File size: 126,540 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T02:10:12.835499Z"
},
"title": "Exploring Morality in Argumentation",
"authors": [
{
"first": "Jonathan",
"middle": [],
"last": "Kobbe",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Mannheim",
"location": {}
},
"email": ""
},
{
"first": "Ines",
"middle": [],
"last": "Rehbein",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Mannheim",
"location": {}
},
"email": ""
},
{
"first": "Ioana",
"middle": [],
"last": "Hulpus",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Mannheim",
"location": {}
},
"email": ""
},
{
"first": "Heiner",
"middle": [],
"last": "Stuckenschmidt",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Mannheim",
"location": {}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Sentiment and stance are two important concepts for the analysis of arguments. We propose to add another perspective to the analysis, namely moral sentiment. We argue that moral values are crucial for ideological debates and can thus add useful information for argument mining. In the paper, we present different models for automatically predicting moral sentiment in debates and evaluate them on a manually annotated test set. We then apply our models to investigate how moral values in arguments relate to argument quality, stance, and audience reactions.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Sentiment and stance are two important concepts for the analysis of arguments. We propose to add another perspective to the analysis, namely moral sentiment. We argue that moral values are crucial for ideological debates and can thus add useful information for argument mining. In the paper, we present different models for automatically predicting moral sentiment in debates and evaluate them on a manually annotated test set. We then apply our models to investigate how moral values in arguments relate to argument quality, stance, and audience reactions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Argumentation mining is a new research field that is closely related to the subfields of stance detection and sentiment analysis (Stede, 2020) , since \"every argument carries a stance towards its topic, often expressed with sentiment\". 1 In addition to stance and sentiment, there is another dimension that can play an important role in debates, namely moral beliefs. A debater's moral beliefs go beyond stance and can be expressed with varying sentiment. They play an important role in ideological debates and cannot be resolved by simply comparing facts but often involve a battle of ideas and a clash of different belief systems. Consider the following arguments on whether or not gay marriage should be legal. 2 (1) The institution of marriage has traditionally been defined as being between a man and a woman.",
"cite_spans": [
{
"start": 129,
"end": 142,
"text": "(Stede, 2020)",
"ref_id": "BIBREF28"
},
{
"start": 714,
"end": 715,
"text": "2",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "(2) Denying some people the option to marry is discriminatory and creates a second class of citizens.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Both arguments are based on moral belief systems. The first argument refers to moral values that promote respect for tradition, while the second focuses on fairness and equal rights. Arguments that express an opposite stance towards the topic usually differ concerning their moral framing. On the other hand, we observe that arguments expressing a similar stance towards a certain topic may still differ with regard to how the argument is framed, as illustrated in example (3) and (4) below. While (3) opposes the legalization of prostitution because it is considered as a harmful form of oppression targeting women, example (4) depicts prostitution as increasing the danger of diseases and contamination. This makes moral framing an interesting ingredient for argument mining.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "(3) Prostitution and human trafficking are forms of gender-based violence.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "(4) Prostitution is the biggest vector of sexually transmitted diseases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In the paper, we argue that identifying moral values in debates has the potential to support argument analysis and to help with different subtasks related to argument mining. Being able to distinguish between arguments with similar stance and sentiment but framed according to different moral categories can help to identify new arguments and can improve camp detection, thus supporting more fine-grained modeling of debaters beyond stance. Furthermore, moral framing is of particular interest for the analysis of political debates (Lakoff, 1997; Roggeband and Vliegenthart, 2007) .",
"cite_spans": [
{
"start": 532,
"end": 546,
"text": "(Lakoff, 1997;",
"ref_id": "BIBREF18"
},
{
"start": 547,
"end": 580,
"text": "Roggeband and Vliegenthart, 2007)",
"ref_id": "BIBREF25"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "1 Cited from the workshop website (https://argmining2020.i3s.unice.fr/). 2 From https://gaymarriage.procon.org (accessed August 25, 2020).",
"cite_spans": [
{
"start": 73,
"end": 74,
"text": "2",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In practice, however, predicting moral sentiment from text poses several challenges. First, morality is a fuzzy concept, and it is difficult to find an operationalization that turns it into measurable data. Moral sentiment is often expressed implicitly and thus hard to detect, based merely on the presence of lexical cues. In addition, human coders might be biased by their own belief systems, which casts doubt on the validity of the annotations used to train or evaluate automatic systems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In the paper, we present work in progress where we evaluate different models for the prediction of moral framing in text on two datasets and assess the benefits of these predictions for the analysis of arguments. Based on three datasets with argumentative text, we investigate whether we can find correlations between moral values and different aspects of argumentation, such as argument quality, stance, or audience approval. We are interested in the following research questions: RQ1: Do debaters that produce high-quality arguments make more or less frequent use of moral framing? RQ2: Is moral framing more strongly related to a positive or negative stance? RQ3: Can we find a positive correlation between the audiences' approval and the use of moral frames?",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our main contributions are the following: (i) We augment the ArgQuality Corpus of Wachsmuth et al. (2017) with annotations for moral values, as a first test set for the evaluation of moral sentiment in argumentation; (ii) We evaluate two methods for the prediction of moral sentiment on the new dataset; (iii) We present a correlation study investigating the relation between moral framing and argument quality, stance, and audience reactions.",
"cite_spans": [
{
"start": 82,
"end": 105,
"text": "Wachsmuth et al. (2017)",
"ref_id": "BIBREF32"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The paper is structured as follows. We first review work on quantifying moral sentiment in text ( \u00a72). In \u00a73, we describe the annotation of our test set and present different approaches to the automatic detection of moral sentiment in debates. Then we discuss our correlation analysis ( \u00a74), and in \u00a76 we summarise our results and conclude.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "As an operationalisation for the concept of moral sentiment, we refer to Moral Foundations Theory (MFT) (Haidt and Joseph, 2004; Graham et al., 2013) . MFT has its roots in social and cultural psychology and assumes the existence of innate and universally available psychological systems that build the foundations of intuitive ethics. These foundations are augmented by culture-specific constructs of virtues and backed up by personal narratives \"that people construct to make sense of their values and beliefs\" (Graham et al., 2013)[p.17] , and are also reinforced by institutional environments. MFT assumes that all moral issues can be described along the following dimensions: Care-Harm, Fairness-Cheating, Loyalty-Betrayal, Authority-Subversion, and Purity-Degradation. 3",
"cite_spans": [
{
"start": 104,
"end": 128,
"text": "(Haidt and Joseph, 2004;",
"ref_id": "BIBREF14"
},
{
"start": 129,
"end": 149,
"text": "Graham et al., 2013)",
"ref_id": "BIBREF10"
},
{
"start": 513,
"end": 540,
"text": "(Graham et al., 2013)[p.17]",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Dictionary-based approaches to MFT The first version of the Moral Foundations Dictionary (MFD) was presented by Graham et al. (2009) and has been used for a content analysis of christian sermons held in liberal and conservative congregations. Each of the five moral foundations listed above has been further split into a vice and a virtue subcategory, reflecting the positive and negative ends of each dimension. Examples are peace*, protect*, compassion* for Care virtue and suffer*, crush*, killer* for Care vice . The MFD includes, on average, 32 words per moral subcategory. Frimer et al. (2019) presents a new version of the MFD with more entries per MF subcategory, selected according to prototypicality estimates for each MF, based on cosine similarity for Word2Vec embeddings for each item in the dictionary. While the authors admit that the construct validity of the MFD 2.0 is not better than for the original MFD, they recommend the use of the MFD 2.0 due to its improved coverage. Other work on expanding the MFD includes Rezapour et al. (2019) who increase the original size of the dictionary to over 4,600 lexical items, using a quality controlled, human in the loop process. 4 The MF dictionary has been used in several studies in the political and social sciences, psychology, and related fields (Takikawa and Sakamoto, 2017; Matsuo et al., 2018; Lewis, 2019 account for unknown words or the different meanings a word can take, nor do they consider that shifter words and negation can change the polarity of an expression. In addition, we expect that moral vocabulary might vary considerably, depending on the speaker's age and other geopolitical, social, and cultural variables. Garten et al. (2016) address the coverage problem of dictionary-based approaches by replacing the terms in the MF dictionary with their averaged vector representations in distributional space. They show that predicting moral foundations based on the cosine similarity of the words in a text to the distributional representations outperforms a naive method that predicts MF based on word counts.",
"cite_spans": [
{
"start": 112,
"end": 132,
"text": "Graham et al. (2009)",
"ref_id": "BIBREF9"
},
{
"start": 1034,
"end": 1056,
"text": "Rezapour et al. (2019)",
"ref_id": "BIBREF24"
},
{
"start": 1190,
"end": 1191,
"text": "4",
"ref_id": null
},
{
"start": 1312,
"end": 1341,
"text": "(Takikawa and Sakamoto, 2017;",
"ref_id": "BIBREF29"
},
{
"start": 1342,
"end": 1362,
"text": "Matsuo et al., 2018;",
"ref_id": "BIBREF21"
},
{
"start": 1363,
"end": 1374,
"text": "Lewis, 2019",
"ref_id": "BIBREF19"
},
{
"start": 1696,
"end": 1716,
"text": "Garten et al. (2016)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Machine learning-based approaches Recent work has applied the framework of MFT to research questions in the social and political sciences (Fulgoni et al., 2016; Johnson and Goldwasser, 2018; Rezapour et al., 2019; Xie et al., 2019) , replacing dictionary-based counts with more sophisticated methods. Johnson and Goldwasser (2018) model moral framing in politicians' tweets, using probabilistic soft logic (Bach et al., 2013) . Lin et al. (2018) improve the prediction of moral foundations by acquiring additional background knowledge from Wikipedia, using information extraction techniques such as entity linking and cross-document knowledge propagation. Xie et al. (2019) study the change of moral sentiment in longitudinal data, presenting a parameter-free model that predicts moral sentiment on three different levels: (i) moral relevance, (ii) moral polarity, and (iii) the ten moral subclasses of the MFD encoding the virtue/vice dimension for each MF. Finally, Resapour et al. 2019show that using dictionary counts for moral sentiment as features in a supervised classification setup can increase results for stance detection.",
"cite_spans": [
{
"start": 138,
"end": 160,
"text": "(Fulgoni et al., 2016;",
"ref_id": "BIBREF7"
},
{
"start": 161,
"end": 190,
"text": "Johnson and Goldwasser, 2018;",
"ref_id": "BIBREF17"
},
{
"start": 191,
"end": 213,
"text": "Rezapour et al., 2019;",
"ref_id": "BIBREF24"
},
{
"start": 214,
"end": 231,
"text": "Xie et al., 2019)",
"ref_id": "BIBREF34"
},
{
"start": 301,
"end": 330,
"text": "Johnson and Goldwasser (2018)",
"ref_id": "BIBREF17"
},
{
"start": 406,
"end": 425,
"text": "(Bach et al., 2013)",
"ref_id": "BIBREF0"
},
{
"start": 428,
"end": 445,
"text": "Lin et al. (2018)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "3 Predicting Moral Sentiment in Tweets and Debates",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "As a test set for evaluating moral framing in English argumentative text, we use the Dagstuhl ArgQuality Corpus (Wachsmuth et al., 2017) . The dataset contains 320 arguments with approx. 22,600 tokens, covering 16 topics, and is balanced for stance. The data was extracted from two online debate platforms by Habernal and Gurevych (2016) . 5 Each instance has been annotated by three coders, using a fine-grained scheme to assess the arguments' quality. The data also provides a majority score for each dimension of argument quality (Wachsmuth et al., 2017) . The authors report a low agreement for the individual annotations (.51 Krippendorff's \u03b1) but a high majority agreement (94%). We further augment the ArgQuality corpus with annotations for moral foundations, manually coded by two of the authors. 6 We chose not to annotate the 10 subclasses encoded in the dictionary but considered the two ends of each dimension (virtue/vice) as one category. The motivation behind this decision is that both are closely related, and it is often unclear which end of the dimension is addressed, particularly for negated sentences. E.g., \"I could never hurt you\" could either be considered as an instance of Harm as it uses vocabulary related to this dimension or could be annotated as the opposite, Care, as it talks about not being able to harm somebody, thus being more strongly related to the virtue class. Table 1 shows inter-annotator agreement (IAA) scores for individual MFs on the ArgQuality dataset. As expected, IAA is low, being roughly in the same range as agreement scores reported for the annotation of emotions (Schuff et al., 2017; Wood et al., 2018) , thus giving evidence for the subjectivity of the task. Our IAA is not directly comparable to Hoover et al. (2020) as they report Fleiss' \u03ba for the 10 fine-grained subclasses, with an avg. of .315 \u03ba over all 10 classes.",
"cite_spans": [
{
"start": 112,
"end": 136,
"text": "(Wachsmuth et al., 2017)",
"ref_id": "BIBREF32"
},
{
"start": 309,
"end": 337,
"text": "Habernal and Gurevych (2016)",
"ref_id": "BIBREF13"
},
{
"start": 340,
"end": 341,
"text": "5",
"ref_id": null
},
{
"start": 533,
"end": 557,
"text": "(Wachsmuth et al., 2017)",
"ref_id": "BIBREF32"
},
{
"start": 805,
"end": 806,
"text": "6",
"ref_id": null
},
{
"start": 1619,
"end": 1640,
"text": "(Schuff et al., 2017;",
"ref_id": "BIBREF26"
},
{
"start": 1641,
"end": 1659,
"text": "Wood et al., 2018)",
"ref_id": "BIBREF33"
},
{
"start": 1755,
"end": 1775,
"text": "Hoover et al. (2020)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 1403,
"end": 1410,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "A New Test Set for Moral Framing in Argumentation",
"sec_num": "3.1"
},
{
"text": "We model moral sentiment prediction as a text classification task and propose two distinct methods for feature generation. The first method is based on a sense-disambiguated version of the MFD and extends its coverage by exploiting relations in Wordnet (WN) (Fellbaum, 2010) . The second method uses the MFD as seed data to learn BERT sequence embeddings that encode moral sentiment. The representations created by each method are fixed-sized vectors that can easily be combined by concatenation.",
"cite_spans": [
{
"start": 258,
"end": 274,
"text": "(Fellbaum, 2010)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "I. Sense-disambiguated features (WN-PPR) The MFD has two main disadvantages that we try to overcome with this method. First, the lexicon contains many words with different word senses, where the moral value only applies to one specific sense. Thus, we link the dictionary entries to their corresponding WN synsets. This way, fair is only considered to be related to the MF fairness-cheating if used as synonym to just or honest, but not if used as synonym to carnival, funfair or attractively feminine. Also, this way we overcome the problems resulting from the use of regular expressions in the MFD (e.g., defenestration would belong to the MF Care because of the entry defen*, and Churchill would trigger Purity because of the entry church*). The second disadvantage of the MFD is its low coverage, which we extend by running Personalized Pagerank (Haveliwala, 2003) on the set of WN synsets that have been linked to dictionary entries. a) Linking MFD entries to WN To create a word sense disambiguated version of the MFD, one expert annotator was presented with the following information: a specific moral foundation; a WN synset whereof at least one word in the synset is part of the respective MF in the MFD; and its definition. With this information, the annotator decided whether the synset is relevant for the moral foundation in question or not. Overall, the resulting lexicon contains, on average, 61 synsets per MF. b) Extending the disambiguated Lexicon We extend the disambiguated lexicon by exploiting relations between synsets in WN, such as hypernym or similar to. Concretely, we run personalized pagerank on the graph consisting of the WN synsets and the relations between them for every MF, using the corresponding lexicon entries as seed nodes. This way, each WN synset is assigned a fixed-sized vector containing scores for each MF, including the category GeneralMorality. 7 We expect that higher scores reflect a stronger correspondence between the synset and the respective MF. c) Extracting features from text Given a short English text, we first extract WN 3.0 synsets using the disambiguation method by Tan (2014). Then we link these synsets to WN 3.1, using the official Wordnet Search Engine 8 and, if necessary, resolving the final mapping manually. For instance, a variety of offensive terms have been removed in WN 3.1, and thus, we had to link terms like darky or tom to black (noun.person) manually. 9 As each of the synsets is assigned a fixed-size score vector in our lexicon, any function to aggregate these vectors is conceivable. To obtain vectors that do not depend on the input text's length, we decided to take their mean. The result is a vector consisting of 6 entries, where each entry represents a MF, including GeneralMorality.",
"cite_spans": [
{
"start": 850,
"end": 868,
"text": "(Haveliwala, 2003)",
"ref_id": "BIBREF15"
},
{
"start": 1893,
"end": 1894,
"text": "7",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "II. Contextualized MF sequence embeddings (SBERT-Wiki) Our second method uses Sentence-BERT (SBERT) (Reimers and Gurevych, 2019) to obtain text representations that encode moral sentiment. SBERT is a modification of the original transformer model, based on siamese networks. The modified SBERT encodes sentence similarity in a human interpretable way where similar sentence pairs can be retrieved efficiently, based on cosine similarity. While previous work has computed BERT-based embeddings for text sequences (i) by averaging (or summing) over all word embeddings for this particular text or (ii) by using the network output at the position of the [CLS] token, Reimers and Gurevych (2019) show that those representations are not well suited to encode sentence similarity and often yield inferior results as compared to using averaged Glove embeddings (Pennington et al., 2014) .",
"cite_spans": [
{
"start": 100,
"end": 128,
"text": "(Reimers and Gurevych, 2019)",
"ref_id": "BIBREF23"
},
{
"start": 664,
"end": 691,
"text": "Reimers and Gurevych (2019)",
"ref_id": "BIBREF23"
},
{
"start": 854,
"end": 879,
"text": "(Pennington et al., 2014)",
"ref_id": "BIBREF22"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "SBERT can also be applied to tasks where an anchor text is compared to a positive and a negative text sample, thus learning to maximize a score based on the similarity of the anchor text to the first sentence (the positive sample) and its distance to the second text (the negative sample). For that, SBERT uses the triplet objective function (equation 1) where d is a distance metric (here: Euclidean distance), and the margin is set to 1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "L = max(d(s anchor \u2212 s pos ) \u2212 d(s anchor \u2212 s neg ) + margin, 0)",
"eq_num": "(1)"
}
],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "We fine-tune SBERT embeddings so that they encode different moral foundations. First, we download all short Wikipedia abstracts from DBpedia 10 and label them with their corresponding MF (if any), using weak supervision. Our approach is based on the MFD and proceeds as follows: For each dictionary entry, we search in Wikipedia for corresponding articles to get a set of candidates consisting of articles whose title is a lexicon entry (including redirections) and articles that are linked by the lexical entry's disambiguation page. From these candidates, we manually select the ones related to the MF and label their abstracts accordingly.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "This approach yields 317 short abstracts from Wikipedia, labeled with moral foundations, extracted from a pool of 4,935,596 unlabelled short Wikipedia abstracts. We iterate over each abstract in the annotated dataset, considering the abstract as the anchor text. First, we retrieve all other abstracts labeled with the same moral foundation as the anchor and create pairs of (anchor, positive sample). Then, for each pair, we randomly select 3 labeled abstracts that belong to a different moral foundation as well as 7 abstracts from the unlabelled pool as negative samples, assuming that the unlabelled abstracts also do, more often than not, either belong to a different moral foundation or do not express any moral content. This gives us a total of 10 negative samples for each pair and results in a weakly supervised dataset with 107,940 instances. We then fine-tune the model on the data, using the same settings as reported in Reimers and Gurevych (2019) . After the training is completed, we use the learned model to retrieve representations for new text sequences from different argumentation datasets, expecting that the finetuned embeddings will now capture some aspects of moral sentiment. We compare our approach with the pretrained SBERT embeddings (bert-base-nli-stsb-mean-tokens) of Reimers and Gurevych (2019) , trained without the fine-tuning step on the weakly supervised Wikipedia abstracts.",
"cite_spans": [
{
"start": 933,
"end": 960,
"text": "Reimers and Gurevych (2019)",
"ref_id": "BIBREF23"
},
{
"start": 1298,
"end": 1325,
"text": "Reimers and Gurevych (2019)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods for the Prediction of Moral Sentiment",
"sec_num": "3.2"
},
{
"text": "multi-label BERT To compare our lexicon-based methods with a state-of-the-art approach to text classification, we train a multi-label text classifier based on BERT. We use a publicly available implementation in pytorch 11 that replaces the cross-entropy loss with a binary cross-entropy with logits to adapt the BERT sequence classifier to the multi-label setup.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "State of the Art and Baselines",
"sec_num": "3.3"
},
{
"text": "The model includes an input embedding layer for the pretrained BERT embeddings, the BERT encoder with 12 attention layers, and, as final layer, a linear transformation, with one dimension for each class. This gives us six output dimensions: the five moral foundations + one class for tweets with non-moral content. Our model uses the pretrained English uncased BERT base embeddings with a vocabulary size of around 30,000. We use the same data splits and preprocessing in all experiments (for details, see \u00a73.4). In contrast to our other models, however, BERT further segments the input text into subword tokens (WordPiece tokenization), which might increase coverage for words not seen in the training data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "State of the Art and Baselines",
"sec_num": "3.3"
},
{
"text": "The Random baseline assigns labels randomly but respecting the class distribution in the training data. Results are averaged over 100 trials.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Random baseline",
"sec_num": null
},
{
"text": "MFD baseline Given a text, we compute frequency counts for each MF, based on the entries in the MFD, and normalize by text length. We use these count-based vectors as features for the text classifier. Similar to WN-PPR, we derive one feature per MF, including general morality. Table 3 : Binary F1-scores on the Dagstuhl ArgQuality Corpus for individual MFs (F1 for the pos. class).",
"cite_spans": [],
"ref_spans": [
{
"start": 278,
"end": 285,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Random baseline",
"sec_num": null
},
{
"text": "We now present the data used for the evaluation of the methods described above ( \u00a73.2) for the prediction of moral sentiment in tweets and debates. As training data for our MF classifiers, we use the Moral Foundations Twitter Corpus (Hoover et al., 2020) , a collection of approximately 35,000 tweets covering seven controversial topical threads: All Lives Matter, Black Lives Matter, the Baltimore protests, the 2016 Presidential election, hate speech & offensive language (Davidson et al., 2017) , Hurricane Sandy, and #MeToo. Each tweet has been annotated with MF by at least three trained annotators. The authors report Fleiss' \u03ba and PABAK, a measure adjusted for prevalence and bias (Sim and Wright, 2005) . IAA is relatively low (with a Fleiss \u03ba in the range of 0.24 -0.46 and PABAK ranging from 0.65 -0.85) and shows considerable variation across the different moral domains and threads. We follow the procedure described in Hoover et al. (2020) to create a gold standard from the annotated tweets and consider a label as gold if it was assigned by at least half of the annotators. Thus, our gold standard includes 6 labels: one for each MF and a sixth one for GeneralMorality. Note that in the MFTC, this label is called Non-moral while we report results for its inverse, which we call Moral. We normalized the tweets using the script available from the Glove website. 12 We noticed that the dataset includes many near-duplicates (e.g., 96 instances of homosexuality is a sin). To ensure that these near-duplicates do not appear in both training and test set, we split the data into the different threads and present results for a seven-fold cross-validation where we train the models on six threads and evaluate on the remaining one. We also evaluate the models trained on the MFTC on out-of-domain data from the ArgQuality Corpus, where we consider all labels assigned by each of the two annotators as ground truth. 13",
"cite_spans": [
{
"start": 233,
"end": 254,
"text": "(Hoover et al., 2020)",
"ref_id": null
},
{
"start": 474,
"end": 497,
"text": "(Davidson et al., 2017)",
"ref_id": "BIBREF1"
},
{
"start": 688,
"end": 710,
"text": "(Sim and Wright, 2005)",
"ref_id": "BIBREF27"
},
{
"start": 932,
"end": 952,
"text": "Hoover et al. (2020)",
"ref_id": null
},
{
"start": 1377,
"end": 1379,
"text": "12",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "3.4"
},
{
"text": "We conduct experiments on the Twitter corpus, testing different traditional classification methods, and report results for the best performing classifiers only. For WN-PPR and MFD-Features, this was a knearest-neighbors classifier, and for SBERT-Base, SBERT-Wiki, as well as for WN-PPR + SBERT-Wiki a Linear Discriminant Analysis. 14 All other results, as well as the correlations reported in \u00a74, refer to these classification methods. Table 2 shows results on the MFTC for our different methods. Not surprisingly, multi-label BERT outperforms all other methods on the Twitter data. However, our lexicon-based methods outperform the random baseline for each category, with the best results obtained by the concatenation of SBERT-Wiki with WN-PPR. WN-PPR on its own only yields poor results, barely outperforming the constant and the MFD baseline.",
"cite_spans": [],
"ref_spans": [
{
"start": 436,
"end": 443,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Results for MF Prediction on Tweets and Debates",
"sec_num": "3.5"
},
{
"text": "When applying the classifiers to the out-of-domain data from the ArgQuality corpus (Table 3) , multilabel-BERT still yields best results, but now SBERT-Wiki outperforms BERT on the Authority, Purity and Moral categories. The lower performance for the Moral class can be explained by the differences in class distribution between the two datasets. In the MFTC, this class makes up for approximately 57% of the training instances, while the amount of moral instances in the debate corpus is much higher (79%). The lexicon-based methods are not sensitive to the class distribution in the training data, which, in this case, makes them more robust. Still, all systems fail to beat the majority baseline for the Moral class which has a binary F1-score of 0.881. 15 WN-PPR again performs poorly with results below the Random baseline, and results for the MFD baseline also fail to outperform Random. This time, results for the concatenation of WN-PPR and BERT-Wiki are considerably worse than for BERT-Wiki alone.",
"cite_spans": [],
"ref_spans": [
{
"start": 83,
"end": 92,
"text": "(Table 3)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results for MF Prediction on Tweets and Debates",
"sec_num": "3.5"
},
{
"text": "To study the impact of moral framing in argumentation, we investigate the correlation between moral sentiment and other properties of argumentative text, namely argument quality, stance, and audience reactions. For this, we use the multilabel-BERT model that yielded the best results on both datasets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Correlation Studies",
"sec_num": "4"
},
{
"text": "The Dagstuhl ArgQuality Corpus contains arguments that are annotated with different dimensions of argument quality, such as cogency and credibility, as well as a score for overall quality. Some of the dimensions are also interesting for contexts other than argument quality, such as clarity and emotional appeal.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "4.1"
},
{
"text": "The IBM Argument Quality Ranking Corpus (Gretz et al., 2019) is used to triangulate our findings on the Dagstuhl ArgQuality Corpus and to investigate the correlation between moral sentiment and an argument's stance. The corpus contains more than 30,000 arguments on 71 topics, labelled for quality (good or bad) and stance (pro or con) by crowd annotators. To obtain ranks for argument quality, the authors apply two different strategies, which both give more weight to the answers of reliable annotators.",
"cite_spans": [
{
"start": 40,
"end": 60,
"text": "(Gretz et al., 2019)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "4.1"
},
{
"text": "We use CORPS (Guerini et al., 2013) to investigate whether moral sentiment in political speeches has an impact on the audience. CORPS includes >3,600 political speeches held by more than 203 different speakers, tagged for audience reactions such as applause, laughter or booing. The motivation for creating the corpus was that such tags might highlight passages in the speech where an attempt has been made by the speaker to persuade the audience, either successful or not. We expect to find a correlation between text passages that triggered a positive audience reaction (i.e. applause) and moral framing, but not for laughter (we focus on these two tags as the other tags are relatively rare in comparison 16 ). We also exclude mixed tags that mark two different reactions for the same text passage (laughter; applause). To test our hypothesis, we predict moral sentiment for the speech passages directly before an audience reaction was triggered. We consider up to 360 tokens of speech context and omit all speech passages where another tag occurs within this context. Table 4 shows Spearman correlations between argument quality, stance, and audience reactions and a) human annotations (HU) and b) labels predicted by multi-label BERT (BM). We observe a weak positive correlation between argument quality and moral sentiment for the two most frequent categories (Moral, Care) on the ArgQuality data. For the other MFs, there are no significant effects. On the IBM-AQR Corpus, we see a consistent and significant positive correlation for Care and Fairness. However, the effect is very weak. For the subdimensions of argument quality, the correlations tend to be similar to the ones for overall quality and are highest for emotional appeal, which seems plausible. Concerning argument stance, we again find slightly positive correlations for Care and Moral. Results on the CORPS Table 4 : Spearman \u03c1 between human annotations (HU) and multi-label BERT predictions (BM), respectively, and quality, stance and audience reactions. Bold values are statistically significant (p < 0.05). + / \u2212 : The correlation to the SBERT-Wiki predictions was considerably higher / lower (by at least 0.05).",
"cite_spans": [
{
"start": 13,
"end": 35,
"text": "(Guerini et al., 2013)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [
{
"start": 1072,
"end": 1079,
"text": "Table 4",
"ref_id": null
},
{
"start": 1880,
"end": 1887,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Data",
"sec_num": "4.1"
},
{
"text": "data are as expected: a positive correlation for applause and a negative one for laughter, but again the effect is very weak.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results for the Correlation Analysis",
"sec_num": "4.2"
},
{
"text": "Correlation with topic To control for topic effects, we computed the correlation between topic and argument quality and between topic and MF in the IBM-AQR. While we found no correlation between topic and argument quality, there was a weak correlation between some topics and specific MFs (see Table 5 ). Table 5 : Topics with correlations to MFs greater than .2 according to multilabel-BERT predictions.",
"cite_spans": [],
"ref_spans": [
{
"start": 294,
"end": 301,
"text": "Table 5",
"ref_id": null
},
{
"start": 305,
"end": 312,
"text": "Table 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results for the Correlation Analysis",
"sec_num": "4.2"
},
{
"text": "A crucial issue for using moral values for argumentation analysis concerns the reliability of the (manual and automatic) annotations. Before we can reliably use moral values for the analysis of arguments, we need to ensure the quality of the annotations, as the low inter-annotator agreement for MF annotation casts doubt on the validity of the findings. While we expect that more extensive training and more detailed guidelines will increase IAA for human annotation at least slightly, we still think that due to the fuzziness of the concept of morality, high agreement scores are not very probable. Thus, we would like to propose a different approach to the annotation of moral foundations where we ground the annotations in lexical semantics. This approach has already been shown to improve IAA for a similarly difficult annotation task, namely the annotation of causal language (Dunietz et al., 2015) . The authors created a lexical resource for terms that can trigger causality in text and instructed annotators to disambiguate instances of those terms in context, showing that their modularized, dictionary-based approach yields substantially increased IAA scores.",
"cite_spans": [
{
"start": 882,
"end": 904,
"text": "(Dunietz et al., 2015)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "5"
},
{
"text": "Inspired by their work, we propose to anchor MF annotations in lexical semantics, using an expanded version of the MFD as seed terms. The annotators will then be presented with instances of these terms and instructed to disambiguate them in context, and also to annotate a small, predefined set of semantic roles, such as Betrayer, Harm doer, Victim, and so on. Setting up the annotation of moral values as a frame semantic annotation task has several advantages. First, the addition of semantic roles would make the annotations more informative by encoding the core participants of moral arguments, i.e., the target of the moral evaluation and the affected party. Second, anchoring the annotations in lexical semantics would make it easier to provide the annotators with precise guidelines. This might not only increase the consistency of the annotations but might also help to control for annotator bias. An open question, however, concerns coverage as it is not yet clear whether this approach would miss too many relevant expressions of moral values, given that it only captures explicitly stated moral evaluation but not implicit judgments. Whether the merits of our proposal outweigh its drawbacks needs to be explored in future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "5"
},
{
"text": "Being able to predict moral values in text reliably can open up new research avenues in argumentation. E.g., recent work in psychology has shown that moral values play an important role in debates on political and social issues (Feinberg and Willer, 2013; Voelkel and Feinberg, 2018; Feinberg and Willer, 2019) . For example, Feinberg and Willer (2013) have shown that debates on environmental issues are often framed in terms of moral values such as Care-Harm, a moral foundation that is at the core of liberal belief systems, while conservatives, in contrast, seem to value all five MFs more similarly (Graham et al., 2009) . This often results in highly polarized discussions, and Feinberg and Willer (2013) argue that reframing such issues in terms of moral values that explicitly address the opponents' belief system might have the potential to depolarize controversial debates and improve understanding between the camps by addressing the \"moral empathy gap\" (see Feinberg and Willer (2019) and references therein).",
"cite_spans": [
{
"start": 228,
"end": 255,
"text": "(Feinberg and Willer, 2013;",
"ref_id": "BIBREF3"
},
{
"start": 256,
"end": 283,
"text": "Voelkel and Feinberg, 2018;",
"ref_id": "BIBREF31"
},
{
"start": 284,
"end": 310,
"text": "Feinberg and Willer, 2019)",
"ref_id": "BIBREF4"
},
{
"start": 326,
"end": 352,
"text": "Feinberg and Willer (2013)",
"ref_id": "BIBREF3"
},
{
"start": 604,
"end": 625,
"text": "(Graham et al., 2009)",
"ref_id": "BIBREF9"
},
{
"start": 684,
"end": 710,
"text": "Feinberg and Willer (2013)",
"ref_id": "BIBREF3"
},
{
"start": 970,
"end": 996,
"text": "Feinberg and Willer (2019)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "5"
},
{
"text": "In the paper, we evaluated different models for predicting moral sentiment in debates, based on Moral Foundations Theory. We then used our models to predict moral values in three argumentation datasets. We investigated whether we could find a correlation between morality and (i) argument quality, (ii) stance, and (iii) audience reactions for political speeches.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Future Work",
"sec_num": "6"
},
{
"text": "We found weak but significant correlations between general morality and argument quality in the ArgQuality data and a consistent positive correlation between moral sentiment and audience approval in CORPS as well as a negative correlation for moral sentiment and laughter. However, our study has several limitations that need to be addressed. First, the annotation experiment has been conducted by two annotators only, not allowing us to retrieve more reliable labels using the wisdom of the crowd. Also, the size of the test set is rather small, thus questioning the reliability of the results. Another problem is the low accuracy of the classifiers for the prediction of moral values. While results were substantially higher than the random baseline and an MFD-based baseline, we still expect a considerable amount of noise in the classifiers' predictions, which might impact the results of the correlation study. It is conceivable that cleaner predictions might increase the effect size of the observed correlations, which would be consistent with the slightly larger correlation coefficients found for the human annotations (HU). This, however, still needs to be confirmed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Future Work",
"sec_num": "6"
},
{
"text": "The next steps should include the creation of larger test sets where the annotations have been validated by more than two annotators as well as the evaluation of semantically grounded approaches to coding moral values, to assess their reliability and validity. Another important task is the development of more accurate and robust classifiers for the prediction of moral sentiment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Future Work",
"sec_num": "6"
},
{
"text": "Another foundation currently under discussion is Liberty-Opression.4 Arguably, a comparison of size is not meaningful, given that the original MFD includes regular expressions that can match an unknown number of instance types while the expanded lexicon includes word forms for unigrams and word compounds.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "www.createdebate.com and convinceme.net. 6 All resources created for this paper are available at https://github.com/dwslab/Morality-in-Arguments",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "GeneralMorality includes terms related to moral concepts that do not fit into one of the five MFs (e.g. ethic, good, evil). 8 http://wordnet-rdf.princeton.edu/json/pwn30/...9 We are aware that this treatment is not optimal. A better solution would link those terms to a synset that captures their offensive usage, similar to the one for Kraut: offensive term for a person of German descent.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://wiki.dbpedia.org/downloads-2016-10 11 The code was adapted from https://medium.com/huggingface/multi-label-text-classificationusing-bert-the-mighty-transformer-69714fa3fb3d and is based on the HuggingFace library (https://github.com/huggingface/pytorch-pretrained-BERT).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://nlp.stanford.edu/projects/glove/preprocess-twitter.rb 13 As the data has been annotated by two of the authors, we can be sure that we do not have to eliminate spammers.14 We use the scikit-learn implementation for these methods. Other methods we tried include Logistic Regression, Decision Trees, Na\u00efve Bayes, Support Vector Machines.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The majority baseline is not included in tables 2 and 3 because its binary F1-score is zero for all classes except Moral.16 Applause: 23,095; Laughter: 5,857; Booing: 532; Cheers: 80; Sustained applause: 61; Spontaneous demonstration: 16.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work has been funded by the Deutsche Forschungsgemeinschaft (DFG) within the project Ex-pLAIN, Grant Number STU 266/14-1, as part of the Priority Program \"Robust Argumentation Machines (RATIO)\" (SPP-1999), as well as within the SFB 884 on the Political Economy of Reforms at the University of Mannheim (projects B6 and C4).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Hinge-loss Markov random fields: Convex inference for structured prediction",
"authors": [
{
"first": "H",
"middle": [],
"last": "Stephen",
"suffix": ""
},
{
"first": "Bert",
"middle": [],
"last": "Bach",
"suffix": ""
},
{
"first": "Ben",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Lise",
"middle": [],
"last": "London",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Getoor",
"suffix": ""
}
],
"year": 2013,
"venue": "Conference on Uncertainty in Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stephen H. Bach, Bert Huang, Ben London, and Lise Getoor. 2013. Hinge-loss Markov random fields: Convex inference for structured prediction. In Conference on Uncertainty in Artificial Intelligence.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Automated hate speech detection and the problem of offensive language",
"authors": [
{
"first": "Thomas",
"middle": [],
"last": "Davidson",
"suffix": ""
},
{
"first": "Dana",
"middle": [],
"last": "Warmsley",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Macy",
"suffix": ""
},
{
"first": "Ingmar",
"middle": [],
"last": "Weber",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1703.04009"
]
},
"num": null,
"urls": [],
"raw_text": "Thomas Davidson, Dana Warmsley, Michael Macy, and Ingmar Weber. 2017. Automated hate speech detection and the problem of offensive language. arXiv preprint arXiv:1703.04009.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Annotating causal language using corpus lexicography of constructions",
"authors": [
{
"first": "Jesse",
"middle": [],
"last": "Dunietz",
"suffix": ""
},
{
"first": "Lori",
"middle": [],
"last": "Levin",
"suffix": ""
},
{
"first": "Jaime",
"middle": [],
"last": "Carbonell",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of The 9th Linguistic Annotation Workshop",
"volume": "",
"issue": "",
"pages": "188--196",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jesse Dunietz, Lori Levin, and Jaime Carbonell. 2015. Annotating causal language using corpus lexicography of constructions. In Proceedings of The 9th Linguistic Annotation Workshop, pages 188-196, Denver, Colorado, USA, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "The moral roots of environmental attitudes",
"authors": [
{
"first": "Matthew",
"middle": [],
"last": "Feinberg",
"suffix": ""
},
{
"first": "Robb",
"middle": [],
"last": "Willer",
"suffix": ""
}
],
"year": 2013,
"venue": "Psychological Science",
"volume": "24",
"issue": "1",
"pages": "56--62",
"other_ids": {
"PMID": [
"23228937"
]
},
"num": null,
"urls": [],
"raw_text": "Matthew Feinberg and Robb Willer. 2013. The moral roots of environmental attitudes. Psychological Science, 24(1):56-62. PMID: 23228937.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Moral reframing: A technique for effective and persuasive communication across political divides",
"authors": [
{
"first": "Matthew",
"middle": [],
"last": "Feinberg",
"suffix": ""
},
{
"first": "Robb",
"middle": [],
"last": "Willer",
"suffix": ""
}
],
"year": 2019,
"venue": "Social Psychology and Personality Compass",
"volume": "",
"issue": "",
"pages": "56--62",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew Feinberg and Robb Willer. 2019. Moral reframing: A technique for effective and persuasive communi- cation across political divides. Social Psychology and Personality Compass, pages 56-62.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Princeton university: About wordnet",
"authors": [
{
"first": "Christiane",
"middle": [],
"last": "Fellbaum",
"suffix": ""
}
],
"year": 2010,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christiane Fellbaum. 2010. Princeton university: About wordnet.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Moral Foundations Dictionary for Linguistic Analyses 2.0. Unpublished manuscript",
"authors": [
{
"first": "Jeremy",
"middle": [
"A"
],
"last": "Frimer",
"suffix": ""
},
{
"first": "Reihane",
"middle": [],
"last": "Boghrati",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Haidt",
"suffix": ""
},
{
"first": "Jesse",
"middle": [],
"last": "Graham",
"suffix": ""
},
{
"first": "Morteza",
"middle": [],
"last": "Dehgani",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jeremy A. Frimer, Reihane Boghrati, Jonathan Haidt, Jesse Graham, and Morteza Dehgani. 2019. Moral Foundations Dictionary for Linguistic Analyses 2.0. Unpublished manuscript. Available from https: //osf.io/xakyw/.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "An Empirical Exploration of Moral Foundations Theory in Partisan News Sources",
"authors": [
{
"first": "Dean",
"middle": [],
"last": "Fulgoni",
"suffix": ""
},
{
"first": "Jordan",
"middle": [],
"last": "Carpenter",
"suffix": ""
},
{
"first": "Lyle",
"middle": [],
"last": "Ungar",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Preo\u0163iuc-Pietro",
"suffix": ""
},
{
"first": ";",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "Khalid",
"middle": [],
"last": "Choukri",
"suffix": ""
},
{
"first": "Thierry",
"middle": [],
"last": "Declerck",
"suffix": ""
},
{
"first": "Sara",
"middle": [],
"last": "Goggi",
"suffix": ""
},
{
"first": "Marko",
"middle": [],
"last": "Grobelnik",
"suffix": ""
},
{
"first": "Bente",
"middle": [],
"last": "Maegaard",
"suffix": ""
},
{
"first": "Joseph",
"middle": [],
"last": "Mariani",
"suffix": ""
}
],
"year": 2016,
"venue": "The 10th International Conference on Language Resources and Evaluation, LREC'16",
"volume": "",
"issue": "",
"pages": "3730--3736",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dean Fulgoni, Jordan Carpenter, Lyle Ungar, and Daniel Preo\u0163iuc-Pietro. 2016. An Empirical Exploration of Moral Foundations Theory in Partisan News Sources. In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Thierry Declerck, Sara Goggi, Marko Grobelnik, Bente Maegaard, Joseph Mariani, Helene Mazo, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, editors, The 10th International Conference on Language Resources and Evaluation, LREC'16, pages 3730-3736, Paris, France, may. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Morality between the lines: Detecting moral sentiment in text",
"authors": [
{
"first": "Justin",
"middle": [],
"last": "Garten",
"suffix": ""
},
{
"first": "Reihane",
"middle": [],
"last": "Boghrati",
"suffix": ""
},
{
"first": "Joe",
"middle": [],
"last": "Hoover",
"suffix": ""
},
{
"first": "Kate",
"middle": [
"M"
],
"last": "Johnson",
"suffix": ""
},
{
"first": "Morteza",
"middle": [],
"last": "Dehghani",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of IJCAI 2016 Workshop on Computational Modeling of Attitudes",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Justin Garten, Reihane Boghrati, Joe Hoover, Kate M. Johnson, and Morteza Dehghani. 2016. Morality be- tween the lines: Detecting moral sentiment in text. In Proceedings of IJCAI 2016 Workshop on Computational Modeling of Attitudes.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Liberals and conservatives rely on different sets of moral foundations",
"authors": [
{
"first": "Joseph",
"middle": [],
"last": "Graham",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Haidt",
"suffix": ""
},
{
"first": "B",
"middle": [
"A"
],
"last": "Nosek",
"suffix": ""
}
],
"year": 2009,
"venue": "Journal of personality and social psychology",
"volume": "96",
"issue": "5",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joseph Graham, Jonathan Haidt, and B. A. Nosek. 2009. Liberals and conservatives rely on different sets of moral foundations. Journal of personality and social psychology, 96(5:1029).",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Chapter Two -Moral Foundations Theory: The Pragmatic Validity of Moral Pluralism",
"authors": [
{
"first": "Jesse",
"middle": [],
"last": "Graham",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Haidt",
"suffix": ""
},
{
"first": "Sena",
"middle": [],
"last": "Koleva",
"suffix": ""
},
{
"first": "Matt",
"middle": [],
"last": "Motyl",
"suffix": ""
},
{
"first": "Ravi",
"middle": [],
"last": "Iyer",
"suffix": ""
},
{
"first": "Sean",
"middle": [
"P"
],
"last": "Wojcik",
"suffix": ""
},
{
"first": "Peter",
"middle": [
"H"
],
"last": "Ditto",
"suffix": ""
}
],
"year": 2013,
"venue": "Advances in Experimental Social Psychology",
"volume": "47",
"issue": "",
"pages": "55--130",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jesse Graham, Jonathan Haidt, Sena Koleva, Matt Motyl, Ravi Iyer, Sean P. Wojcik, and Peter H. Ditto. 2013. Chapter Two -Moral Foundations Theory: The Pragmatic Validity of Moral Pluralism. Advances in Experi- mental Social Psychology, 47:55 -130.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Ranit Aharonov, and Noam Slonim. 2019. 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": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1911.11408"
]
},
"num": null,
"urls": [],
"raw_text": "Shai Gretz, Roni Friedman, Edo Cohen-Karlik, Assaf Toledo, Dan Lahav, Ranit Aharonov, and Noam Slonim. 2019. A large-scale dataset for argument quality ranking: Construction and analysis. arXiv preprint arXiv:1911.11408.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "The New Release of CORPS: A Corpus of Political Speeches Annotated with Audience Reactions",
"authors": [
{
"first": "Marco",
"middle": [],
"last": "Guerini",
"suffix": ""
},
{
"first": "Danilo",
"middle": [],
"last": "Giampiccolo",
"suffix": ""
},
{
"first": "Giovanni",
"middle": [],
"last": "Moretti",
"suffix": ""
},
{
"first": "Rachele",
"middle": [],
"last": "Sprugnoli",
"suffix": ""
},
{
"first": "Carlo",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2013,
"venue": "Multimodal Communication in Political Speech. Shaping Minds and Social Action",
"volume": "",
"issue": "",
"pages": "86--98",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marco Guerini, Danilo Giampiccolo, Giovanni Moretti, Rachele Sprugnoli, and Carlo Strapparava. 2013. The New Release of CORPS: A Corpus of Political Speeches Annotated with Audience Reactions. In Isabella Poggi, Francesca D'Errico, Laura Vincze, and Alessandro Vinciarelli, editors, Multimodal Communication in Political Speech. Shaping Minds and Social Action, pages 86-98, Berlin, Heidelberg. Springer Berlin Heidelberg.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "What makes a convincing argument? empirical analysis and detecting attributes of convincingness in Web argumentation",
"authors": [
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1214--1223",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ivan Habernal and Iryna Gurevych. 2016. What makes a convincing argument? empirical analysis and detect- ing attributes of convincingness in Web argumentation. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1214-1223, Austin, Texas, November. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Intuitive ethics: How innately prepared intuitions generate culturally variable virtues",
"authors": [
{
"first": "Jonathan",
"middle": [],
"last": "Haidt",
"suffix": ""
},
{
"first": "Craig",
"middle": [],
"last": "Joseph",
"suffix": ""
}
],
"year": 2004,
"venue": "Daedalus",
"volume": "133",
"issue": "4",
"pages": "55--66",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jonathan Haidt and Craig Joseph. 2004. Intuitive ethics: How innately prepared intuitions generate culturally variable virtues. Daedalus, 133(4):55-66.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Topic-sensitive pagerank: A context-sensitive ranking algorithm for web search",
"authors": [
{
"first": "H",
"middle": [],
"last": "Taher",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Haveliwala",
"suffix": ""
}
],
"year": 2003,
"venue": "IEEE transactions on knowledge and data engineering",
"volume": "15",
"issue": "4",
"pages": "784--796",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Taher H. Haveliwala. 2003. Topic-sensitive pagerank: A context-sensitive ranking algorithm for web search. IEEE transactions on knowledge and data engineering, 15(4):784-796.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Jun Yen Leung, Arineh Mirinjian, and Morteza Dehghani. 2020. Moral Foundations Twitter Corpus: A Collection of 35k Tweets Annotated for Moral Sentiment",
"authors": [
{
"first": "Joe",
"middle": [],
"last": "Hoover",
"suffix": ""
},
{
"first": "Gwenyth",
"middle": [],
"last": "Portillo-Wightman",
"suffix": ""
},
{
"first": "Leigh",
"middle": [],
"last": "Yeh",
"suffix": ""
},
{
"first": "Shreya",
"middle": [],
"last": "Havaldar",
"suffix": ""
},
{
"first": "Aida",
"middle": [
"Mostafazadeh"
],
"last": "Davani",
"suffix": ""
},
{
"first": "Ying",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Brendan",
"middle": [],
"last": "Kennedy",
"suffix": ""
},
{
"first": "Mohammad",
"middle": [],
"last": "Atari",
"suffix": ""
},
{
"first": "Zahra",
"middle": [],
"last": "Kamel",
"suffix": ""
},
{
"first": "Madelyn",
"middle": [],
"last": "Mendlen",
"suffix": ""
},
{
"first": "Gabriela",
"middle": [],
"last": "Moreno",
"suffix": ""
},
{
"first": "Christina",
"middle": [],
"last": "Park",
"suffix": ""
},
{
"first": "Tingyee",
"middle": [
"E"
],
"last": "Chang",
"suffix": ""
},
{
"first": "Jenna",
"middle": [],
"last": "Chin",
"suffix": ""
},
{
"first": "Christian",
"middle": [],
"last": "Leong",
"suffix": ""
}
],
"year": null,
"venue": "Social Psychological and Personality Science",
"volume": "0",
"issue": "0",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joe Hoover, Gwenyth Portillo-Wightman, Leigh Yeh, Shreya Havaldar, Aida Mostafazadeh Davani, Ying Lin, Brendan Kennedy, Mohammad Atari, Zahra Kamel, Madelyn Mendlen, Gabriela Moreno, Christina Park, Tingyee E. Chang, Jenna Chin, Christian Leong, Jun Yen Leung, Arineh Mirinjian, and Morteza Dehghani. 2020. Moral Foundations Twitter Corpus: A Collection of 35k Tweets Annotated for Moral Sentiment. Social Psychological and Personality Science, 0(0):0.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Classification of Moral Foundations in Microblog Political Discourse",
"authors": [
{
"first": "Kristen",
"middle": [],
"last": "Johnson",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Goldwasser",
"suffix": ""
}
],
"year": 2018,
"venue": "The 56th Annual Meeting of the Association for Computational Linguistics, ACL'18",
"volume": "",
"issue": "",
"pages": "720--730",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kristen Johnson and Dan Goldwasser. 2018. Classification of Moral Foundations in Microblog Political Dis- course. In The 56th Annual Meeting of the Association for Computational Linguistics, ACL'18, pages 720-730, July.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Moral Politics: What Conservatives Know That Liberals Don't",
"authors": [
{
"first": "George",
"middle": [],
"last": "Lakoff",
"suffix": ""
}
],
"year": 1997,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "George Lakoff. 1997. Moral Politics: What Conservatives Know That Liberals Don't. University of Chicago Press.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Moral Foundations in the 2015-16 U.S. Presidential Primary Debates: The Positive and Negative Moral Vocabulary of Partisan Elites",
"authors": [
{
"first": "G",
"middle": [],
"last": "Paul",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Lewis",
"suffix": ""
}
],
"year": 2019,
"venue": "Social Sciences",
"volume": "8",
"issue": "233",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Paul G. Lewis. 2019. Moral Foundations in the 2015-16 U.S. Presidential Primary Debates: The Positive and Negative Moral Vocabulary of Partisan Elites. Social Sciences, 8(233).",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Acquiring Background Knowledge to Improve Moral Value Prediction",
"authors": [
{
"first": "Ying",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Joe",
"middle": [],
"last": "Hoover",
"suffix": ""
},
{
"first": "Gwenyth",
"middle": [],
"last": "Portillo-Wightman",
"suffix": ""
},
{
"first": "Christina",
"middle": [],
"last": "Park",
"suffix": ""
},
{
"first": "Morteza",
"middle": [],
"last": "Dehghani",
"suffix": ""
},
{
"first": "Heng",
"middle": [],
"last": "Ji",
"suffix": ""
}
],
"year": 2018,
"venue": "IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining",
"volume": "",
"issue": "",
"pages": "552--559",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ying Lin, Joe Hoover, Gwenyth Portillo-Wightman, Christina Park, Morteza Dehghani, and Heng Ji. 2018. Ac- quiring Background Knowledge to Improve Moral Value Prediction. In 2018 IEEE/ACM International Confer- ence on Advances in Social Networks Analysis and Mining (ASONAM), pages 552-559.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Development of the Japanese Moral Foundations Dictionary: Procedures and Applications",
"authors": [
{
"first": "Akiko",
"middle": [],
"last": "Matsuo",
"suffix": ""
},
{
"first": "Kazutoshi",
"middle": [],
"last": "Sasahara",
"suffix": ""
},
{
"first": "Yasuhiro",
"middle": [],
"last": "Taguchi",
"suffix": ""
},
{
"first": "Minoru",
"middle": [],
"last": "Karasawa",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Akiko Matsuo, Kazutoshi Sasahara, Yasuhiro Taguchi, and Minoru Karasawa. 2018. Development of the Japanese Moral Foundations Dictionary: Procedures and Applications. CoRR, abs/1804.00871.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Glove: Global vectors for word representation",
"authors": [
{
"first": "Jeffrey",
"middle": [],
"last": "Pennington",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Socher",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "1532--1543",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word represen- tation. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532-1543.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
"authors": [
{
"first": "Nils",
"middle": [],
"last": "Reimers",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"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": "3982--3992",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982-3992, Hong Kong, China, November. Association for Computational Linguistics.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Enhancing the Measurement of Social Effects by Capturing Morality",
"authors": [
{
"first": "Rezvaneh",
"middle": [],
"last": "Rezapour",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Saumil",
"suffix": ""
},
{
"first": "Jana",
"middle": [],
"last": "Shah",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Diesner",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the Tenth Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis",
"volume": "",
"issue": "",
"pages": "35--45",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rezvaneh Rezapour, Saumil H. Shah, and Jana Diesner. 2019. Enhancing the Measurement of Social Effects by Capturing Morality. In Proceedings of the Tenth Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 35-45, Minneapolis, USA, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Divergent framing: The public debate on migration in the Dutch parliament and media",
"authors": [
{
"first": "Conny",
"middle": [],
"last": "Roggeband",
"suffix": ""
},
{
"first": "Rens",
"middle": [],
"last": "Vliegenthart",
"suffix": ""
}
],
"year": 1995,
"venue": "West European Politics",
"volume": "3",
"issue": "30",
"pages": "524--548",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Conny Roggeband and Rens Vliegenthart. 2007. Divergent framing: The public debate on migration in the Dutch parliament and media, 1995-2004. West European Politics, 3(30):524-548.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Annotation, Modelling and Analysis of Fine-Grained Emotions on a Stance and Sentiment Detection Corpus",
"authors": [
{
"first": "Hendrik",
"middle": [],
"last": "Schuff",
"suffix": ""
},
{
"first": "Jeremy",
"middle": [],
"last": "Barnes",
"suffix": ""
},
{
"first": "Julian",
"middle": [],
"last": "Mohme",
"suffix": ""
},
{
"first": "Sebastian",
"middle": [],
"last": "Pad\u00f3",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Klinger",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis",
"volume": "",
"issue": "",
"pages": "13--23",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hendrik Schuff, Jeremy Barnes, Julian Mohme, Sebastian Pad\u00f3, and Roman Klinger. 2017. Annotation, Mod- elling and Analysis of Fine-Grained Emotions on a Stance and Sentiment Detection Corpus. In Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 13-23, Copenhagen, Denmark, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "The Kappa Statistic in Reliability Studies: Use, Interpretation, and Sample Size Requirements",
"authors": [
{
"first": "Julius",
"middle": [],
"last": "Sim",
"suffix": ""
},
{
"first": "Chris",
"middle": [
"C"
],
"last": "Wright",
"suffix": ""
}
],
"year": 2005,
"venue": "Physical Therapy",
"volume": "85",
"issue": "3",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Julius Sim and Chris C. Wright. 2005. The Kappa Statistic in Reliability Studies: Use, Interpretation, and Sample Size Requirements. Physical Therapy, 85(3):257-268, 03.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "Automatic argumentation mining and the role of stance and sentiment",
"authors": [
{
"first": "Manfred",
"middle": [],
"last": "Stede",
"suffix": ""
}
],
"year": 2020,
"venue": "Journal of Argumentation in Context",
"volume": "9",
"issue": "1",
"pages": "19--41",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Manfred Stede. 2020. Automatic argumentation mining and the role of stance and sentiment. Journal of Argu- mentation in Context, 9(1):19-41.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "Moral Foundations of Political Discourse: Comparative Analysis of the Speech Records of the US Congress and the Japanese Diet",
"authors": [
{
"first": "Hiroki",
"middle": [],
"last": "Takikawa",
"suffix": ""
},
{
"first": "Takuto",
"middle": [],
"last": "Sakamoto",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hiroki Takikawa and Takuto Sakamoto. 2017. Moral Foundations of Political Discourse: Comparative Analysis of the Speech Records of the US Congress and the Japanese Diet. CoRR, abs/1704.06903.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "Pywsd: Python Implementations of Word Sense Disambiguation (WSD) Technologies",
"authors": [
{
"first": "Liling",
"middle": [],
"last": "Tan",
"suffix": ""
}
],
"year": 2014,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Liling Tan. 2014. Pywsd: Python Implementations of Word Sense Disambiguation (WSD) Technologies [soft- ware]. https://github.com/alvations/pywsd.",
"links": null
},
"BIBREF31": {
"ref_id": "b31",
"title": "Morally reframed arguments can affect support for political candidates",
"authors": [
{
"first": "Jan",
"middle": [
"G"
],
"last": "Voelkel",
"suffix": ""
},
{
"first": "Matthew",
"middle": [],
"last": "Feinberg",
"suffix": ""
}
],
"year": 2018,
"venue": "Social Psychological and Personality Science",
"volume": "9",
"issue": "8",
"pages": "917--924",
"other_ids": {
"PMID": [
"30595808"
]
},
"num": null,
"urls": [],
"raw_text": "Jan G. Voelkel and Matthew Feinberg. 2018. Morally reframed arguments can affect support for political candi- dates. Social Psychological and Personality Science, 9(8):917-924. PMID: 30595808.",
"links": null
},
"BIBREF32": {
"ref_id": "b32",
"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",
"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. 2017. Computational Argumentation Quality Assessment in Natural Lan- guage. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 176-187. Association for Computational Linguistics.",
"links": null
},
"BIBREF33": {
"ref_id": "b33",
"title": "A Comparison Of Emotion Annotation Schemes And A New Annotated Data Set",
"authors": [
{
"first": "Ian",
"middle": [],
"last": "Wood",
"suffix": ""
},
{
"first": "John",
"middle": [
"P"
],
"last": "Mccrae",
"suffix": ""
},
{
"first": "Vladimir",
"middle": [],
"last": "Andryushechkin",
"suffix": ""
},
{
"first": "Paul",
"middle": [],
"last": "Buitelaar",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ian Wood, John P. McCrae, Vladimir Andryushechkin, and Paul Buitelaar. 2018. A Comparison Of Emotion Annotation Schemes And A New Annotated Data Set. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan, May. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF34": {
"ref_id": "b34",
"title": "Text-based inference of moral sentiment change",
"authors": [
{
"first": "Jing",
"middle": [],
"last": "Yi Xie",
"suffix": ""
},
{
"first": "Renato Ferreira Pinto",
"middle": [],
"last": "Junior",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Hirst",
"suffix": ""
},
{
"first": "Yang",
"middle": [],
"last": "Xu",
"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": "4654--4663",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jing Yi Xie, Renato Ferreira Pinto Junior, Graeme Hirst, and Yang Xu. 2019. Text-based inference of moral sentiment change. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Pro- cessing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4654-4663, Hong Kong, China, November. Association for Computational Linguistics.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"type_str": "table",
"content": "<table/>",
"html": null,
"text": "Inter-Annotator Agreement for the five MFs and for a binary label (Moral: yes/no).",
"num": null
},
"TABREF3": {
"type_str": "table",
"content": "<table><tr><td colspan=\"8\">Method Moral Care Fairness Loyalty Authority Purity Average (excl. Moral)</td></tr><tr><td>Random baseline</td><td>.658</td><td>.257</td><td>.179</td><td>.096</td><td>.134</td><td>.105</td><td>.154</td></tr><tr><td>MFD baseline</td><td>.853</td><td>.056</td><td>.237</td><td>.043</td><td>.200</td><td>.086</td><td>.124</td></tr><tr><td>multi-label BERT</td><td>.444</td><td>.517</td><td>.519</td><td>.138</td><td>.157</td><td>.208</td><td>.308</td></tr><tr><td>WN-PPR</td><td>.756</td><td>.118</td><td>.253</td><td>.049</td><td>.105</td><td>.029</td><td>.111</td></tr><tr><td>SBERT-Base</td><td>.703</td><td>.280</td><td>.342</td><td>.065</td><td>.148</td><td>.133</td><td>.194</td></tr><tr><td>SBERT-Wiki</td><td>.730</td><td>.339</td><td>.246</td><td>.125</td><td>.233</td><td>.318</td><td>.252</td></tr><tr><td>WN-PPR + SBERT-Wiki</td><td>.686</td><td>.298</td><td>.351</td><td>.067</td><td>.040</td><td>.135</td><td>.178</td></tr></table>",
"html": null,
"text": "Binary F1-scores on the MFTC for individual MFs (F1 for the positive class). The last column shows the average over the F1 scores for the five MFs (excluding Moral).",
"num": null
},
"TABREF4": {
"type_str": "table",
"content": "<table><tr><td colspan=\"2\">Care</td><td colspan=\"2\">Fairness</td><td colspan=\"2\">Loyalty</td><td colspan=\"2\">Authority</td><td colspan=\"2\">Purity</td><td>Moral</td></tr><tr><td colspan=\"3\">HU BM HU</td><td>BM</td><td>HU</td><td>BM</td><td>HU</td><td>BM</td><td>HU</td><td>BM</td><td>HU</td><td>BM</td></tr><tr><td>Dagstuhl ArgQuality Corpus</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>overall quality .25</td><td>.15</td><td>.10</td><td>.08</td><td colspan=\"4\">.05 .10 \u2212 -.09 .05 +</td><td>.03</td><td>.07 \u2212</td><td>.19</td><td>.21</td></tr><tr><td>local acceptability .18</td><td>.09</td><td colspan=\"9\">.00 -.04 .03</td><td>.09</td></tr><tr><td>appropriateness .30</td><td>.17</td><td>-.01</td><td>.03</td><td colspan=\"4\">-.02 .05 \u2212 -.09 .00 +</td><td>.01</td><td>.02</td><td>.19</td><td>.15</td></tr><tr><td>arrangement .24</td><td>.16</td><td>.08</td><td>.03</td><td colspan=\"4\">.03 .08 \u2212 -.06 -.01 +</td><td>.04</td><td>.05</td><td>.16</td><td>.17</td></tr><tr><td>clarity .17</td><td>.17</td><td>.02</td><td>.02</td><td colspan=\"4\">.05 .12 \u2212 -.03 -.01 +</td><td>.03</td><td>.06</td><td>.09 .21 \u2212</td></tr><tr><td>cogency .24</td><td>.16</td><td>.05</td><td>.06</td><td colspan=\"4\">-.02 .03 \u2212 -.10 .05 +</td><td>.01</td><td>.03 \u2212</td><td>.10</td><td>.18</td></tr><tr><td>effectiveness .25</td><td>.17</td><td>.09</td><td>.09</td><td colspan=\"4\">-.05 .07 \u2212 -.10 -.02 +</td><td>.04</td><td>.04</td><td>.21</td><td>.17</td></tr><tr><td>global acceptability .23</td><td>.12</td><td>.05</td><td>.05</td><td colspan=\"4\">-.01 .07 \u2212 -.12 .02 +</td><td>.01</td><td>.04</td><td>.12</td><td>.13</td></tr><tr><td>global relevance .15</td><td>.06</td><td>.11</td><td>.09</td><td colspan=\"4\">.02 .07 \u2212 -.11 .00 +</td><td>.04</td><td>.05</td><td>.12</td><td>.11</td></tr><tr><td>global sufficiency .19</td><td>.11</td><td>.11</td><td>.11</td><td colspan=\"4\">-.01 .06 \u2212 -.04 -.03 +</td><td>.07</td><td>.05 \u2212</td><td>.19 .14 +</td></tr><tr><td>reasonableness .23</td><td>.17</td><td>.09</td><td>.08</td><td colspan=\"3\">.02 .08 \u2212 -.11</td><td>.04</td><td>.06</td><td>.07 \u2212</td><td>.16</td><td>.18</td></tr><tr><td>local relevance .18</td><td>.14</td><td>.08</td><td>.03</td><td colspan=\"4\">.01 .01 \u2212 -.10 .02 +</td><td>.02</td><td>-.02</td><td>.12</td><td>.13</td></tr><tr><td>credibility .22</td><td>.07</td><td>.06</td><td>.02 \u2212</td><td>.05</td><td>-.01</td><td colspan=\"2\">-.13 .01 +</td><td colspan=\"2\">-.01 .00 \u2212</td><td>.09</td><td>.08</td></tr><tr><td>emotional appeal .32</td><td>.22</td><td>.16</td><td>.12 +</td><td colspan=\"4\">.14 .02 \u2212 -.01 .10 +</td><td>-.01</td><td>.02</td><td>.31</td><td>.25</td></tr><tr><td>sufficiency .25</td><td>.18</td><td>.06</td><td>.09 \u2212</td><td colspan=\"4\">.00 .04 \u2212 -.10 .03 +</td><td>.07</td><td>.06 \u2212</td><td>.15 .19 +</td></tr><tr><td>IBM-AQR</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>quality (WA)</td><td>.08</td><td/><td>.06</td><td/><td>.01</td><td/><td>.00</td><td/><td>-.02</td><td>.08 \u2212</td></tr><tr><td>quality (MACE-P)</td><td>.08</td><td/><td>.05</td><td/><td>.01</td><td/><td>.00</td><td/><td>-.01</td><td>.07 \u2212</td></tr><tr><td>stance</td><td>.07</td><td/><td>.01</td><td/><td>-.03</td><td/><td>.01</td><td/><td>-.03 +</td><td>.04</td></tr><tr><td>CORPS</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>applause</td><td>.02</td><td/><td>.04</td><td/><td>.07</td><td/><td>.05</td><td/><td>.01</td><td>.10</td></tr><tr><td>laughter</td><td>-.07</td><td/><td>-.05</td><td/><td>-.05</td><td/><td>-.03</td><td/><td>-.02</td><td>-.11</td></tr></table>",
"html": null,
"text": ".00 .04 \u2212 -.15 -.01 + -.03 .07 \u2212",
"num": null
}
}
}
} |