File size: 127,693 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 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:28:29.496680Z"
},
"title": "Amplifying the Range of News Stories with Creativity: Methods and their Evaluation, in Portuguese",
"authors": [
{
"first": "Rui",
"middle": [],
"last": "Mendes",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Coimbra",
"location": {
"country": "Portugal"
}
},
"email": ""
},
{
"first": "Hugo",
"middle": [
"Gon\u00e7alo"
],
"last": "Oliveira",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Coimbra",
"location": {
"country": "Portugal"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Headlines are key for attracting people to a story, but writing appealing headlines requires time and talent. This work aims to automate the production of creative short texts (e.g., news headlines) for an input context (e.g., existing headlines), thus amplifying its range. Well-known expressions (e.g., proverbs, movie titles), which typically include word-play and resort to figurative language, are used as a starting point. Given an input text, they can be recommended by exploiting Semantic Textual Similarity (STS) techniques, or adapted towards higher relatedness. For the latter, three methods that exploit static word embeddings are proposed. Experimentation in Portuguese lead to some conclusions, based on human opinions: STS methods that look exclusively at the surface text, recommend more related expressions; resulting expressions are somewhat related to the input, but adaptation leads to higher relatedness and novelty; humour can be an indirect consequence, but most outputs are not funny.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Headlines are key for attracting people to a story, but writing appealing headlines requires time and talent. This work aims to automate the production of creative short texts (e.g., news headlines) for an input context (e.g., existing headlines), thus amplifying its range. Well-known expressions (e.g., proverbs, movie titles), which typically include word-play and resort to figurative language, are used as a starting point. Given an input text, they can be recommended by exploiting Semantic Textual Similarity (STS) techniques, or adapted towards higher relatedness. For the latter, three methods that exploit static word embeddings are proposed. Experimentation in Portuguese lead to some conclusions, based on human opinions: STS methods that look exclusively at the surface text, recommend more related expressions; resulting expressions are somewhat related to the input, but adaptation leads to higher relatedness and novelty; humour can be an indirect consequence, but most outputs are not funny.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Each minute, new stories are published online, listed in news aggregators, and spread through social media. With the amount of information each of us is constantly bombed with, most people end up looking at the headlines and, only sporadically, opening and reading the full text. We may thus say that headlines play a key role in this process: the more appealing they are, the higher the probability of someone actually reading their story. And while personal preferences are relevant, creativity and familiarity (e.g., text resembles a funny saying or situation) certainly contribute to higher appeal.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A common strategy for making the headline more catchy is to reuse expressions known by a general audience, sometimes also attempting at a humorous effect. If the expression is related enough, it can be used directly, but it may also suffer minor adaptations, to become more related to the context and still resemble the original saying. This is also common in news satire tv shows, like The Daily Show. While the host is telling a story, in one of the top corners of the screen, a short text (e.g., a book, movie title, proverb, idiomatic expression, or their adaptation), often accompanied by an image, complements the scene.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "However, writing catchy headlines requires time and talent. Even if resorting to well-known expressions is a possible shortcut, it is still a knowledge intensive task (i.e., on folk or pop culture) and requires creativity skills. Therefore, in this paper, we propose to automate this process and assess different unsupervised methods for, given a short text (e.g., a news headline): (i) recommending a related known expression (e.g., a proverb) from a set; (ii) adapting a known expression so that it becomes related to the input. In any case, it should be possible to use the resulting expression as an alternative, though more creative, headline, sub-title, or, at least, a comment in social media. Methods tested for recommendation are mostly baselines for Semantic Textual Similarity (STS) (Agirre et al., 2012) , including some based on the surface text, others based on static word embeddings, and on BERT (Devlin et al., 2019) contextual embeddings. As for the adaptation, static word embeddings are exploited for replacing some words in the expressions, taken directly from the headline, or based on similarity or analogy.",
"cite_spans": [
{
"start": 794,
"end": 815,
"text": "(Agirre et al., 2012)",
"ref_id": "BIBREF0"
},
{
"start": 912,
"end": 933,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Proposed methods were tested with a set of news headlines, as input text, and a set of proverbs and movie titles. We have worked with Portuguese, and even though the resources and tools used are for the target language, theoretically, the methods are language-independent. The development of automatic approaches for producing creative artefacts, sometimes based on a given context or cur-rent events, is not new. The proposed adaptation methods have some novelty, when compared to the well-established methods for recommendation. However, here they are used with proverbs, which poses additional challenges, such as the frequent utilisation of figurative language. Finally, we target Portuguese, a language for which work of this kind is still scarce.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "For better understanding how successful this approach was, results of this Portuguese instantiation were assessed. Due to the underlying subjectivity in the appreciation of the results, evaluations (one for recommendation, another for adaptation) are based on the opinion of human judges, who scored pairs of headlines and expressions by different methods. This further enabled to compare methods and draw some conclusions, such as: on average, results of recommendation methods are not significantly different from each other, but using expressions that share words with the headline seems to increase the perceived relatedness; resulting expressions are somewhat related to the input, and adaptation leads to higher relatedness and novelty; humour can be an indirect consequence, but most outputs are not funny.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In the remainder of this paper, we overview work on the generation of creative text, focusing on those inspired by news or current events. We then describe our approaches for amplifying news stories with creative text and their evaluation: first, for the recommendation of known expressions, including the comparison of different methods; second, we propose three methods for adapting such expressions. We conclude with a brief discussion and some cues for future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Computational Creativity (CC) is at the intersection of Artificial Intelligence and scientific areas like cognitive psychology, philosophy, and the arts. One of its ends is to develop creative systems, i.e., computational systems that exhibit behaviours deemed as creative by unbiased observers (Colton and Wiggins, 2012) , e.g., they can produce new music or visual art, among others, including text.",
"cite_spans": [
{
"start": 295,
"end": 321,
"text": "(Colton and Wiggins, 2012)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Work on linguistically-creative systems has tackled the generation of textual artefacts like narrative (Gerv\u00e1s et al., 2019) , poetry Chrismartin and Manurung, 2015; Gon\u00e7alo Oliveira, 2017) , memorable news headlines (Gatti et al., 2015; Veale et al., 2017; Alnajjar et al., 2019) and slogans (Alnajjar and Toivonen, 2020) , or humour (Binsted and Ritchie, 1997; Valitutti et al., 2016) . Among those, some were adapted for producing new textual content based on current events. Poetry generation has been: guided by dependency relations in a single news story (Chrismartin and Manurung, 2015) ; inspired by the mood and related similes in a set of news stories, possibly including sentences from one of them ; or inspired by Twitter trends, associated words and semantic relations (Gon\u00e7alo Oliveira, 2017) .",
"cite_spans": [
{
"start": 103,
"end": 124,
"text": "(Gerv\u00e1s et al., 2019)",
"ref_id": "BIBREF17"
},
{
"start": 134,
"end": 165,
"text": "Chrismartin and Manurung, 2015;",
"ref_id": "BIBREF10"
},
{
"start": 166,
"end": 189,
"text": "Gon\u00e7alo Oliveira, 2017)",
"ref_id": "BIBREF18"
},
{
"start": 217,
"end": 237,
"text": "(Gatti et al., 2015;",
"ref_id": "BIBREF16"
},
{
"start": 238,
"end": 257,
"text": "Veale et al., 2017;",
"ref_id": null
},
{
"start": 258,
"end": 280,
"text": "Alnajjar et al., 2019)",
"ref_id": "BIBREF2"
},
{
"start": 293,
"end": 322,
"text": "(Alnajjar and Toivonen, 2020)",
"ref_id": "BIBREF3"
},
{
"start": 335,
"end": 362,
"text": "(Binsted and Ritchie, 1997;",
"ref_id": "BIBREF7"
},
{
"start": 363,
"end": 386,
"text": "Valitutti et al., 2016)",
"ref_id": "BIBREF33"
},
{
"start": 561,
"end": 593,
"text": "(Chrismartin and Manurung, 2015)",
"ref_id": "BIBREF10"
},
{
"start": 782,
"end": 806,
"text": "(Gon\u00e7alo Oliveira, 2017)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Work on increasing the creativity, and thus memorability, of news headlines often resorts to wellknown expressions, with which readers are familiar, including slogans, movie or song titles. On this context, the creativity of an automated journalism system was increased with the injection of known expressions and figurative language (Alnajjar et al., 2019) , e.g., the headline \"Biggest gains for The Christian Democrats across Lapin vaalipiiri\" may become \"Biggest gains for The Christian Democrats, as powerful as a soldier, across Lapin vaalipiiri'. Human-written headlines were also blended with well-known expressions, through the substitution of words in the expression with keywords from the headline (Gatti et al., 2015) , e.g., \"What the Euro is coming to\" for the original headline \"UK anger at 1.7bn EU cash demand\". To reduce the risk of producing outputs with a detached meaning, a threshold was set on the cosine similarity between the original and the produced headline. Also relying on vector semantics and on the cosine, previously generated metaphors have been paired with the current news (Veale et al., 2017) , e.g., the news tweet \"@newtgingrich says 'the country will become enraged' if the violent protests at @realDonaldTrump rallies continue\" was paired with metaphors like What is a radical but a crusading demagogue?.",
"cite_spans": [
{
"start": 334,
"end": 357,
"text": "(Alnajjar et al., 2019)",
"ref_id": "BIBREF2"
},
{
"start": 709,
"end": 729,
"text": "(Gatti et al., 2015)",
"ref_id": "BIBREF16"
},
{
"start": 1109,
"end": 1129,
"text": "(Veale et al., 2017)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "When it comes to the generation of creative text, word embeddings are useful resources. In fact, the operations of similarity, neighborhood, theme and analogy have been formalised for adapting text by lexical replacement, with increased creativity, given a set of intentions (Bay et al., 2017) . In our work, we also propose methods for adapting wellknown expressions, based on a textual input (context), more like Gatti et al. (2015) , with word embeddings exploited in the selection of suitable replacements. Although any text should work, news headlines were also used.",
"cite_spans": [
{
"start": 275,
"end": 293,
"text": "(Bay et al., 2017)",
"ref_id": "BIBREF6"
},
{
"start": 415,
"end": 434,
"text": "Gatti et al. (2015)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Humour can be a consequence of the former approaches, but it is rarely tackled specifically. There is, however, work on humour generation, also relying on lexical substitution in short texts (Valitutti et al., 2016) . In this case, messages, e.g., \"Later we go where to eat?\" may become \"Later we go where to shit?\". Replacement words should have the same part-of-speech of the original, a similar sound or writing and, the more efficient constraint, is that they are taboos. Hossain et al. (2019) present a corpus of original news headlines and their manuallyedited funny versions. While such a corpus may be useful for many tasks, including humor generation, funny headlines are often out-of-context (e.g., \"EU says summit with Turkey provides no answers to concerns\" becomes \"EU says gravy with Turkey provides no answers to concerns\"), which is different from our goal.",
"cite_spans": [
{
"start": 191,
"end": 215,
"text": "(Valitutti et al., 2016)",
"ref_id": "BIBREF33"
},
{
"start": 476,
"end": 497,
"text": "Hossain et al. (2019)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Another important difference of our work is that we experiment and present results for the Portuguese language, while the majority of the aforementioned systems processes and produces English text. For Portuguese, related work is less, but still covers: the automatic generation of poetry (Gon\u00e7alo Oliveira, 2017); riddles (Gon\u00e7alo Oliveira and Rodrigues, 2018), with the help of lexical-semantic knowledge; and memes (Gon\u00e7alo Oliveira et al., 2016) , based on current news and rules for selecting an image macro and adapting the text.",
"cite_spans": [
{
"start": 418,
"end": 449,
"text": "(Gon\u00e7alo Oliveira et al., 2016)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "A related task is that of headline generation for a given document (see e.g., (Takase et al., 2016) ), which has some similarities with automatic summarization (Banko et al., 2000) and may involve a deeper understanding of the news story. This is, however, different from our goal, as our starting point are existing headlines, for which new creative alternatives are recommended or produced.",
"cite_spans": [
{
"start": 78,
"end": 99,
"text": "(Takase et al., 2016)",
"ref_id": "BIBREF32"
},
{
"start": 160,
"end": 180,
"text": "(Banko et al., 2000)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Alternatively to text generation, some systems simply recommend (famous) quotes (Ahn et al., 2016) , or retrieve interactions in movie subtitles (Ameixa et al., 2014) , to be used in dialogues. This may involve training an encoder-decoder network (Shang et al., 2015) in dialogues where target quotes are used. Though, an unsupervised retrievalbased approach is also possible, e.g., relying on vector semantics and pretrained language models for computing Semantic Textual Similarity (Agirre et al., 2012; Cer et al., 2017) , and retrieving the most similar quote. In this work, we also test different unsupervised methods for the direct recom-mendation of well-known expressions.",
"cite_spans": [
{
"start": 80,
"end": 98,
"text": "(Ahn et al., 2016)",
"ref_id": "BIBREF1"
},
{
"start": 145,
"end": 166,
"text": "(Ameixa et al., 2014)",
"ref_id": "BIBREF4"
},
{
"start": 247,
"end": 267,
"text": "(Shang et al., 2015)",
"ref_id": "BIBREF31"
},
{
"start": 484,
"end": 505,
"text": "(Agirre et al., 2012;",
"ref_id": "BIBREF0"
},
{
"start": 506,
"end": 523,
"text": "Cer et al., 2017)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "The first part of this work tackled the recommendation of suitable expressions for a given short text. A good example of our goal is taken from the 17th May 2020 edition of the Portuguese satire news tv show Isto\u00e9 Gozar com Quem Trabalha: when presenting a story about the plan of the Portuguese President to take a swim in the sea after the Covid-19 lockdown, they used the text \"O Velho e o Mar\" (The Old Man and the Sea, a book by Ernest Hemingway). For this, we tested different methods for computing the semantic similarity between news headlines, the input text, and Portuguese proverbs, the expressions to output. All methods are unsupervised and can be seen as baselines for STS. Once all available proverbs are ranked according to the input, the first, which maximises similarity, is recommended. This section describes the tested methods, how they were used in this experimentation, their results and evaluation by human judges.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Recommendation of Creative Text in Context",
"sec_num": "3"
},
{
"text": "The following STS methods were covered: shallow methods that consider only the surface text; methods based on sparse vector representations of the text, with word counts and TF-IDF weighting; methods based on static word embeddings, with sentences represented by the average embedding, weighted with TF-IDF or not; and methods based on contextual embeddings that encode each sentence in a single vector. Overall, eight methods were tested: Jaccard Similarity; CountVectorizer; TfIdfVectorizer; GloVe (Pennington et al., 2014) embeddings, with and without TF-IDF; FastText-CBOW (Bojanowski et al., 2017) embeddings with and without TF-IDF; and BERT (Devlin et al., 2019) embeddings. All methods return a value between 0 and 1, proportional to the similarity between the two sentences. By definition, Jaccard Similarity already does this, while, for the other methods, this value is given by the cosine similarity between the vector representations of each text.",
"cite_spans": [
{
"start": 500,
"end": 525,
"text": "(Pennington et al., 2014)",
"ref_id": "BIBREF27"
},
{
"start": 577,
"end": 602,
"text": "(Bojanowski et al., 2017)",
"ref_id": "BIBREF8"
},
{
"start": 648,
"end": 669,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "3.1"
},
{
"text": "With the News API 1 , a set of 60 news headlines was gathered from online editions of Portuguese news-papers. The source of expressions was a corpus of 1,617 Portuguese proverbs, obtained from project Natura 2 . For all methods but BERT, headlines and proverbs were tokenized with the NLPyPort package (Ferreira et al., 2019) , a layer on top of NLTK (Loper and Bird, 2002) for better handling Portuguese.",
"cite_spans": [
{
"start": 302,
"end": 325,
"text": "(Ferreira et al., 2019)",
"ref_id": "BIBREF15"
},
{
"start": 351,
"end": 373,
"text": "(Loper and Bird, 2002)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "3.2"
},
{
"text": "Sparse-vector representations of sentences and the computation of TF-IDF were based on the corpus of proverbs. For GloVe and FastText, we used pretrained models for Portuguese, both with 300sized vectors, respectively from NILC (Hartmann et al., 2017) and fastText.cc 3 repositories. Finally, for BERT, we used a pretrained multilingual model that covers 104 languages, including Portuguese, BERT-Base, Multilingual Cased 4 .",
"cite_spans": [
{
"start": 228,
"end": 251,
"text": "(Hartmann et al., 2017)",
"ref_id": "BIBREF22"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "3.2"
},
{
"text": "All methods were implemented in Python, with the help of the following packages:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "3.2"
},
{
"text": "scikit-learn (Pedregosa et al., 2011) , for the CountVectorizer (Count) and TfIdfVectorizer (TFIDF); gensim (\u0158eh\u016f\u0159ek and Sojka, 2010), for handling the static word embeddings; and bert-as-a-service 5 , for loading and using BERT for encoding sentences.",
"cite_spans": [
{
"start": 13,
"end": 37,
"text": "(Pedregosa et al., 2011)",
"ref_id": "BIBREF26"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "3.2"
},
{
"text": "As it happens with many other creative outputs, assessing the suitability of a proverb to a headline is a subjective task, which cannot be automatised. Therefore, to compare the performance of each method in the proposed scenario, we relied on human opinions. For this, proverbs were recommended by each method for each of the 60 headlines in the gathered set. Several surveys were then created with Google Forms, each having ten of those headlines followed by the headline's recommended proverbs, at most eight, randomly distributed, without repetitions or any identification with regard to the method.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "We then asked 24 Portuguese-speaking volunteers to answer the surveys. Each survey was answered by four different judges, who scored the proverbs recommended for ten headlines, according to their relatedness and funniness, with the following questions: (a) \"How would you rate the relation between the proverbs and the news title?\" [Not related (1); Remotely related (2); Considerably related (3); Extremely related (4)]; (b) \"In re-lation to the headline, how funny is each proverb?\" [Not funny (1); Remotely funny (2); Considerably funny (3); Extremely funny (4)]. Volunteers were not informed that the expressions were proverbs nor that they had been recommended automatically. Table 1 shows the distribution of scores for recommendations by each method, according to human opinions. It also includes the median (x), which is, nevertheless, not very discriminant. We omit both GloVe and FastText with TF-IDF because their scores were not much different.",
"cite_spans": [],
"ref_spans": [
{
"start": 681,
"end": 688,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "Out of the possible scores for relatedness, not related (1) was always the most common. A curious outcome is that pretrained embeddings, like GloVe, FastText and, especially, BERT, do not make much qualitative difference on the results. In fact, most judges gave higher scores to proverbs that share one or more words with the headline, which does not always happen when word or sentence embeddings are involved (e.g., first two examples in table 2). We recall that most proverbs are highly figurative, meaning that models trained in general language may struggle to interpret them.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "A deeper look shows that only two methods had at least 10% recommendations with average relatedness scores higher than 3.5, namely the Jaccard Similarity (12%) and TFIDF (10%); and only three recommendations had the highest average relatedness, namely the first three examples in table 2.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "Funniness scores are not much different, and only two recommendations got the highest average score from all judges, including the last two examples in table 2. Though not intended, both use taboo words, which should have contributed to their high score. This also suggests that we can increase funniness by forcing the presence of such words (Valitutti et al., 2016) . On the other hand, it comes at the cost of lower relatedness.",
"cite_spans": [
{
"start": 343,
"end": 367,
"text": "(Valitutti et al., 2016)",
"ref_id": "BIBREF33"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "3.3"
},
{
"text": "According the previous evaluation, it is not easy to find suitable proverbs for a given headline. This could possibly be improved if the corpus of proverbs were increased. However, headlines can be significantly different and, virtually, on any topic, so another option is to start from any known expression and adapt it, so that it becomes more related to the headline, while still resembling the original expression.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Adaptation of Expressions to a Context",
"sec_num": "4"
},
{
"text": "This is not new and is also a commonly adopted strategy in news satires. For instance, in the 5th N\u00e3o contes com o ovo no cu da galinha.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Adaptation of Expressions to a Context",
"sec_num": "4"
},
{
"text": "(Do not count with the egg in the chicken's butt.) Jaccard 1.5 4 April 2020 edition of Isto\u00e9 Gozar com Quem Trabalha, the expression \"Droga de Elite\" (Elite drug), an adaptation of the Brazilian movie title \"Tropa de Elite\" (Elite Squad), illustrated a story on Covid-19 in the Brazilian favelas, where drug dealers were ensuring that residents followed the sanitary rules. This section describes three methods that explore static word embeddings in the automatic adaptation of a given expression, to suit, as much as possible, an input short text, in such a way that it can be used for transmitting or complementing the same idea, though more creatively. After presenting the methods, we describe an experiment where Portuguese proverbs and movie titles were adapted for news headlines, followed by their results and evaluation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Adaptation of Expressions to a Context",
"sec_num": "4"
},
{
"text": "We propose three automatic methods for adapting a known expression for a given short text. Besides a list of well-known expressions, to be modified according to the input text (e.g., news headline), all methods: (i) exploit a pretrained model of static word embeddings; (ii) assume that the most relevant words in a text are previously computed; (iii) go through all the expressions in a list and try to make adaptations guided by the most relevant words of both the expressions and the input texts. Methods only differ on the adopted strategies for selecting the word(s) to replace.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "The first method, Substitution, replaces the most relevant word in the expression, a, by a word from the input text, b, or by a word similar to b. Our intuition is that, by using a relevant word of the input text, the meaning of the expression becomes more semantically-related to the given context.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "The second method, Analogy, relies on a common operation for computing analogies in word embeddings, i.e., b \u2212 a + a * = b * (Mikolov et al., 2013) , phrased as b * is to b as a * is to a. The strategy is to use the two most relevant words in the expression as a and a * , the most relevant word in the input as b, and then: (i) from the previous three, compute a new word b * ; (ii) in the original expression, replace a and a * by b and b * , respectively. Given that both pairs of words are analogouslyrelated, our intuition is that the result will still make sense and be more related to the input text.",
"cite_spans": [
{
"start": 125,
"end": 147,
"text": "(Mikolov et al., 2013)",
"ref_id": "BIBREF25"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "The third method, Vector Difference (VecDiff), also selects the two most relevant words in the input text, b and b * , and then: (i) computes the vector between the previous b \u2212 b * ; (ii) identifies the pair of open-class words in the expression, a and a * , such that a \u2212 a * maximises the (cosine) similarity with b \u2212 b * ; (iii) replaces a and a * by b and b * , respectively. Our intuition is that the new text will not only use two words of the input, and thus be more related, but also that they will be included in such a way that their relation is roughly preserved.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "To avoid syntactic inconsistencies, in any method, replacement candidates must match the morphology of the replaced word, including partof-speech (PoS), gender and number, according to a morphology lexicon or a PoS tagger. If morphology does not match, the lexicon can be further used to inflect the candidate to the target form. Moreover, the set of possible replacements can be augmented by considering not only the relevant words in the input, but also their most semantically-similar, computed in the embeddings, e.g., in the Substitution method, a can be replaced by a word different but semantically similar to b. Table 3 shows an example of the application of each method, including the original headline, a proverb and the resulting output. Replaced words and their replacements are underlined. In the first example, b = bancos replaces a = amigos. In the second, comeces = apontar \u2212 deixes + f azer.",
"cite_spans": [],
"ref_spans": [
{
"start": 620,
"end": 627,
"text": "Table 3",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "In the third, f ere \u2212 f erido \u2248 f inge \u2212 detido.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods",
"sec_num": "4.1"
},
{
"text": "Although the proposed methods are language dependent, we focused again on Portuguese. This time, we used the same \u22481,600 proverbs as in the previous section, but added about 2,500 movie titles in Portuguese, obtained from IMDB 6 . For better identification of the titles, we discarded the 25% oldest and all others with less than four words. Moreover, to avoid the inclusion of English names, all words in the title had to be in a Portuguese morphology lexicon.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "4.2"
},
{
"text": "Regarding pre-processing, NLPyPort (Ferreira et al., 2019) was used for tokenization, PoS tagging and lemmatization, and the morphology lexicon LABEL-Lex (Ranchhod et al., 1999) for information on word inflections. Relevant words were considered to be the least frequent, but appearing, in the newspaper corpus CETEMP\u00fablico (Rocha and Santos, 2000) . For word embeddings, we used the same GloVe model as in the previous section.",
"cite_spans": [
{
"start": 35,
"end": 58,
"text": "(Ferreira et al., 2019)",
"ref_id": "BIBREF15"
},
{
"start": 154,
"end": 177,
"text": "(Ranchhod et al., 1999)",
"ref_id": "BIBREF28"
},
{
"start": 324,
"end": 348,
"text": "(Rocha and Santos, 2000)",
"ref_id": "BIBREF29"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "4.2"
},
{
"text": "For experimentation, we used a set of 100 news headlines on different topics, posted between April and May 2020, in the Twitter accounts of Portuguese newspapers. An initial set was randomly selected, but darker headlines (e.g., about death) were manually excluded later, to make 100. For each headline, new texts were produced with the previous three methods. However, application to each headline could result in several new texts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "4.2"
},
{
"text": "Even if, due to the morphology constraints, some expressions end up not being used, others will. Plus, for each relevant word, we considered the top-5 similar. Thus, the same method may produce several variations of the same text for the same input, one for each.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "4.2"
},
{
"text": "For selecting a single expression, the recommendation methods described in section ?? are used in the beginning and in the end of the process. First, given a headline, they select the subset of the full corpus of expressions, in which each adaptation method will be applied to. Since the previous evaluation (section 3.3) did not help in the choice of a single recommendation method, we used two significantly different methods for this selection: TF-IDF and BERT. More precisely, given the headline, this subset will contain the top-30 expressions recommended by TF-IDF, the top-30 recommended by BERT, plus a random selection of 30, for higher diversity. This should still result in several (adapted) expressions, where a single one has to be selected from. This final selection is again made with the help of TF-IDF or BERT, which rank the results, allowing us to use only the top-ranked, i.e., the most similar to the headline, as long as it is not equals to an existing expression. Thus, depending on the final recommendation method, the resulting expression may be different.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimentation Setup",
"sec_num": "4.2"
},
{
"text": "To get insights on the suitability of the proposed methods, we relied again on human opinions. Since text was being changed automatically, it was now important to assess syntax, i.e., whether the resulting expression had grammatical or structural issues, that could make it difficult to interpret. However, syntax is less subjective than the other criteria, so it was assessed by the authors in a set of expressions produced for another subset of 30 headlines. This was enough to conclude that there were only syntactic issues in a minority of the produced expressions -3% for Substitution, 8% for Analogy, 13% in Vector Difference -and the majority of these issues did not have much impact on interpretability.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "Furthermore, we decided to assess an important aspect of creativity, which is novelty. In addition to being related to the input text, results should also be novel, in the sense that they have not been produced before or associated with the input text. Ideally, in addition to those, it would be nice again if the new expression had the potential of making the reader laugh (funniness).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "To have resulting expressions scored according to the evaluation criteria, a survey was again created, with a headline in each page, followed by expressions from eight different processes, namely the result of adapting the TF-IDF or BERT recommendations to the output of the three adaptation methods (6), plus expressions recommended directly by TF-IDF and BERT, with no adaptation (2). The latter were included to enable a comparison between reusing well-known expressions directly or with an adaptation, also having in mind that, in section 3, the headlines were different and so was the evaluation scale.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "For each resulting expression, human judges were asked to score the relatedness, by selecting one of the following: (1) There is no relation at all between the generated expression and the input;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "(2) The expression is somewhat related to the input, because it shares one or two words, or other contextual aspects; (3) The expression is clearly related to the input's context, could replace it or be used as a comment. Novelty, could be scored as: (1) I knew the expression well before reading it;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "(2) Reminds me of some expression, but has some changes; (3) The expression is completely new to me. Finally, for funniness, the options were:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "(1) The expression is not funny and should make no one laugh; (2) The expression is somewhat funny or could be, depending on the reader's subjective view;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "(3) The expression is very funny and has a great potential to make people laugh.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "This time, we had 100 Portuguese-speaking volunteers, each answering the previous questions for five headlines and their eight expressions, in such a way that each expression was scored by five different judges, which were not informed that the expressions were produced automatically. Table 4 has the distribution of scores given in the human evaluation, for text produced by each method.",
"cite_spans": [],
"ref_spans": [
{
"start": 286,
"end": 293,
"text": "Table 4",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "We recall that one of the motivations for adapting text instead of simply reusing it, was to increase relatedness. Even though the impact was not as high as expected, this trend is confirmed by the higher proportion of the highest scores (2 and 3) for the adaptation methods on this criteria. This is more clear if we look at results using the same method for the final recommendation. Moreover, figures suggest again that TF-IDF leads to better relatedness, because texts recommended by this method have a higher proportion of results scored with 3 than those by BERT. In section 3.3, we pointed out that this can be due to BERT recommendations sharing less words with the context. Yet, in the future we should also explore different ways of using BERT for this purpose.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "When using TF-IDF in the final recommendation, about a third of the results is highly related to the headline, which is ok, especially if, before utilisation, results can be manually selected out of the top-ranked. Curiously, the proportion of completely related results is higher for the simplest method, Substitution, and lower for the Vector Difference. The latter is also the only of the three methods that, in this scenario, has more than a third of results considered unrelated to their headline. Table 5 presents three high-scored examples, with the adaptation method, the input headline, the original expression and the output expression, with considered words underlined and average scores in the three assessed criteria.",
"cite_spans": [],
"ref_spans": [
{
"start": 503,
"end": 510,
"text": "Table 5",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "On the remaining criteria, adapted expressions are more novel, a key aspect for a creative system that shows we can go further than simply reusing text. This was expected because we are comparing newly created expressions with proverbs and movie titles, most of which are part of our culture and known by many people. At the same time, even though they are new, resulting expressions should still resemble existing expressions, which is why not all are scored with 3. For the same reason, novelty of the recommendation methods was lower, but not always 1, possibly because some judges did not know all the expressions. This, however, could have also influenced, positively, the scores of the adaptation methods, i.e., if the judge does not know the original expression, they will also not associate its adaptation with an expression they previously heard. On this criteria, BERT did not make much difference when applied to the adaptation methods, but novelty of its recommendations is significantly higher than for TF-IDF. We also note that the ranking of the adaptation methods according to novelty is the inverse of the relatedness, possibly because some judges rated the novelty when compared to the headline. The middle example in table 5 is one of the best scored results on novelty.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "About a half of the results of each method is just not funny. On the other hand, the proportion of clearly funny results is between 19% and 26%, which, considering that this criteria is not explicitly tackled, is not bad, and suggests that, indeed, humour can be a consequence of this word-play. All the examples in table 5 have average funniness of 2 or higher. In the first, funniness is more subjective, but the result may suggest too much promiscuity between banks and the government. The last two make unexpected associations, like people with a herd, or honey / love with despair.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "4.3"
},
{
"text": "Aiming at the amplification of news stories, we explored a set of automatic methods for making headlines more creative and appealing, with the exploitation of well-known expressions and word embeddings. When tested with Portuguese news headlines, human opinions suggest that our goals were achieved with relative success. Results are somewhat related to input headlines, especially if known expressions are adapted, and not used directly, which also increases novelty, intimately related to creativity. For the best adaptation methods, about a third of the results was clearly related to the headline. Although, in a few cases, humour was an indirect consequence, most of the outputs were not so funny. In the future, funniness may benefit from ranking candidate results with a humour classifier, based on different humour-relevant features, such as the one recently proposed for Portuguese (Clem\u00eancio et al., 2019) .",
"cite_spans": [
{
"start": 891,
"end": 915,
"text": "(Clem\u00eancio et al., 2019)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "5"
},
{
"text": "All the methods are unsupervised and exploring supervision was never our goal, mainly because the lack of training data, e.g., different Portuguese headlines for the same news scored according to their creativity, possibly also including some created specifically for this purpose. While such a dataset is not available for Portuguese, for English, a crowdsourced corpus of 15k original news headlines and their manually-edited funny versions was recently presented (Hossain et al., 2019) , making thus room for a data-driven approach for our task.",
"cite_spans": [
{
"start": 466,
"end": 488,
"text": "(Hossain et al., 2019)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "5"
},
{
"text": "Another observation of our work was that, in this context, methods that consider exclusively the surface text retrieve proverbs that are perceived to be more related. This happens because recommended proverbs tend to use some words of the input, immediately suggesting some relation. Although state-of-the-art BERT embeddings would be better at representing sentence meanings, they are not apt to deal with the figurative language in proverbs, at least if they are not fine-tuned for this, or possibly trained only for the target language. In the future, it would be interesting to make a deeper study on STS and figurative language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "5"
},
{
"text": "The proposed methods can be integrated in a tool for suggesting alternative headlines that are still related to a story, possibly useful for journalists. Even if not all results have the necessary quality, different options may be produced (e.g., the top-n for each method), faster than if it were a human, who may still make the final decision or additional adaptations. To some extent, enabling human interaction would make such a tool co-creative, as other interactive systems developed for writing stories (Roemmele and Gordon, 2015) , song lyrics (Watanabe et al., 2017) or poetry , among others.",
"cite_spans": [
{
"start": 510,
"end": 537,
"text": "(Roemmele and Gordon, 2015)",
"ref_id": "BIBREF30"
},
{
"start": 552,
"end": 575,
"text": "(Watanabe et al., 2017)",
"ref_id": "BIBREF35"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "5"
},
{
"text": "Finally, this work should also contribute to the development of more creative chatbots, e.g., capable of providing creative responses to out-ofdomain interactions. In the meantime, the TECo Twitterbot 7 is already following several Portuguese newspapers and retweeting some of their publications together with comments produced by the proposed methods.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "5"
},
{
"text": "https://newsapi.org/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "natura.di.uminho.pt/wiki/doku.php 3 https://fasttext.cc/ 4 github.com/google-research/bert 5 github.com/hanxiao/bert-as-service",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://www.imdb.com/interfaces/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "SemEval-2012 task 6: A pilot on semantic textual similarity",
"authors": [
{
"first": "Eneko",
"middle": [],
"last": "Agirre",
"suffix": ""
},
{
"first": "Mona",
"middle": [],
"last": "Diab",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Cer",
"suffix": ""
},
{
"first": "Aitor",
"middle": [],
"last": "Gonzalez-Agirre",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of 1st Joint Conference on Lexical and Computational Semantics",
"volume": "1",
"issue": "",
"pages": "385--393",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eneko Agirre, Mona Diab, Daniel Cer, and Aitor Gonzalez-Agirre. 2012. SemEval-2012 task 6: A pi- lot on semantic textual similarity. In Proceedings of 1st Joint Conference on Lexical and Computa- tional Semantics-Vol. 1: Proceedings of main con- ference and shared task, and Vol. 2: Proceedings of Sixth International Workshop on Semantic Eval- uation, pages 385-393. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Quote recommendation for dialogs and writings",
"authors": [
{
"first": "Yeonchan",
"middle": [],
"last": "Ahn",
"suffix": ""
},
{
"first": "Hanbit",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Heesik",
"middle": [],
"last": "Jeon",
"suffix": ""
},
{
"first": "Seungdo",
"middle": [],
"last": "Ha",
"suffix": ""
},
{
"first": "Sang-Goo",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of 3rd Workshop on New Trends in Content-Based Recommender Systems, CBRecSys@ RecSys",
"volume": "",
"issue": "",
"pages": "39--42",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yeonchan Ahn, Hanbit Lee, Heesik Jeon, Seungdo Ha, and Sang-goo Lee. 2016. Quote recommendation for dialogs and writings. In Proceedings of 3rd Workshop on New Trends in Content-Based Recom- mender Systems, CBRecSys@ RecSys, pages 39-42.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "TextoEmContexto ating colourful and factual multilingual news headlines",
"authors": [
{
"first": "Khalid",
"middle": [],
"last": "Alnajjar",
"suffix": ""
},
{
"first": "Leo",
"middle": [],
"last": "Lepp\u00e4nen",
"suffix": ""
},
{
"first": "Hannu",
"middle": [],
"last": "Toivonen",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of 10th International Conference on Computational Creativity, ICCC 2019",
"volume": "",
"issue": "",
"pages": "258--265",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Khalid Alnajjar, Leo Lepp\u00e4nen, and Hannu Toivonen. 2019. No time like the present: Methods for gener- 7 https://twitter.com/TextoEmContexto ating colourful and factual multilingual news head- lines. In Proceedings of 10th International Con- ference on Computational Creativity, ICCC 2019, pages 258-265. Association for Computational Cre- ativity.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Computational generation of slogans",
"authors": [
{
"first": "Khalid",
"middle": [],
"last": "Alnajjar",
"suffix": ""
},
{
"first": "Hannu",
"middle": [],
"last": "Toivonen",
"suffix": ""
}
],
"year": 2020,
"venue": "Natural Language Engineering",
"volume": "",
"issue": "",
"pages": "1--33",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Khalid Alnajjar and Hannu Toivonen. 2020. Compu- tational generation of slogans. Natural Language Engineering, pages 1-33.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Luke, i am your father: dealing with out-of-domain requests by using movies subtitles",
"authors": [
{
"first": "David",
"middle": [],
"last": "Ameixa",
"suffix": ""
},
{
"first": "Luisa",
"middle": [],
"last": "Coheur",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Fialho",
"suffix": ""
},
{
"first": "Paulo",
"middle": [],
"last": "Quaresma",
"suffix": ""
}
],
"year": 2014,
"venue": "International Conference on Intelligent Virtual Agents",
"volume": "",
"issue": "",
"pages": "13--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David Ameixa, Luisa Coheur, Pedro Fialho, and Paulo Quaresma. 2014. Luke, i am your father: dealing with out-of-domain requests by using movies sub- titles. In International Conference on Intelligent Vir- tual Agents, pages 13-21. Springer.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Headline generation based on statistical translation",
"authors": [
{
"first": "Michele",
"middle": [],
"last": "Banko",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Vibhu",
"suffix": ""
},
{
"first": "Michael",
"middle": [
"J"
],
"last": "Mittal",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Witbrock",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "318--325",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Michele Banko, Vibhu O Mittal, and Michael J Wit- brock. 2000. Headline generation based on statist- ical translation. In Proceedings of the 38th Annual Meeting of the Association for Computational Lin- guistics, pages 318-325.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Text transformation via constraints and word embedding",
"authors": [
{
"first": "Benjamin",
"middle": [],
"last": "Bay",
"suffix": ""
},
{
"first": "Paul",
"middle": [],
"last": "Bodily",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Ventura",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of 8th International Conference on Computational Creativity",
"volume": "",
"issue": "",
"pages": "49--56",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Benjamin Bay, Paul Bodily, and Dan Ventura. 2017. Text transformation via constraints and word em- bedding. In Proceedings of 8th International Con- ference on Computational Creativity, ICCC 2017, pages 49-56.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Computational rules for generating punning riddles",
"authors": [
{
"first": "Kim",
"middle": [],
"last": "Binsted",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Ritchie",
"suffix": ""
}
],
"year": 1997,
"venue": "Humor: International Journal of Humor Research",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kim Binsted and Graeme Ritchie. 1997. Computa- tional rules for generating punning riddles. Humor: International Journal of Humor Research.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Enriching word vectors with subword information",
"authors": [
{
"first": "Piotr",
"middle": [],
"last": "Bojanowski",
"suffix": ""
},
{
"first": "Edouard",
"middle": [],
"last": "Grave",
"suffix": ""
},
{
"first": "Armand",
"middle": [],
"last": "Joulin",
"suffix": ""
},
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
}
],
"year": 2017,
"venue": "Transactions of the Association for Computational Linguistics",
"volume": "5",
"issue": "",
"pages": "135--146",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. Enriching word vectors with subword information. Transactions of the Associ- ation for Computational Linguistics, 5:135-146.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Semeval-2017 task 1: Semantic textual similarity multilingual and crosslingual focused evaluation",
"authors": [
{
"first": "Daniel",
"middle": [],
"last": "Cer",
"suffix": ""
},
{
"first": "Mona",
"middle": [],
"last": "Diab",
"suffix": ""
},
{
"first": "Eneko",
"middle": [],
"last": "Agirre",
"suffix": ""
},
{
"first": "Inigo",
"middle": [],
"last": "Lopez-Gazpio",
"suffix": ""
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of 11th International Workshop on Semantic Evaluation (SemEval-2017)",
"volume": "",
"issue": "",
"pages": "1--14",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Daniel Cer, Mona Diab, Eneko Agirre, Inigo Lopez- Gazpio, and Lucia Specia. 2017. Semeval-2017 task 1: Semantic textual similarity multilingual and crosslingual focused evaluation. In Proceedings of 11th International Workshop on Semantic Eval- uation (SemEval-2017), pages 1-14. Association for Computational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "A chart generation system for topical meaningful metrical poetry",
"authors": [
{
"first": "Berty",
"middle": [],
"last": "Chrismartin",
"suffix": ""
},
{
"first": "Ruli",
"middle": [],
"last": "Manurung",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of The 6th International Conference on Computational Creativity, ICCC 2015",
"volume": "",
"issue": "",
"pages": "308--314",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Berty Chrismartin and Ruli Manurung. 2015. A chart generation system for topical meaningful metrical poetry. In Proceedings of The 6th International Con- ference on Computational Creativity, ICCC 2015, pages 308-314, Park City, UT, USA.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Recognizing humor in Portuguese: First steps",
"authors": [
{
"first": "Andr\u00e9",
"middle": [],
"last": "Clem\u00eancio",
"suffix": ""
},
{
"first": "Ana",
"middle": [],
"last": "Alves",
"suffix": ""
},
{
"first": "Hugo Gon\u00e7alo",
"middle": [],
"last": "Oliveira",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of 19th EPIA Conference on Artificial Intelligence",
"volume": "2019",
"issue": "",
"pages": "744--756",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andr\u00e9 Clem\u00eancio, Ana Alves, and Hugo Gon\u00e7alo Oli- veira. 2019. Recognizing humor in Portuguese: First steps. In Proceedings of 19th EPIA Conference on Artificial Intelligence, EPIA 2019, Vila Real, Por- tugal, September 3-6, 2019, Part II, volume 11805 of LNCS/LNAI, pages 744-756. Springer.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Full FACE poetry generation",
"authors": [
{
"first": "Simon",
"middle": [],
"last": "Colton",
"suffix": ""
},
{
"first": "Jacob",
"middle": [],
"last": "Goodwin",
"suffix": ""
},
{
"first": "Tony",
"middle": [],
"last": "Veale",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of 3rd International Conference on Computational Creativity (ICCC)",
"volume": "",
"issue": "",
"pages": "95--102",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Simon Colton, Jacob Goodwin, and Tony Veale. 2012. Full FACE poetry generation. In Proceedings of 3rd International Conference on Computational Creativ- ity (ICCC), pages 95-102, Dublin, Ireland.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Computational creativity: the final frontier?",
"authors": [
{
"first": "Simon",
"middle": [],
"last": "Colton",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Geraint",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Wiggins",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 20th European Conference on Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "21--26",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Simon Colton and Geraint A Wiggins. 2012. Computa- tional creativity: the final frontier? In Proceedings of the 20th European Conference on Artificial Intel- ligence, pages 21-26. IOS Press.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "4171--4186",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understand- ing. In Proceedings of 2019 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171- 4186. Association for Computational Linguistics.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Improving NLTK for processing Portuguese",
"authors": [
{
"first": "Jo\u00e3o",
"middle": [],
"last": "Ferreira",
"suffix": ""
},
{
"first": "Hugo",
"middle": [
"Gon\u00e7alo"
],
"last": "Oliveira",
"suffix": ""
},
{
"first": "Ricardo",
"middle": [],
"last": "Rodrigues",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of 8th Symposium on Languages, Applications and Technologies",
"volume": "74",
"issue": "",
"pages": "1--18",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jo\u00e3o Ferreira, Hugo Gon\u00e7alo Oliveira, and Ricardo Rodrigues. 2019. Improving NLTK for processing Portuguese. In Proceedings of 8th Symposium on Languages, Applications and Technologies (SLATE 2019), volume 74 of OASIcs, pages 18:1-18:9. Schloss Dagstuhl.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Slogans are not forever: Adapting linguistic expressions to the news",
"authors": [
{
"first": "Lorenzo",
"middle": [],
"last": "Gatti",
"suffix": ""
},
{
"first": "Marco",
"middle": [],
"last": "G\u00f6zde\u00f6zbal",
"suffix": ""
},
{
"first": "Oliviero",
"middle": [],
"last": "Guerini",
"suffix": ""
},
{
"first": "Carlo",
"middle": [],
"last": "Stock",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings 24th International Joint Conference on Artificial Intelligence, IJCAI 2015",
"volume": "",
"issue": "",
"pages": "2452--2458",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lorenzo Gatti, G\u00f6zde\u00d6zbal, Marco Guerini, Oliviero Stock, and Carlo Strapparava. 2015. Slogans are not forever: Adapting linguistic expressions to the news. In Proceedings 24th International Joint Conference on Artificial Intelligence, IJCAI 2015, pages 2452- 2458. AAAI Press.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "The long path to narrative generation",
"authors": [
{
"first": "Pablo",
"middle": [],
"last": "Gerv\u00e1s",
"suffix": ""
},
{
"first": "Eugenio",
"middle": [],
"last": "Concepci\u00f3n",
"suffix": ""
},
{
"first": "Carlos",
"middle": [],
"last": "Le\u00f3n",
"suffix": ""
},
{
"first": "Gonzalo",
"middle": [],
"last": "M\u00e9ndez",
"suffix": ""
},
{
"first": "Pablo",
"middle": [],
"last": "Delatorre",
"suffix": ""
}
],
"year": 2019,
"venue": "IBM Journal of Research and Development",
"volume": "63",
"issue": "1",
"pages": "8--9",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pablo Gerv\u00e1s, Eugenio Concepci\u00f3n, Carlos Le\u00f3n, Gonzalo M\u00e9ndez, and Pablo Delatorre. 2019. The long path to narrative generation. IBM Journal of Research and Development, 63(1):8-1.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "O Poeta Artificial 2.0: Increasing meaningfulness in a poetry generation Twitter bot",
"authors": [
{
"first": "",
"middle": [],
"last": "Hugo Gon\u00e7alo Oliveira",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the Workshop on Computational Creativity in Natural Language Generation (CC-NLG 2017)",
"volume": "",
"issue": "",
"pages": "11--20",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hugo Gon\u00e7alo Oliveira. 2017. O Poeta Artificial 2.0: Increasing meaningfulness in a poetry generation Twitter bot. In Proceedings of the Workshop on Computational Creativity in Natural Language Gen- eration (CC-NLG 2017), pages 11-20, Santiago de Compostela, Spain. Association for Computational Linguistics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "One does not simply produce funny memes! -explorations on the automatic generation of Internet humor",
"authors": [
{
"first": "Diogo",
"middle": [],
"last": "Hugo Gon\u00e7alo Oliveira",
"suffix": ""
},
{
"first": "Alexandre",
"middle": [],
"last": "Costa",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Pinto",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of 7th International Conference on Computational Creativity, ICCC 2016",
"volume": "",
"issue": "",
"pages": "238--245",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hugo Gon\u00e7alo Oliveira, Diogo Costa, and Alexandre Pinto. 2016. One does not simply produce funny memes! -explorations on the automatic genera- tion of Internet humor. In Proceedings of 7th In- ternational Conference on Computational Creativ- ity, ICCC 2016, pages 238-245, Paris, France.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Co-PoeTryMe: a co-creative interface for the composition of poetry",
"authors": [
{
"first": "Tiago",
"middle": [],
"last": "Hugo Gon\u00e7alo Oliveira",
"suffix": ""
},
{
"first": "Ana",
"middle": [],
"last": "Mendes",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Boavida",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of 10th International Conference on Natural Language Generation",
"volume": "",
"issue": "",
"pages": "70--71",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hugo Gon\u00e7alo Oliveira, Tiago Mendes, and Ana Boavida. 2017. Co-PoeTryMe: a co-creative inter- face for the composition of poetry. In Proceedings of 10th International Conference on Natural Lan- guage Generation, INLG 2017, pages 70-71, San- tiago de Compostela, Spain. ACL Press.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Exploring lexical-semantic knowledge in the generation of novel riddles in Portuguese",
"authors": [
{
"first": "Gon\u00e7alo",
"middle": [],
"last": "Hugo",
"suffix": ""
},
{
"first": "Ricardo",
"middle": [],
"last": "Oliveira",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Rodrigues",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 3rd Workshop on Computational Creativity in Natural Language Generation",
"volume": "",
"issue": "",
"pages": "17--25",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hugo Gon\u00e7alo Oliveira and Ricardo Rodrigues. 2018. Exploring lexical-semantic knowledge in the gener- ation of novel riddles in Portuguese. In Proceed- ings of the 3rd Workshop on Computational Creativ- ity in Natural Language Generation, CC-NLG 2018, pages 17-25, Tilburg, The Netherlands. Association for Computational Linguistics.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Portuguese word embeddings: Evaluating on word analogies and natural language tasks",
"authors": [
{
"first": "Nathan",
"middle": [
"S"
],
"last": "Hartmann",
"suffix": ""
},
{
"first": "Erick",
"middle": [
"R"
],
"last": "Fonseca",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Shulby",
"suffix": ""
},
{
"first": "Marcos",
"middle": [
"V"
],
"last": "Treviso",
"suffix": ""
},
{
"first": "J\u00e9ssica",
"middle": [
"S"
],
"last": "Rodrigues",
"suffix": ""
},
{
"first": "Sandra",
"middle": [
"M"
],
"last": "Alu\u00edsio",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of 11th Brazilian Symposium in Information and Human Language Technology",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nathan S. Hartmann, Erick R. Fonseca, Christopher D. Shulby, Marcos V. Treviso, J\u00e9ssica S. Rodrigues, and Sandra M. Alu\u00edsio. 2017. Portuguese word em- beddings: Evaluating on word analogies and natural language tasks. In Proceedings of 11th Brazilian Symposium in Information and Human Language Technology (STIL 2017).",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "President Vows to Cut <Taxes> Hair\": Dataset and analysis of creative text editing for humorous headlines",
"authors": [
{
"first": "Nabil",
"middle": [],
"last": "Hossain",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Krumm",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Gamon",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "133--142",
"other_ids": {
"DOI": [
"10.18653/v1/N19-1012"
]
},
"num": null,
"urls": [],
"raw_text": "Nabil Hossain, John Krumm, and Michael Gamon. 2019. \"President Vows to Cut <Taxes> Hair\": Dataset and analysis of creative text editing for hu- morous headlines. In Proceedings of the 2019 Con- ference of the North American Chapter of the Asso- ciation for Computational Linguistics: Human Lan- guage Technologies, Volume 1 (Long and Short Pa- pers), pages 133-142, Minneapolis, Minnesota. As- sociation for Computational Linguistics.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Nltk: The natural language toolkit",
"authors": [
{
"first": "Edward",
"middle": [],
"last": "Loper",
"suffix": ""
},
{
"first": "Steven",
"middle": [],
"last": "Bird",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the ACL-02 Workshop on Effective Tools and Methodologies for Teaching Natural Language Processing and Computational Linguistics",
"volume": "",
"issue": "",
"pages": "63--70",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Edward Loper and Steven Bird. 2002. Nltk: The nat- ural language toolkit. In Proceedings of the ACL-02 Workshop on Effective Tools and Methodologies for Teaching Natural Language Processing and Compu- tational Linguistics, pages 63-70.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Linguistic regularities in continuous space word representations",
"authors": [
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "Yih",
"middle": [],
"last": "Wen-Tau",
"suffix": ""
},
{
"first": "Geoffrey",
"middle": [],
"last": "Zweig",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of 2013 Conference of the North American Chapter of the ACL: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "746--751",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig. 2013. Linguistic regularities in continuous space word representations. In Proceedings of 2013 Con- ference of the North American Chapter of the ACL: Human Language Technologies, pages 746-751. As- sociation for Computational Linguistics.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Scikit-learn: Machine learning in Python",
"authors": [
{
"first": "F",
"middle": [],
"last": "Pedregosa",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Varoquaux",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Gramfort",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Michel",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Thirion",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Grisel",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Blondel",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Prettenhofer",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Weiss",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Dubourg",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Vanderplas",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Passos",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Cournapeau",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Brucher",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Perrot",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Duchesnay",
"suffix": ""
}
],
"year": 2011,
"venue": "Journal of Machine Learning Research",
"volume": "12",
"issue": "",
"pages": "2825--2830",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12:2825-2830.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "Glove: Global vectors for word representation",
"authors": [
{
"first": "Jeffrey",
"middle": [],
"last": "Pennington",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Socher",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "1532--1543",
"other_ids": {
"DOI": [
"10.3115/v1/D14-1162"
]
},
"num": null,
"urls": [],
"raw_text": "Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. Glove: Global vectors for word rep- resentation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Pro- cessing (EMNLP), pages 1532-1543, Doha, Qatar. Association for Computational Linguistics.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "A computational lexicon of Portuguese for automatic text parsing",
"authors": [
{
"first": "Elisabete",
"middle": [],
"last": "Ranchhod",
"suffix": ""
},
{
"first": "Cristina",
"middle": [],
"last": "Mota",
"suffix": ""
},
{
"first": "Jorge",
"middle": [],
"last": "Baptista",
"suffix": ""
}
],
"year": 1999,
"venue": "Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks",
"volume": "",
"issue": "",
"pages": "45--50",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Elisabete Ranchhod, Cristina Mota, and Jorge Baptista. 1999. A computational lexicon of Portuguese for automatic text parsing. In Proceedings of SIGLEX99 Workshop: Standardizing Lexical Re- sources. Association for Computational Linguistics. Radim\u0158eh\u016f\u0159ek and Petr Sojka. 2010. Software Framework for Topic Modelling with Large Cor- pora. In Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks, pages 45- 50, Valletta, Malta. ELRA. http://is.muni.cz/ publication/884893/en.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "CETEMP\u00fablico: Um corpus de grandes dimens\u00f5es de linguagem jornal\u00edstica portuguesa",
"authors": [
{
"first": "Paulo",
"middle": [
"Alexandre"
],
"last": "Rocha",
"suffix": ""
},
{
"first": "Diana",
"middle": [],
"last": "Santos",
"suffix": ""
}
],
"year": 2000,
"venue": "V Encontro para o processamento computacional da l\u00edngua portuguesa escrita e falada (PROPOR 2000)",
"volume": "",
"issue": "",
"pages": "131--140",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Paulo Alexandre Rocha and Diana Santos. 2000. CETEMP\u00fablico: Um corpus de grandes dimens\u00f5es de linguagem jornal\u00edstica portuguesa. In V Encon- tro para o processamento computacional da l\u00edngua portuguesa escrita e falada (PROPOR 2000), pages 131-140, S\u00e3o Paulo. ICMC/USP.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "Creative help: a story writing assistant",
"authors": [
{
"first": "Melissa",
"middle": [],
"last": "Roemmele",
"suffix": ""
},
{
"first": "Andrew S",
"middle": [],
"last": "Gordon",
"suffix": ""
}
],
"year": 2015,
"venue": "International Conference on Interactive Digital Storytelling",
"volume": "",
"issue": "",
"pages": "81--92",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Melissa Roemmele and Andrew S Gordon. 2015. Cre- ative help: a story writing assistant. In Interna- tional Conference on Interactive Digital Storytelling, pages 81-92. Springer.",
"links": null
},
"BIBREF31": {
"ref_id": "b31",
"title": "Neural responding machine for short-text conversation",
"authors": [
{
"first": "Lifeng",
"middle": [],
"last": "Shang",
"suffix": ""
},
{
"first": "Zhengdong",
"middle": [],
"last": "Lu",
"suffix": ""
},
{
"first": "Hang",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing",
"volume": "1",
"issue": "",
"pages": "1577--1586",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lifeng Shang, Zhengdong Lu, and Hang Li. 2015. Neural responding machine for short-text conver- sation. In Proceedings of the 53rd Annual Meet- ing of the Association for Computational Linguistics and the 7th International Joint Conference on Nat- ural Language Processing (Volume 1: Long Papers), pages 1577-1586.",
"links": null
},
"BIBREF32": {
"ref_id": "b32",
"title": "Neural headline generation on abstract meaning representation",
"authors": [
{
"first": "Jun",
"middle": [],
"last": "Sho Takase",
"suffix": ""
},
{
"first": "Naoaki",
"middle": [],
"last": "Suzuki",
"suffix": ""
},
{
"first": "Tsutomu",
"middle": [],
"last": "Okazaki",
"suffix": ""
},
{
"first": "Masaaki",
"middle": [],
"last": "Hirao",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Nagata",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 conference on empirical methods in natural language processing",
"volume": "",
"issue": "",
"pages": "1054--1059",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sho Takase, Jun Suzuki, Naoaki Okazaki, Tsutomu Hirao, and Masaaki Nagata. 2016. Neural head- line generation on abstract meaning representation. In Proceedings of the 2016 conference on empir- ical methods in natural language processing, pages 1054-1059.",
"links": null
},
"BIBREF33": {
"ref_id": "b33",
"title": "Computational generation and dissection of lexical replacement humor",
"authors": [
{
"first": "Alessandro",
"middle": [],
"last": "Valitutti",
"suffix": ""
},
{
"first": "Antoine",
"middle": [],
"last": "Doucet",
"suffix": ""
},
{
"first": "Jukka",
"middle": [
"M"
],
"last": "Toivanen",
"suffix": ""
},
{
"first": "Hannu",
"middle": [],
"last": "Toivonen",
"suffix": ""
}
],
"year": 2016,
"venue": "Natural Language Engineering",
"volume": "22",
"issue": "5",
"pages": "727--749",
"other_ids": {
"DOI": [
"10.1017/S1351324915000145"
]
},
"num": null,
"urls": [],
"raw_text": "Alessandro Valitutti, Antoine Doucet, Jukka M. Toivanen, and Hannu Toivonen. 2016. Compu- tational generation and dissection of lexical re- placement humor. Natural Language Engineering, 22(5):727-749.",
"links": null
},
"BIBREF34": {
"ref_id": "b34",
"title": "2017. I read the news today, oh boy",
"authors": [
{
"first": "Tony",
"middle": [],
"last": "Veale",
"suffix": ""
},
{
"first": "Hanyang",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Guofu",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": null,
"venue": "International Conference on Distributed, Ambient, and Pervasive Interactions",
"volume": "",
"issue": "",
"pages": "696--709",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tony Veale, Hanyang Chen, and Guofu Li. 2017. I read the news today, oh boy. In International Conference on Distributed, Ambient, and Pervasive Interactions, pages 696-709. Springer.",
"links": null
},
"BIBREF35": {
"ref_id": "b35",
"title": "Lyrisys: An interactive support system for writing lyrics based on topic transition",
"authors": [
{
"first": "Kento",
"middle": [],
"last": "Watanabe",
"suffix": ""
},
{
"first": "Yuichiroh",
"middle": [],
"last": "Matsubayashi",
"suffix": ""
},
{
"first": "Kentaro",
"middle": [],
"last": "Inui",
"suffix": ""
},
{
"first": "Tomoyasu",
"middle": [],
"last": "Nakano",
"suffix": ""
},
{
"first": "Satoru",
"middle": [],
"last": "Fukayama",
"suffix": ""
},
{
"first": "Masataka",
"middle": [],
"last": "Goto",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 22nd international conference on intelligent user interfaces",
"volume": "",
"issue": "",
"pages": "559--563",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kento Watanabe, Yuichiroh Matsubayashi, Kentaro Inui, Tomoyasu Nakano, Satoru Fukayama, and Masataka Goto. 2017. Lyrisys: An interactive sup- port system for writing lyrics based on topic trans- ition. In Proceedings of the 22nd international con- ference on intelligent user interfaces, pages 559- 563.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"text": "Human evaluation of recommendation approaches.",
"num": null,
"content": "<table><tr><td>Headline</td><td>Proverb</td><td colspan=\"3\">Method Rel Fun</td></tr><tr><td>Mal\u00e1sia devolve 150 contentores ilegais de lixo a pa\u00edses</td><td>Quem faz de si lixo, pisam-no as galin-</td><td>TFIDF</td><td>4</td><td>3.5</td></tr><tr><td>subdesenvolvidos (Malaysia returns 150 illegal trash containers to</td><td>has (Whoever makes themselves trash, will be</td><td/><td/><td/></tr><tr><td>underdeveloped countries)</td><td>trampled by the chicken)</td><td/><td/><td/></tr><tr><td>Tempestade 'Gl\u00f3ria' fez 12 mortos em Espanha. Gov-</td><td>A culpa morre solteira (Guilt dies single)</td><td>TFIDF</td><td>4</td><td>3.25</td></tr><tr><td>erno culpa altera\u00e7\u00f5es clim\u00e1ticas ('Gloria' storm made 12 cas-</td><td/><td/><td/><td/></tr><tr><td>ualities in Spain. Government blames climate change)</td><td/><td/><td/><td/></tr><tr><td>Ainda n\u00e3o\u00e9 demasiado tarde para salvarmos os oceanos</td><td>N\u00e3o deixe para amanh\u00e3 o que voc\u00ea</td><td>BERT</td><td>4</td><td>2.5</td></tr><tr><td>(It is not too late to save the oceans)</td><td>pode fazer hoje (Do not leave for tomorrow</td><td/><td/><td/></tr><tr><td/><td>what you can do today)</td><td/><td/><td/></tr><tr><td>Veredicto abre a porta a protec\u00e7\u00e3o para 'refugiados</td><td>Para tr\u00e1s mija a burra. (The female don-</td><td>Jaccard</td><td>2.5</td><td>4</td></tr><tr><td>clim\u00e1ticos' (Veredict opens door for protection to 'climate refugees')</td><td>key pisses backwards)</td><td/><td/><td/></tr></table>",
"html": null,
"type_str": "table"
},
"TABREF2": {
"text": "",
"num": null,
"content": "<table/>",
"html": null,
"type_str": "table"
},
"TABREF4": {
"text": "Running examples of the application of each adaptation method.",
"num": null,
"content": "<table/>",
"html": null,
"type_str": "table"
},
"TABREF5": {
"text": "Vector Difference 34.6 35.6 29.8 2 17.0 35.4 47.6 2 53.4 27.0 19.6 1 Recommendation only 38.4 32.0 29.6 2 33.8 34.6 31.6 2 53.0 27.8 19,2 1",
"num": null,
"content": "<table><tr><td>Method</td><td>Relatedness (%) 1 2 3x</td><td>1</td><td>Novelty (%) 2 3x</td><td>Funniness (%) 1 2 3x</td></tr><tr><td colspan=\"2\">Final recommendation by TF-IDF</td><td/><td/><td/></tr><tr><td colspan=\"5\">Substitution 25.8 38.0 36.2 2 17.6 40.8 41.6 2 45.2 29.4 25.4 2</td></tr><tr><td colspan=\"5\">Analogy 29.0 37.8 33.2 2 17.4 36.4 46.2 2 44.6 29.8 25.6 2</td></tr><tr><td colspan=\"2\">Final recommendation by BERT</td><td/><td/><td/></tr><tr><td colspan=\"5\">Substitution 44.8 34.0 21.2 2 22.0 36.4 41.6 2 49.2 30.6 20.2 1</td></tr><tr><td colspan=\"5\">Analogy 38.8 36.6 24.6 2 20.0 33.8 46.2 2 52.0 28.2 19,8 1</td></tr><tr><td colspan=\"5\">Vector Difference 38.4 35.6 26.0 2 16.4 32.2 51.4 3 51.8 24.2 24.0 1</td></tr><tr><td colspan=\"5\">Recommendation only 52.5 27.8 19.8 1 22.0 35.0 43.0 2 46.0 29.8 24.2 2</td></tr></table>",
"html": null,
"type_str": "table"
},
"TABREF6": {
"text": "Human evaluation of the adaptation and recommendation approaches.",
"num": null,
"content": "<table/>",
"html": null,
"type_str": "table"
},
"TABREF7": {
"text": "Bancos dizem que as condi\u00e7\u00f5es das linhas de cr\u00e9dito foram definidas pelo governo (Banks claim that credit conditions were defined by the government) Dar a C\u00e9sar o que\u00e9 de C\u00e9sar e a Deus o qu\u00e9 e de Deus (To Caesar what is Caesar's and to God what is God's) Dar a governo o que\u00e9 de governo e a Deus o que\u00e9 de Deus (To the government what is the government's and to God what is God's)",
"num": null,
"content": "<table><tr><td>Method</td><td>Input</td><td>Original Expression</td><td>Output</td><td>Scores</td></tr><tr><td>Subs</td><td/><td/><td/><td>Rel= 3</td></tr><tr><td>+</td><td/><td/><td/><td>Nov= 2</td></tr><tr><td>TF-IDF</td><td/><td/><td/><td>Fun= 2.5</td></tr><tr><td>Analogy</td><td>Uma simples conversa gera</td><td>Uma ovelha m\u00e1 p\u00f5e o</td><td>Uma got\u00edcula suspensa</td><td>Rel= 2.25</td></tr><tr><td>+</td><td>got\u00edculas que podem ficar</td><td>rebanho a perder (A bad</td><td>p\u00f5e o rebanho a perder</td><td>Nov= 2.75</td></tr><tr><td>BERT</td><td>suspensas no ar at\u00e9 14 minutos</td><td>sheep makes the herd lose)</td><td>(A suspended droplet makes the</td><td>Fun= 2.5</td></tr><tr><td/><td>(A simple conversation generates droplets</td><td/><td>herd lose)</td><td/></tr><tr><td/><td>that can be suspended in the air for up to 14</td><td/><td/><td/></tr><tr><td/><td>minutes)</td><td/><td/><td/></tr><tr><td>VecDiff</td><td>Noivos desesperam por terminar</td><td>Ouro\u00e9 o que ouro vale</td><td>Mel desespera o que mel</td><td>Rel= 2.25</td></tr><tr><td>+</td><td>lua de mel, mesmo estando \"presos\"</td><td>(Gold is what gold is worth)</td><td>vale (Honey despairs honey is</td><td>Nov= 2.5</td></tr><tr><td>BERT</td><td>nas Maldivas (Newlyweds are desperate</td><td/><td>worth)</td><td>Fun= 2</td></tr><tr><td/><td>to end their honeymoon, even though they are</td><td/><td/><td/></tr><tr><td/><td>\"stuck\" in the Maldives)</td><td/><td/><td/></tr></table>",
"html": null,
"type_str": "table"
},
"TABREF8": {
"text": "Examples of high-scored expressions according to the human judges.",
"num": null,
"content": "<table/>",
"html": null,
"type_str": "table"
}
}
}
} |