File size: 85,323 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 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T02:09:50.292256Z"
},
"title": "Creating a Domain-diverse Corpus for Theory-based Argument Quality Assessment",
"authors": [
{
"first": "Lily",
"middle": [],
"last": "Ng",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Anne",
"middle": [],
"last": "Lauscher",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Mannheim",
"location": {
"country": "Germany"
}
},
"email": ""
},
{
"first": "Joel",
"middle": [],
"last": "Tetreault",
"suffix": "",
"affiliation": {},
"email": "3jtetreault@dataminr.com"
},
{
"first": "Courtney",
"middle": [],
"last": "Napoles",
"suffix": "",
"affiliation": {},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Computational models of argument quality (AQ) have focused primarily on assessing the overall quality or just one specific characteristic of an argument, such as its convincingness or its clarity. However, previous work has claimed that assessment based on theoretical dimensions of argumentation could benefit writers, but developing such models has been limited by the lack of annotated data. In this work, we describe GAQCorpus, the first large, domain-diverse annotated corpus of theory-based AQ. We discuss how we designed the annotation task to reliably collect a large number of judgments with crowdsourcing, formulating theory-based guidelines that helped make subjective judgments of AQ more objective. We demonstrate how to identify arguments and adapt the annotation task for three diverse domains. Our work will inform research on theory-based argumentation annotation and enable the creation of more diverse corpora to support computational AQ assessment.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Computational models of argument quality (AQ) have focused primarily on assessing the overall quality or just one specific characteristic of an argument, such as its convincingness or its clarity. However, previous work has claimed that assessment based on theoretical dimensions of argumentation could benefit writers, but developing such models has been limited by the lack of annotated data. In this work, we describe GAQCorpus, the first large, domain-diverse annotated corpus of theory-based AQ. We discuss how we designed the annotation task to reliably collect a large number of judgments with crowdsourcing, formulating theory-based guidelines that helped make subjective judgments of AQ more objective. We demonstrate how to identify arguments and adapt the annotation task for three diverse domains. Our work will inform research on theory-based argumentation annotation and enable the creation of more diverse corpora to support computational AQ assessment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "The notion of Argumentation Quality (AQ) plays an important role in many existing argument-related downstream applications, such as argumentative writing support (Stab and Gurevych, 2017) , automatic essay grading (Persing and Ng, 2013) , and debate systems (Toledo et al., 2019) . For some of these applications, the idea is to automatically give feedback to users to help them improve their writing skills or assess their writing capabilities. For others, assessing AQ is an important step in a more complex pipeline for retrieving high-quality arguments.",
"cite_spans": [
{
"start": 162,
"end": 187,
"text": "(Stab and Gurevych, 2017)",
"ref_id": "BIBREF13"
},
{
"start": 214,
"end": 236,
"text": "(Persing and Ng, 2013)",
"ref_id": "BIBREF7"
},
{
"start": 258,
"end": 279,
"text": "(Toledo et al., 2019)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "While grading overall AQ (Toledo et al., 2019) or a specific conceptualization of AQ, such as prompt adherence (Persing and Ng, 2014) is relatively well explored, researchers have noted the lack of work in so-called theory-based AQ 1 (Wachsmuth et al., 2017b) , which can be represented with a taxonomy characterizing overall AQ into several subdimensions and aspects, for instance, as logic and rhetoric, which therefore provides a more informative and targeted perspective. However, this holistic approach comes with the downside of higher complexity, especially when it comes to annotating textual corpora, which are required for training and developing common computational approaches (see, e.g., Gretz et al. (2020) ). In a small study, Wachsmuth et al. (2017a) demonstrate that theory-based AQ annotations can be done both by trained experts and by crowd annotators, though the authors acknowledge the high complexity and subjectivity of the problem and call for the simplification of theory-based AQ annotation in order to reliably create larger-scale corpora. To date, no work has tackled this challenge and accordingly, no larger-scale and no domain-diverse corpus of this kind exists. We aim to close this gap by describing our efforts to create Grammarly Argument Quality Corpus (GAQCorpus) (Lauscher et al., 2020) , the largest and the only domain-diverse corpus consisting of 5,285 English arguments annotated with theory-based AQ scores across four dimensions.",
"cite_spans": [
{
"start": 25,
"end": 46,
"text": "(Toledo et al., 2019)",
"ref_id": "BIBREF14"
},
{
"start": 111,
"end": 133,
"text": "(Persing and Ng, 2014)",
"ref_id": "BIBREF8"
},
{
"start": 234,
"end": 259,
"text": "(Wachsmuth et al., 2017b)",
"ref_id": "BIBREF16"
},
{
"start": 701,
"end": 720,
"text": "Gretz et al. (2020)",
"ref_id": null
},
{
"start": 742,
"end": 766,
"text": "Wachsmuth et al. (2017a)",
"ref_id": "BIBREF15"
},
{
"start": 1302,
"end": 1325,
"text": "(Lauscher et al., 2020)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Building on Wachsmuth et al. (2017a) , in this work, we modify the annotation task to be suitable for both experts and the crowd while preserving the theoretical basis of the taxonomy. We collect and annotate argumentative texts from web debate forums, as well as community questions and answers forums (CQA), and review forum texts, which are still understudied in computational AQ. The latter domains can consist of rather non-canonical arguments in that they exhibit a lack of explicitness of certain argumentative components; are topic-wise more subjective; or consist of longer, more convoluted text. This makes assessing the quality of such arguments even more challenging, but downstream can result in a more robust model of computational AQ.",
"cite_spans": [
{
"start": 12,
"end": 36,
"text": "Wachsmuth et al. (2017a)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Given all these challenges, we work closely with trained linguists to adapt the annotation task, iterating over how best to approach these novel domains and simplify the annotation guidelines for crowdsourcing, allowing us to collect a large number of judgments efficiently. We hope that our work fuels further research on theory-based computational AQ. Our approach to building GAQCorpus can inspire and inform AQ annotation in new domains, enriching the domain-diversity of linguistic resources available in this space and consequently expanding computational approaches to AQ.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Structure. We start by surveying previous AQ annotation studies ( \u00a72). Next, we describe our efforts to adapt and simplify the annotation task ( \u00a73), which is followed by a discussion of the data domains ( \u00a74).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u00a75 presents an analysis of the resulting corpus. Finally, we conclude our work and provide directions for future research ( \u00a76). Figure 1 : The taxonomy of theory-based argument quality aspects (Wachsmuth et al., 2017b) .",
"cite_spans": [
{
"start": 194,
"end": 219,
"text": "(Wachsmuth et al., 2017b)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [
{
"start": 129,
"end": 137,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Most argumentation annotation studies have been conducted on student essays or web debates. Student essays have been annotated for thesis clarity (Persing and Ng, 2013) , organization (Persing et al., 2010) , and prompt adherence (Persing and Ng, 2014), and Persing and Ng (2015) model argument strength rated on a 4-point Likert scale. Similarly, Stab and Gurevych (2016) annotate the absence of opposing arguments and Stab and Gurevych (2017) predict insufficient premise support in arguments. For web debates, Habernal and Gurevych (2016) conduct an annotation study in which they present debate arguments pairwise to crowd annotators, who then can choose the more convincing argument. Persing and Ng (2017) also annotate the reasons why an argument receives a low persuasive power score. Wachsmuth et al. (2017b) developed a taxonomy of AQ synthesized from traditional works in argumentation theory, such as Aristotle (trans 2007) . The full taxonomy is depicted in Figure 1 , and defines the Overall AQ to consist of the following three subdimensions, each of which is itself defined by several finer-grained AQ aspects:",
"cite_spans": [
{
"start": 146,
"end": 168,
"text": "(Persing and Ng, 2013)",
"ref_id": "BIBREF7"
},
{
"start": 184,
"end": 206,
"text": "(Persing et al., 2010)",
"ref_id": "BIBREF11"
},
{
"start": 258,
"end": 279,
"text": "Persing and Ng (2015)",
"ref_id": "BIBREF9"
},
{
"start": 348,
"end": 372,
"text": "Stab and Gurevych (2016)",
"ref_id": "BIBREF12"
},
{
"start": 420,
"end": 444,
"text": "Stab and Gurevych (2017)",
"ref_id": "BIBREF13"
},
{
"start": 689,
"end": 710,
"text": "Persing and Ng (2017)",
"ref_id": "BIBREF10"
},
{
"start": 792,
"end": 816,
"text": "Wachsmuth et al. (2017b)",
"ref_id": "BIBREF16"
},
{
"start": 922,
"end": 934,
"text": "(trans 2007)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 970,
"end": 978,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "(1) Cogency relates to the logical aspects of AQ, for instance, whether the an argument's premises are acceptable (local acceptability) or whether they can be seen as relevant for the conclusion (local relevance).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "(2) Effectiveness indicates the rhetorical aspects of an argument. Aspects of effectiveness include, for instance, its clarity or its emotional appeal.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "(3) Reasonableness reflects the quality of an argument in the overall context of the discussion, as, for instance, its relevance towards arriving at a resolution of the issue (global relevance). Wachsmuth et al. (2017a) conducted a study in which crowd workers annotated 304 arguments for all 15 quality dimensions ( Figure 1) , and demonstrated that the theory-based and practical AQ assessments match to a large extent. Their findings indicate that theory-based annotations can be crowdsourced and that theory-based approaches can inform the practical view, especially. Most importantly, the authors conclude that the annotation task should be simplified to guarantee a reliable crowd-annotation process.",
"cite_spans": [
{
"start": 195,
"end": 219,
"text": "Wachsmuth et al. (2017a)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [
{
"start": 317,
"end": 326,
"text": "Figure 1)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Most recently, Toledo et al. (2019) and Gretz et al. (2020) crowdsourced overall argument quality by presenting pairwise arguments to annotators, who then had to select the argument \"they would recommend a friend to use that argument as is in a speech supporting/contesting the topic.\" This is an extreme simplification of the task, which does not seem to lead to better agreement: the authors (Gretz et al., 2020) report an average inter-annotator agreement of = 0.12 and attribute the low score to the high subjectivity of the task. The authors conducted a theory-based annotation study in the spirit of Wachsmuth et al. (2017b) on a subset of the data (100 arguments) which indicated the highest correlation of the annotations with the effectiveness dimension. Later on, Lauscher et al. (2020) empirically confirmed this observation using computational model predictions across the whole corpus.",
"cite_spans": [
{
"start": 15,
"end": 35,
"text": "Toledo et al. (2019)",
"ref_id": "BIBREF14"
},
{
"start": 40,
"end": 59,
"text": "Gretz et al. (2020)",
"ref_id": null
},
{
"start": 394,
"end": 414,
"text": "(Gretz et al., 2020)",
"ref_id": null
},
{
"start": 774,
"end": 796,
"text": "Lauscher et al. (2020)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Building on this large body of work, we aim to facilitate the annotation of theory-based AQ in diverse domains of real-world argumentative writing and compare expert vs. crowd annotations. Our study results in the largest English corpus annotated with theory-based argumentative quality scores.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "In this section, we detail how we developed and designed our annotation task to enable efficient, reliable collection of theory-based AQ judgments with crowdsourcing. We validate Wachsmuth et al. (2017a) 's hypothesis that crowdsourced annotation of theory-based AQ is possible if the task is simplified.",
"cite_spans": [
{
"start": 179,
"end": 203,
"text": "Wachsmuth et al. (2017a)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Study",
"sec_num": "3"
},
{
"text": "Before collecting any crowdsourced annotations, we conducted 14 pilot experiments with a group of four \"expert\" annotators, simplifying the TVSP task design through their feedback and observations, as they provided both a deep understanding of the argumentation theory and practical experience annotating the arguments. Each expert annotator was a fluent or native English speaker with an advanced degree in linguistics. Experts underwent training, which included studying guidelines and participating in calibration tasks to analyze debate arguments from three sources: Dagstuhl-ArgQuality-Corpus-V2 2 , originally from UKPConvArgRank (Habernal and Gurevych, 2016) ; the Internet Argument corpus V2 3 (IAC) (Abbott et al., 2016); and ChangeMyView, 4 a Reddit forum. Through the pilots and subsequent debriefs with the experts, we made the following modifications to the annotation task of Wachsmuth et al. (2017a):",
"cite_spans": [
{
"start": 636,
"end": 665,
"text": "(Habernal and Gurevych, 2016)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Simplifying the task",
"sec_num": "3.1"
},
{
"text": "(1) Reduce taxonomy complexity. While TVSP defined the task to score all 11 AQ subaspects (Local Acceptability, Local Relevance, etc.), 3 dimensions (Cogency, Effectiveness, Reasonableness), and overall AQ, we reduced the number of qualities scored by only focusing on the 3 higher-level dimensions plus overall AQ. As a result, annotators assessed an argumentative text in terms of 4 scores instead of 15 scores, and instead of 3 different AQ levels, the simplified taxonomy is reduced to 2.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplifying the task",
"sec_num": "3.1"
},
{
"text": "(2) Instruction Modifications. We reworded the TVSP dimension descriptions and added several examples to make the guidelines more understandable. As the annotators were not rating the 11 AQ subaspects, we experimented with different methods to incorporate the subaspects into the guidelines. Instead of explaining the subdimensions in the guidelines and trusting crowd annotators to bear them in mind, we represented each subdimension as a yes/no question in the annotation task itself (Table 1) . Our pilot experiments showed that presenting the questions without asking for a response eased the perceived complexity of the task while not affecting agreement.",
"cite_spans": [],
"ref_spans": [
{
"start": 486,
"end": 495,
"text": "(Table 1)",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Simplifying the task",
"sec_num": "3.1"
},
{
"text": "(3) Five-point scale. While TVSP collected judgments with a three-point rating scale (low, medium, high), we employ a five-point scale (very low, low, medium, high, very high, plus cannot judge) to allow for more nuanced judgments, as the expert annotators found too great of a distance between the items on a three-point scale. Scales with 5-9 items have been shown to be optimal, balancing the informational needs of the researcher and the capacity of the raters (Cox III, 1980) . We experimented with both threeand five-point scales and found that the larger scale did not negatively affect inter-annotator agreement.",
"cite_spans": [
{
"start": 465,
"end": 480,
"text": "(Cox III, 1980)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Simplifying the task",
"sec_num": "3.1"
},
{
"text": "Our finalized task design is as follows. First, annotators decide whether a text is argumentative. If yes, the three high-level dimensions are scored on a five-point scale and subaspect questions are presented to guide the annotator's judgment. The Overall AQ is scored last, also on a five-point scale. Before collecting annotations from the crowd, we validated our modifications subjectively and objectively. First, we ran a series of pilot tasks with our expert annotators. They initially annotated using the TVSP guidelines and next worked with the simplified taxonomy. In follow-up discussions, the experts confirmed that the new task design reduced the time and cognitive load necessary to rate arguments, and that the guidelines were more understandable. These modifications make the task more approachable, which is vital when presenting it to (untrained) crowd-workers for larger-scale annotation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Validating the Task Design",
"sec_num": "3.2"
},
{
"text": "We validated the simplifications quantitatively by reproducing the study of TVSP, which compared their crowd and \"expert\" annotations. To this end, we randomly sampled 200 arguments from Dagstuhl-ArgQuality-Corpus-V2, which come with author-annotated \"gold\" ratings. We collected ratings from a crowd (10 ratings per item), following our simplified design 5 ( \u00a73.1). All crowd contributors were native or fluent English speakers engaged through Appen (formerly Figure Eight) . Crowd contributors did not participate in calibration meetings and all feedback was relayed to contributors through a liaison.",
"cite_spans": [],
"ref_spans": [
{
"start": 461,
"end": 474,
"text": "Figure Eight)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Validating the Task Design",
"sec_num": "3.2"
},
{
"text": "We average the crowd ratings to obtain a single score for each argument and computed the interannotator agreement (IAA) with the \"gold\" annotations using Krippendorff's (Krippendorff, 2007) (Table 2) . Even though the annotation scores are not strong, the IAA between our crowd annotators and the gold annotations generally surpasses the agreement scores reported by TVSP. This is a highly nuanced and subjective task, which is reflected in the agreement levels. Based on these results and annotator observations, we conclude that our task guidelines and design allow for better (or at least comparable) quality crowdsourcing of theory-based AQ annotations.",
"cite_spans": [
{
"start": 154,
"end": 189,
"text": "Krippendorff's (Krippendorff, 2007)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 190,
"end": 199,
"text": "(Table 2)",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Validating the Task Design",
"sec_num": "3.2"
},
{
"text": "In this work, we consider three domains: Debate forums, CQA forums, and Review forums. While Debates are generally well-explored in computational AQ, we are unaware of any work involving CQA and Reviews. For each of these domains, we first identified items likely to be argumentative and then adjusted the guidelines in consultation with expert annotators, as described below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "Debate forums. Of these three domains, Debates is the most straightforward to annotate. Given a topic or motion, users can define their stance (pro/contra) and write an argument which supports it. We included data from two online debate forums. ConvinceMe (CM) is a subset of the IAC, where users share their Stance on a topic and discuss their point of view, with replies aiming to change the view of the original poster. Change My View (CMV) is a Reddit forum in which participants post their opinion on a topic and ask others to post replies to change their mind. We sampled original posts from CMV, skipping any moderator posts, and the first reply to an original post from CM, in order to limit the context that annotators must consider when evaluating arguments. CMV posts always include the author's perspective in the title, while CM posts may or may not include a stance in the title. In the guidelines, we instruct annotators to judge a post by how successfully it justifies the author's claim.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "CQA. In community questions and answers forums, users post questions or ask for advice, which other users can address. We experimented with arguments from Yahoo! Answers 6 (YA). When posting a question, a user can provide background information for their question (context) and can later indicate which response is the best answer to their question. The forum's looser structure provides for a wide variety of content, which is appealing as a potential source of non-standard arguments, but challenging as many of the posts do not contain any arguments. Through manual analysis, we identified three categories that frequently contained controversial topics, hypothesizing they would have a higher incidence of debates: Social Science > Sociology, Society & Culture > Other, and Politics & Government > Law & Ethics. We empirically selected the category with the highest proportion of arguments in a study on Amazon Mechanical Turk (MTurk). Qualified annotators 7 decided if question and best-answer pairs were argumentative. We collected 10 judgments for 100 pairs from each category and aggregated judgments with a simple majority. Law & Ethics had the most argumentative posts (70%, compared to Sociology with 40% and Society & Culture with 34%), so we sampled posts from this category to annotate.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "In the guidelines for this domain, we asked annotators to judge the argumentative strength of an answer with respect to how well it addressed the given question. The guidelines and subdimension questions were altered to encourage this. One obstacle in pilot studies with expert annotators was posts offering, as many users solicited legal advice in the Law & Ethics forum. We decided to consider advice as argumentative as long as the author supported the advice with justification, which mirrors our general approach to the Argumentative dimension.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "Reviews. The third domain consists of restaurant reviews from the Yelp-Challenge-Dataset 8 . On Yelp, users write reviews of businesses and rate the quality of their experience from 1 (low) to 5 (high) stars. Unlike the Debate and Q&A forums, the format of Yelp does not support dialogue between users (i.e., users cannot directly reply to other users or posts), and so it is possible to present each post in isolation as a self-contained argument. As most posts do not explicitly state a claim, we pose the star rating as a claim the user is making about the business, and the review as the argument supporting it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "Yelp reviews can be highly subjective in that each review is based on a single user's experience. For instance, a user may rate a restaurant as 5-stars and write only The food was delicious in their review. To address this subjectivity, we asked annotators to judge the argumentative quality of each review with respect to how well it supported the rating provided. Another challenge was defining what constituted a counterargument, as these have a very different character than counterarguments in debates (for example, Everyone says that the pizza crust is too thin here but that's authentic!). In consultation with our experts, we defined counterarguments by the following characteristics: 1) addressing and rebutting the viewpoints of other reviews, 2) addressing and rebutting points that discredit the author's rating, and 3) bringing up favorable points in an unfavorable review and vice versa.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "Experts completed a series of pilots before each domain was presented to the crowd, using the task design described in \u00a73.1. Expert agreement on novel domains (YA and Yelp) are shown in Table 4 . Feedback on the task and guidelines was gathered during calibration meetings with experts, and they were iteratively altered to be more clear and specific. Applying the annotation task design and data selection described above, we created GAQCorpus, containing 5,285 arguments across three domains, annotated for theory-based dimensions. All arguments were limited to have a length between 70 and 200 characters. Ratings were provided by the two groups of annotators described above, Experts ( \u00a73.1) and the Crowd ( \u00a73.2). Each group judged 3,000 arguments, with about 1,000 arguments annotated by both groups for comparison. The size of the corpus is described in Table 3 . Annotators worked with the domains in the following order: Debate forums, CQA forums, and Review forums. Before switching to a new domain, annotators completed a small study for calibration. All data and guidelines are available from https://github.com/grammarly/gaqcorpus. We assessed the quality of the crowd annotations by calculating the agreement between the experts and crowd workers on the overlapping portions of GAQCorpus using the mean scores (Table 5) . For debate forums, the agreement is weak with \u2264 0.21, while for the CQA forums, the agreement is higher: 0.42-0.53. These results suggest that the difficulty of the task is highly dependent on the domain. While our Debates data and the DS data both consist of web debate arguments, the difference in IAA is high, which might be attributed to different complexities of the web debates data. While TVSP only look at single arguments in isolation, often consisting of a single sentence only.",
"cite_spans": [],
"ref_spans": [
{
"start": 186,
"end": 193,
"text": "Table 4",
"ref_id": "TABREF5"
},
{
"start": 861,
"end": 868,
"text": "Table 3",
"ref_id": "TABREF8"
},
{
"start": 1324,
"end": 1333,
"text": "(Table 5)",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "One area of disagreement centered on arguments which were sarcastic, ironic, or included rhetorical questions. Consider the argument given in Figure 2 , over which the expert annotators expressed disagreement. This argument appears to support the stance that a government has the right to censor speech, but several linguistic cues indicate that the argument might be ironic: (a) Punctuation: ellipsis indicates thinking/searching for justifications; similarly, (b) the filler um; (c) capitalization: the noun phrase Our Leader is capitalized, indicating hyperbolic apotheosis; and finally, (d) the phrase (...) so I have to argue for this side. acts like an apologia, which is put in front of the actual argument. Annotators 1 and 2 based their judgments on an interpretation of this text that related to the estimated degree of irony in the post. While Annotator 1 did not perceive irony and judged the argument as very weak in Effectiveness, Annotator 2 considered it to be highly effective as in their view, the irony positively underlined the perceived stance. Annotator 3 gave medium scores across the board. Such disagreements were regularly discussed and usually revealed that multiple opinions may exist according to how the texts were interpreted, highlighting the high subjectivity of the task. Another area of disagreement was how to judge arguments on topics that were deemed \"less worthy\" of being discussed, and which usually were humorous in nature or had trivial consequences, such as Batman vs Superman, in which users argued for the the superiority of either superhero. In pilots, some experts provided lower ratings of arguments on a topic that they considered less worthy Others thought that writing a strong, serious argument on a less worthy topic was especially difficult, and thus provided higher ratings for such arguments.",
"cite_spans": [],
"ref_spans": [
{
"start": 142,
"end": 150,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Data Domains",
"sec_num": "4"
},
{
"text": "The distributions of mean scores across domains and annotator groups in GAQCorpus are depicted in Figures 3a and 3b . In general, the interquartile range of the expert scores was higher than the crowd, suggesting that experts were more specific when scoring items, which is also reflected in the medians: while the crowd exhibits a tendency to score variables equally, expert annotations exhibit more differentiation.",
"cite_spans": [],
"ref_spans": [
{
"start": 98,
"end": 115,
"text": "Figures 3a and 3b",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Analysis of Scores",
"sec_num": "5.2"
},
{
"text": "To understand the interrelations between Overall AQ and the dimensions, we compute Pearson correlations between the mean scores (Figure 4) . Generally, the trends are similar across all three domains. For instance, for Debates (Figures 4d and 4a) , the crowd annotations exhibit stronger correlations between the different dimension scores than the experts, with 0.83 \u2264 \u2264 0.96. Interestingly, the variance among the Pearson scores is lower, indicating that the crowd tends to distribute ratings for a single instance more consistently while the experts seem to put more weight on differentiating the dimensions.",
"cite_spans": [],
"ref_spans": [
{
"start": 128,
"end": 138,
"text": "(Figure 4)",
"ref_id": "FIGREF4"
},
{
"start": 227,
"end": 246,
"text": "(Figures 4d and 4a)",
"ref_id": "FIGREF4"
}
],
"eq_spans": [],
"section": "Analysis of Scores",
"sec_num": "5.2"
},
{
"text": "Expert ratings of Overall AQ have substantially stronger correlation with the dimensions than any of the dimension scores with each other, further indicating that experts are more discerning in their scores than the crowd. Across both annotator groups and all domains, the correlation between Overall AQ and Reasonableness is highest, which is consistent with earlier observations (Wachsmuth et al., 2017b ",
"cite_spans": [
{
"start": 381,
"end": 405,
"text": "(Wachsmuth et al., 2017b",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Analysis of Scores",
"sec_num": "5.2"
},
{
"text": "We next examine low-scoring arguments from all domains to understand how AQ is perceived differently, focusing on the Reasonableness dimension. Table 6 shows a low-scoring argument from each domain. The Debate argument raises a counterargument but does not rebut it and additionally neglects to address an obvious counterargument (i.e., the many ethical implications of such a policy). On the other hand, the CQA and Review arguments do not raise or address any counterarguments and are not judged Reasonable for other reasons: the CQA argument jokes about the original poster's question and accuses the poster of malignant behavior, while the Review argument delves into a personal experience that does not contribute to the discussion about the quality of the business.",
"cite_spans": [],
"ref_spans": [
{
"start": 144,
"end": 151,
"text": "Table 6",
"ref_id": "TABREF11"
}
],
"eq_spans": [],
"section": "Qualitative Analysis",
"sec_num": "5.3"
},
{
"text": "Theory-based AQ assessment provides a holistic and targeted perspective on AQ, but its high complexity makes annotation difficult. In this work, we describe our efforts to create GAQCorpus, a multi-domain corpus of 5,295 arguments annotated for quality along theory-based AQ dimensions. We demonstrate that it is possible to collect complex annotations with crowdsourcing in three domains: Debate, CQA, and Review forums. Drawing from the initial study of Wachsmuth et al. (2017a) , which suggested the general feasibility, we relied on the intuition of trained linguists to simplify the task and guidelines while preserving the theoretical basis of the task. The agreement between experts and the crowd was higher than the agreement in earlier studies (Wachsmuth et al., 2017a) , validating our approach and indicating that it is possible to collect complex ratings using a crowd.",
"cite_spans": [
{
"start": 456,
"end": 480,
"text": "Wachsmuth et al. (2017a)",
"ref_id": "BIBREF15"
},
{
"start": 753,
"end": 778,
"text": "(Wachsmuth et al., 2017a)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "GAQCorpus and the findings of our annotation study will serve as a basis for future corpus development and computational model development in theory-based AQ. They are available for download from https://github.com/grammarly/gaqcorpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "2.0 Effectiveness 1.7 Reasonableness 1.0 Overall 1.3",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "Title: Should you need to pass an IQ test to have kids?-Stance: Dumb parents lead to more dumb kids. Text: I have a strong opinion that before having children, the prospective parents should have to pass a series of background and IQ tests. Kids being brought into this world need a good foundation to start a successful life with. You may have that limited case where the parents are morons and the kids strive to be different then their failure parents, but in most cases it is an endless line of parasites on our world. We need more smart people.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "CQA Cogency 2.7 Effectiveness 2.0 Reasonableness 1.7 Overall 2.0",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "Question: Bounced CHECK? Context: Does the company holding the bounced check have to send you a certified letter before issuing a warrant for your arrest. I feel almost certain that they do but i am not sure. Answer: I always make sure my checks are not printed on rubber. they are just too expensive and not worth it. We all make a mistake from time to time, and usually it is no big deal except for the extreme annoyance and all the bounced check fees. But if you are worried about an arrest warrant then I am sure you are doing this deliberately and trying to defraud the company. You have probably sent them a couple of bad checks already in an attempt to string them along so your guilt is probably pretty well established. You can hope that you do not have to share a jail cell with a gross deviate of some sort.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "Reviews Cogency 1.0 Effectiveness 1.0 Reasonableness 1.0 Overall 1.0",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "Title: Business review: 2.0 Stars. Business name: Cook Out. City: Charlotte. Categories: Restaurants, Desserts, Food, Fast Food, American (Traditional), Hot Dogs, Burgers Review: Burgers are good but I like those other 5 guys burgers instead oh and I guess if your not from around here don't even think about going thru the drive thru it's like the biggest most unreadable confusing hurried crazy thing ever if I ever go again hell with drive thru until I've lived here for at least 5 maybe 10 years and can be a veteran drive thru person I'm walking in it's like if I mix up all the letters in this review and give you 1 minute to read it and figure it out then you gotta move on. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cogency",
"sec_num": null
},
{
"text": "In the following, we adopt the term \"theory-based AQ,\" which was proposed byWachsmuth et al. (2017b) to indicate that the conception of AQ is specifically grounded in argumentation theoretic literature (and not in CL or NLP).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://argumentation.bplaced.net/arguana/data 3 https://nlds.soe.ucsc.edu/iac2 4 https://www.reddit.com/r/changemyview/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The only difference is that we used a 3-point scale to more fairly compare to the gold.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://answers.yahoo.com/ 7 HIT approval rate >= 97; HITs approved > 500; Location = US 8 https://www.yelp.com/dataset",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "The work of Anne Lauscher is supported by the Eliteprogramm of the Baden-W\u00fcrttemberg Stiftung (AGREE grant). We thank our linguistic expert annotators for providing interesting insights and discussions as well as the anonymous reviewers for their helpful comments. We also thank Henning Wachsmuth for consulting us w.r.t. his previous work and Yahoo! for granting us access to their data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Internet argument corpus 2.0: An SQL schema for dialogic social media and the corpora to go with it",
"authors": [
{
"first": "Rob",
"middle": [],
"last": "Abbott",
"suffix": ""
},
{
"first": "Brian",
"middle": [],
"last": "Ecker",
"suffix": ""
},
{
"first": "Pranav",
"middle": [],
"last": "Anand",
"suffix": ""
},
{
"first": "Marilyn",
"middle": [],
"last": "Walker",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016)",
"volume": "",
"issue": "",
"pages": "4445--4452",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rob Abbott, Brian Ecker, Pranav Anand, and Marilyn Walker. 2016. Internet argument corpus 2.0: An SQL schema for dialogic social media and the corpora to go with it. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016), pages 4445-4452, Portoro\u017e, Slovenia, May. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "On Rhetoric: A Theory of Civic Discourse",
"authors": [
{
"first": "Aristotle",
"middle": [],
"last": "Trans",
"suffix": ""
}
],
"year": 2007,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Aristotle. trans. 2007. On Rhetoric: A Theory of Civic Discourse. Oxford University Press, Oxford, UK. Trans- lated by George A. Kennedy.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "The optimal number of response alternatives for a scale: A review",
"authors": [
{
"first": "Eli",
"middle": [
"P"
],
"last": "Cox",
"suffix": ""
},
{
"first": "Iii",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 1980,
"venue": "Journal of Marketing Research",
"volume": "17",
"issue": "4",
"pages": "407--422",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eli P. Cox III. 1980. The optimal number of response alternatives for a scale: A review. Journal of Marketing Research, 17(4):407-422.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Ranit Aharonov, and Noam Slonim. 2020. A large-scale dataset for argument quality ranking: Construction and analysis",
"authors": [
{
"first": "Shai",
"middle": [],
"last": "Gretz",
"suffix": ""
},
{
"first": "Roni",
"middle": [],
"last": "Friedman",
"suffix": ""
},
{
"first": "Edo",
"middle": [],
"last": "Cohen-Karlik",
"suffix": ""
},
{
"first": "Assaf",
"middle": [],
"last": "Toledo",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Lahav",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of AAAI2020",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shai Gretz, Roni Friedman, Edo Cohen-Karlik, Assaf Toledo, Dan Lahav, Ranit Aharonov, and Noam Slonim. 2020. A large-scale dataset for argument quality ranking: Construction and analysis. In Proceedings of AAAI2020.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Which argument is more convincing? analyzing and predicting convincingness of web arguments using bidirectional LSTM",
"authors": [
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1589--1599",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ivan Habernal and Iryna Gurevych. 2016. Which argument is more convincing? analyzing and predicting con- vincingness of web arguments using bidirectional LSTM. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1589-1599, Berlin, Germany, Au- gust. Association for Computational Linguistics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Computing krippendorff's alpha-reliability",
"authors": [
{
"first": "Klaus",
"middle": [],
"last": "Krippendorff",
"suffix": ""
}
],
"year": 2007,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Klaus Krippendorff. 2007. Computing krippendorff's alpha-reliability. Technical report, University of Pennsyl- vania, Annenberg School for Communication.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Rhetoric, logic, and dialectic: Advancing theory-based argument quality assessment in natural language processing",
"authors": [
{
"first": "Anne",
"middle": [],
"last": "Lauscher",
"suffix": ""
},
{
"first": "Lily",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "Courtney",
"middle": [],
"last": "Napoles",
"suffix": ""
},
{
"first": "Joel",
"middle": [],
"last": "Tetreault",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 28th International Conference on Computational Linguistics (COLING 2020",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Anne Lauscher, Lily Ng, Courtney Napoles, and Joel Tetreault. 2020. Rhetoric, logic, and dialectic: Advancing theory-based argument quality assessment in natural language processing. In Proceedings of the 28th Interna- tional Conference on Computational Linguistics (COLING 2020).",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Modeling thesis clarity in student essays",
"authors": [
{
"first": "Isaac",
"middle": [],
"last": "Persing",
"suffix": ""
},
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "260--269",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isaac Persing and Vincent Ng. 2013. Modeling thesis clarity in student essays. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 260-269, Sofia, Bulgaria, August. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Modeling prompt adherence in student essays",
"authors": [
{
"first": "Isaac",
"middle": [],
"last": "Persing",
"suffix": ""
},
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1534--1543",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isaac Persing and Vincent Ng. 2014. Modeling prompt adherence in student essays. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1534-1543, Baltimore, Maryland, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Modeling argument strength in student essays",
"authors": [
{
"first": "Isaac",
"middle": [],
"last": "Persing",
"suffix": ""
},
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing",
"volume": "1",
"issue": "",
"pages": "543--552",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isaac Persing and Vincent Ng. 2015. Modeling argument strength in student essays. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 543-552.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Why can't you convince me? modeling weaknesses in unpersuasive arguments",
"authors": [
{
"first": "Isaac",
"middle": [],
"last": "Persing",
"suffix": ""
},
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 26th International Joint Conference on Artificial Intelligence, IJCAI'17",
"volume": "",
"issue": "",
"pages": "4082--4088",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isaac Persing and Vincent Ng. 2017. Why can't you convince me? modeling weaknesses in unpersuasive argu- ments. In Proceedings of the 26th International Joint Conference on Artificial Intelligence, IJCAI'17, pages 4082-4088. AAAI Press.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Modeling organization in student essays",
"authors": [
{
"first": "Isaac",
"middle": [],
"last": "Persing",
"suffix": ""
},
{
"first": "Alan",
"middle": [],
"last": "Davis",
"suffix": ""
},
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, EMNLP '10",
"volume": "",
"issue": "",
"pages": "229--239",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isaac Persing, Alan Davis, and Vincent Ng. 2010. Modeling organization in student essays. In Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, EMNLP '10, pages 229-239, Stroudsburg, PA, USA. Association for Computational Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Recognizing the absence of opposing arguments in persuasive essays",
"authors": [
{
"first": "Christian",
"middle": [],
"last": "Stab",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Third Workshop on Argument Mining (ArgMining2016)",
"volume": "",
"issue": "",
"pages": "113--118",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christian Stab and Iryna Gurevych. 2016. Recognizing the absence of opposing arguments in persuasive essays. In Proceedings of the Third Workshop on Argument Mining (ArgMining2016), pages 113-118.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Recognizing insufficiently supported arguments in argumentative essays",
"authors": [
{
"first": "Christian",
"middle": [],
"last": "Stab",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "980--990",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christian Stab and Iryna Gurevych. 2017. Recognizing insufficiently supported arguments in argumentative essays. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 980-990, Valencia, Spain, April. Association for Computational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Automatic argument quality assessment-new datasets and methods",
"authors": [
{
"first": "Assaf",
"middle": [],
"last": "Toledo",
"suffix": ""
},
{
"first": "Shai",
"middle": [],
"last": "Gretz",
"suffix": ""
},
{
"first": "Edo",
"middle": [],
"last": "Cohen-Karlik",
"suffix": ""
},
{
"first": "Roni",
"middle": [],
"last": "Friedman",
"suffix": ""
},
{
"first": "Elad",
"middle": [],
"last": "Venezian",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Lahav",
"suffix": ""
},
{
"first": "Michal",
"middle": [],
"last": "Jacovi",
"suffix": ""
},
{
"first": "Ranit",
"middle": [],
"last": "Aharonov",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Slonim",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
"volume": "",
"issue": "",
"pages": "5629--5639",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Assaf Toledo, Shai Gretz, Edo Cohen-Karlik, Roni Friedman, Elad Venezian, Dan Lahav, Michal Jacovi, Ranit Aharonov, and Noam Slonim. 2019. Automatic argument quality assessment-new datasets and methods. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Inter- national Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5629-5639.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Argumentation quality assessment: Theory vs. practice",
"authors": [
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Nona",
"middle": [],
"last": "Naderi",
"suffix": ""
},
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Yufang",
"middle": [],
"last": "Hou",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Hirst",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "250--255",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Henning Wachsmuth, Nona Naderi, Ivan Habernal, Yufang Hou, Graeme Hirst, Iryna Gurevych, and Benno Stein. 2017a. Argumentation quality assessment: Theory vs. practice. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 250-255, Vancouver, Canada, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Computational argumentation quality assessment in natural language",
"authors": [
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Nona",
"middle": [],
"last": "Naderi",
"suffix": ""
},
{
"first": "Yufang",
"middle": [],
"last": "Hou",
"suffix": ""
},
{
"first": "Yonatan",
"middle": [],
"last": "Bilu",
"suffix": ""
},
{
"first": "Vinodkumar",
"middle": [],
"last": "Prabhakaran",
"suffix": ""
},
{
"first": "Tim",
"middle": [
"Alberdingk"
],
"last": "Thijm",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Hirst",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "176--187",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Henning Wachsmuth, Nona Naderi, Yufang Hou, Yonatan Bilu, Vinodkumar Prabhakaran, Tim Alberdingk Thijm, Graeme Hirst, and Benno Stein. 2017b. Computational argumentation quality assessment in natural language. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguis- tics: Volume 1, Long Papers, pages 176-187, Valencia, Spain, April. Association for Computational Linguistics.",
"links": null
}
},
"ref_entries": {
"FIGREF1": {
"num": null,
"text": "Example argument exhibiting disagreement in the Effectiveness dimension.",
"uris": null,
"type_str": "figure"
},
"FIGREF3": {
"num": null,
"text": "Score distributions by domain for expert and crowd annotators.",
"uris": null,
"type_str": "figure"
},
"FIGREF4": {
"num": null,
"text": "Mean score correlations between the different dimensions for expert and crowd annotators across the three domains (Pearson's ).",
"uris": null,
"type_str": "figure"
},
"TABREF1": {
"num": null,
"html": null,
"text": "Subdimensions represented as questions in the annotation task of debates.",
"type_str": "table",
"content": "<table/>"
},
"TABREF3": {
"num": null,
"html": null,
"text": "Agreement Dagstuhl \"gold\" annotations and our crowdsourced annotations (Ours) compared to TVSP.",
"type_str": "table",
"content": "<table/>"
},
"TABREF5": {
"num": null,
"html": null,
"text": "Agreement(Krippendorff's ) between experts on pilot studies for CQA, Debates, and 150, and 50 arguments, respectively).",
"type_str": "table",
"content": "<table><tr><td/><td colspan=\"4\">Cogency Effectiveness Reasonableness Overall</td></tr><tr><td>CQA</td><td>0.42</td><td>0.52</td><td>0.52</td><td>0.53</td></tr><tr><td colspan=\"2\">Debates 0.14</td><td>0.11</td><td>0.21</td><td>0.19</td></tr><tr><td colspan=\"2\">Reviews 0.32</td><td>0.32</td><td>0.31</td><td>0.33</td></tr></table>"
},
"TABREF6": {
"num": null,
"html": null,
"text": "IAA between the mean expert and crowd scores for Cogency, Effectiveness, Reasonableness, and Overall AQ.",
"type_str": "table",
"content": "<table/>"
},
"TABREF8": {
"num": null,
"html": null,
"text": "Number of arguments annotated by experts and the crowd and the number of overlapping instances (annotated by both experts and the crowd) by domain.",
"type_str": "table",
"content": "<table/>"
},
"TABREF9": {
"num": null,
"html": null,
"text": "My government doesn't give me freedom of speech, so I have to argue for this side. Freedom of speech is bad because ... um ... then Our Leader's beliefs could be challenged. No one wants that. I mean, if everyone would just say and believe what Our Leader says to, we wouldn't need those firing squads altogether! Everyone wins.",
"type_str": "table",
"content": "<table><tr><td>5.1 Inter-annotator Agreements (IAA)</td><td/><td/><td/><td/></tr><tr><td/><td colspan=\"4\">Cogency Effectiveness Reasonableness Overall</td></tr><tr><td>Annotator 1</td><td>4</td><td>1</td><td>1</td><td>2</td></tr><tr><td>Annotator 2</td><td>4</td><td>5</td><td>3</td><td>4</td></tr><tr><td>Annotator 3</td><td>2</td><td>2</td><td>2</td><td>2</td></tr></table>"
},
"TABREF10": {
"num": null,
"html": null,
"text": ").",
"type_str": "table",
"content": "<table><tr><td>overall</td><td>1</td><td>0.85</td><td>0.8</td><td>0.84</td><td/><td>overall</td><td>1</td><td>0.78</td><td>0.81</td><td>0.83</td><td/><td>overall</td><td>1</td><td>0.87</td><td>0.85</td><td>0.87</td><td/></tr><tr><td>cogency</td><td>0.85</td><td>1</td><td>0.62</td><td>0.72</td><td>0.88 0.96</td><td>cogency</td><td>0.78</td><td>1</td><td>0.63</td><td>0.67</td><td>0.88 0.96</td><td>cogency</td><td>0.87</td><td>1</td><td>0.74</td><td>0.76</td><td>0.90 0.96</td></tr><tr><td>effectiveness reasonableness</td><td>0.8 0.84</td><td>0.62 0.72</td><td>1 0.6</td><td>0.6 1</td><td>0.64 0.72 0.80</td><td>effectiveness reasonableness</td><td>0.81 0.83</td><td>0.63 0.67</td><td>1 0.68</td><td>0.68 1</td><td>0.64 0.72 0.80</td><td>reasonableness effectiveness</td><td>0.87 0.85</td><td>0.76 0.74</td><td>0.69 1</td><td>1 0.69</td><td>0.72 0.78 0.84</td></tr><tr><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/></tr><tr><td/><td colspan=\"4\">(a) Experts on Debates.</td><td/><td/><td colspan=\"4\">(b) Experts on Reviews.</td><td/><td/><td colspan=\"4\">(c) Experts on CQA</td><td/></tr><tr><td>overall</td><td>1</td><td>0.93</td><td>0.93</td><td>0.95</td><td/><td>overall</td><td>1</td><td>0.95</td><td>0.94</td><td>0.96</td><td/><td>overall</td><td>1</td><td>0.96</td><td>0.96</td><td>0.96</td><td/></tr><tr><td/><td/><td/><td/><td/><td>1.000</td><td/><td/><td/><td/><td/><td>1.000</td><td/><td/><td/><td/><td/><td/></tr><tr><td>cogency</td><td>0.93</td><td>1</td><td>0.87</td><td>0.89</td><td>0.950 0.975</td><td>cogency</td><td>0.95</td><td>1</td><td>0.88</td><td>0.91</td><td>0.950 0.975</td><td>cogency</td><td>0.96</td><td>1</td><td>0.93</td><td>0.93</td><td>0.975 0.990</td></tr><tr><td>effectiveness reasonableness</td><td>0.93 0.95</td><td>0.87 0.89</td><td>1 0.88</td><td>0.88 1</td><td>0.875 0.900 0.925</td><td>effectiveness reasonableness</td><td>0.94 0.96</td><td>0.88 0.91</td><td>1 0.9</td><td>0.9 1</td><td>0.900 0.925</td><td>reasonableness effectiveness</td><td>0.96 0.96</td><td>0.93 0.93</td><td>0.93 1</td><td>1 0.93</td><td>0.930 0.945 0.960</td></tr><tr><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/><td/><td>overall</td><td>cogency</td><td colspan=\"2\">effectiveness reasonableness</td><td/></tr><tr><td/><td colspan=\"4\">(d) Crowd on Debates.</td><td/><td/><td colspan=\"4\">(e) Crowd on Reviews.</td><td/><td/><td colspan=\"4\">(f) Crowd on CQA.</td><td/></tr></table>"
},
"TABREF11": {
"num": null,
"html": null,
"text": "Low-scoring arguments from all domains",
"type_str": "table",
"content": "<table/>"
}
}
}
} |