type_str": "table",
"html": null,
"num": null,
"content": "<table><tr><td>strong neural language model (BERT) successfully</td></tr><tr><td>manage to classify the examples.</td></tr></table>",
"text": "Accuracy, Precision, Recall and F1 results of participants and baselines. Results where split by type. General results are included in column 'Global'. All results were calculated by the task organisers(Breit et al., 2020) as participants have not access to test labels. Best performance for each global metric is marked in bold for automatic systems."
}
}
}
} |