File size: 94,022 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 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T01:06:14.325898Z"
},
"title": "damo_nlp at MEDIQA 2021: Knowledge-based Preprocessing and Coverage-oriented Reranking for Medical Question Summarization",
"authors": [
{
"first": "Yifan",
"middle": [],
"last": "He",
"suffix": "",
"affiliation": {
"laboratory": "Alibaba Group",
"institution": "",
"location": {}
},
"email": "y.he@alibaba-inc.com"
},
{
"first": "Mosha",
"middle": [],
"last": "Chen",
"suffix": "",
"affiliation": {
"laboratory": "Alibaba Group",
"institution": "",
"location": {}
},
"email": "chenmosha.cms@alibaba-inc.com"
},
{
"first": "Songfang",
"middle": [],
"last": "Huang",
"suffix": "",
"affiliation": {
"laboratory": "Alibaba Group",
"institution": "",
"location": {}
},
"email": "songfang.hsf@alibaba-inc.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Medical question summarization is an important but difficult task, where the input is often complex and erroneous while annotated data is expensive to acquire. We report our participation in the MEDIQA 2021 question summarization task in which we are required to address these challenges. We start from pre-trained conditional generative language models, use knowledge bases to help correct input errors, and rerank single system outputs to boost coverage. Experimental results show significant improvement in stringbased metrics.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "Medical question summarization is an important but difficult task, where the input is often complex and erroneous while annotated data is expensive to acquire. We report our participation in the MEDIQA 2021 question summarization task in which we are required to address these challenges. We start from pre-trained conditional generative language models, use knowledge bases to help correct input errors, and rerank single system outputs to boost coverage. Experimental results show significant improvement in stringbased metrics.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Question summarization for medical forum is important for medical knowledge discovery and retrieval and facilitates downstream tasks such as biomedical question answering (Jin et al., 2021) . Medical questions are often complex, scattered with non-medical information, and can sometimes be erroneous because forum users are not domain experts (Ben Abacha and Demner-Fushman, 2019) . In addition, annotation in the medical domain is harder to acquire than in the general domain. These challenges make medical question summarization an important and difficult task where annotation is often scarce. The MEDIQA 2021 shared task 1 (Ben Abacha et al., 2021), medical question summarization, requires participants to build summarization systems for noisy medical forum texts with limited annotation data. The official training set of the task is the MeQSum dataset (Ben Abacha and Demner-Fushman, 2019), which is composed of 1,000 medical questions and their corresponding summaries. The validation and test sets consist of 50 and 100 questions respectively and topic words are sometimes misspelled.",
"cite_spans": [
{
"start": 171,
"end": 189,
"text": "(Jin et al., 2021)",
"ref_id": null
},
{
"start": 343,
"end": 380,
"text": "(Ben Abacha and Demner-Fushman, 2019)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Scarcity of data, noisy input, and complexity and redundancy of text all pose challenges for ques-tion summarization systems. We try to address these challenges using a combination of knowledgebased error correction, pre-trained generative language models, and output reranking.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Knowledge-based error correction leverages multiple levels of lexical resources and a high coverage knowledge base to correct errors in input. Our experiments show that knowledge-based error correction helps downstream summarization performance according to the Rouge metric.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Pre-trained generative language models are transformer-based language models trained with loss functions that facilitate sequence to sequence generation. Models such as Pegasus (Zhang et al., 2020a) , BART (Lewis et al., 2020) , and T5 (Raffel et al., 2020) achieve state-of-the-art performance on various text generation tasks and are shown to perform well on few-shot generation scenarios (Goodwin et al., 2020) . We finetune pre-trained language models to obtain baseline systems with limited amount of training data.",
"cite_spans": [
{
"start": 177,
"end": 198,
"text": "(Zhang et al., 2020a)",
"ref_id": "BIBREF19"
},
{
"start": 206,
"end": 226,
"text": "(Lewis et al., 2020)",
"ref_id": "BIBREF9"
},
{
"start": 236,
"end": 257,
"text": "(Raffel et al., 2020)",
"ref_id": "BIBREF15"
},
{
"start": 391,
"end": 413,
"text": "(Goodwin et al., 2020)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Output reranking picks the best output among multiple systems. The availability of different language models offers a diverse set of summaries to choose from. We observe difference in summarization styles between the training and the validation set and devise a simple heuristic to pick the best output based on this observation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In the rest of the paper, we describe these components and report evaluation results on the validation and the test set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The MEDIQA question summarization task requires participants to summarize user generated medical queries into shorter, more focused questions. We present an example from the MEDIQA 2021 task 1 validation set in Figure 1 (a). We note that the name of the disease \"folliculitis\" is spelled Hi, Please can you help -I am writing from South Africa. My daughter suffers with acute folliculitus, and has been since the age of 13. She is now 20 and is in so much distress as nothing seems to alleviate the itching and soreness... I am writing to you for any help you could give me to try and assist her. Could you recommend a specialist and someone who could help us with research? Please could you point us in the right direction? I am happy to send through her lab tests -please let me know. Thanks How can we find a specialist or clinical trial for chronic folliculitis? incorrectly in the input question and the question contains a lot of irrelevant information. We attempt to correct misspellings with a dedicated module in our system. As useful information is often scattered in different sentences in the input, abstractive summarization suits this task better than extractive summarization. We perform abstractive summarization with pre-trained language models. We illustrate the architecture of our submission in Figure 1 (b): we first try to correct spell errors in the input; then summarize each question with three generative LMs: Pegasus, BART, and T5; finally, for each question, we pick the best output with a feature-based reranker and the best output is chosen as the summarization of the question.",
"cite_spans": [],
"ref_spans": [
{
"start": 211,
"end": 219,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 1315,
"end": 1323,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Task and Architecture Overview",
"sec_num": "2"
},
{
"text": "Misspellings are prevalent in medical forums, where non-expert users discuss highly specialized medical topics. Uncorrected misspellings can lead to mismatch between the source text and the summary during training and cause errors if copied verbatim during prediction. These errors are penalized heavily by string matching-based metrics like Rouge as they break n-grams.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Knowledge-based Error Correction",
"sec_num": "3"
},
{
"text": "In this shared task, we conservatively correct misspelled words in input by reusing a cascade of candidate generation modules from an entity linking system. Entity linking is the task to link entity mentions in text to entities in a knowledge base (KB). Candidate generation is an intermediate step in entity linking to generate candidate KB entities from potentially abbreviated, misspelled, or alias text mentions (see e.g. (Charton et al., 2014) ). Our method is also comparable to previous work on Levenshtein distance-based (Levenshtein, 1966) medical query correction (Soualmia et al., 2012), but we augment that approach with cascaded knowledge sources and an alias table.",
"cite_spans": [
{
"start": 426,
"end": 448,
"text": "(Charton et al., 2014)",
"ref_id": "BIBREF3"
},
{
"start": 529,
"end": 548,
"text": "(Levenshtein, 1966)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Knowledge-based Error Correction",
"sec_num": "3"
},
{
"text": "Error correction can be implemented easier and with possibly higher quality if search suggestions from online search engines (Zhou et al., 2015) are utilized. We use in-house error correction to keep the submission offline.",
"cite_spans": [
{
"start": 125,
"end": 144,
"text": "(Zhou et al., 2015)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Knowledge-based Error Correction",
"sec_num": "3"
},
{
"text": "The error correction module relies on the following resources:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Resources",
"sec_num": "3.1"
},
{
"text": "\u2022 Medical word list. We collect tokens from the English side of~20K bilingual medical phrases collected from dictionaries and drug names.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Resources",
"sec_num": "3.1"
},
{
"text": "\u2022 Wikipedia dump. We use a 20210101 dump of the English Wikipedia as the knowledge base and alias table.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Resources",
"sec_num": "3.1"
},
{
"text": "\u2022 High frequency word list. We use the top 10,000 words in the Google 1T corpus 1 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Resources",
"sec_num": "3.1"
},
{
"text": "We use Wikipedia instead of a medical KB because of its broad coverage. Edges (redirects, links etc.) in the Wikipedia KB can be used as an alias table to capture common misspellings and aliases. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Resources",
"sec_num": "3.1"
},
{
"text": "During error correction, we handle tokens composed entirely of alphabetical characters and allow at most 2 edits in similarity searches. We only consider tokens that share 3-prefix or 3-suffix with the query to limit search space.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "Error correction consists of the following steps:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Index construction. We build a token index of Wikipedia. We only index titles with no more than two tokens and tokens more than 5 characters long. We use the first token to represent the title. When a token can map to more than one titles, we map it to the title with the lowest id.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Spell checking. We pass the text through a spell checker with medical terms 2 to detect potential errors. The flagged tokens are the query words for the error correction pipeline.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Wikipedia match. If the query has an exact match in the Wikipedia token index, we link the query to the token and its corresponding Wikipedia title. Note that a title can either be an entity or an alias, which we resolve later in the name resolution step.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Medical word search. We search the medical word list to find medical terms that spell similarly to the query. We choose the medical term if a result is found.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Frequent word search. We search the high frequency word list to recall common English words that spell similarly to the query. We choose the word if a result is found.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Wikipedia search. We search the Wikipedia token index for queries longer than 5. To further constrain search space, we only consider tokens that share 5-prefix, 5-suffix, or all consonants with the query. We choose the token with the highest sequence matching ratio 3 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "\u2022 Name resolution. For corrected tokens retrieved from the medical word list and the Wikipedia, we search the Wikipedia dump to check if it is an alias of another entity and maps it to its canonical form.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "Consider the example in Figure 2 . Input queries of the error correction pipeline are the misspelled words identified by the spell checker. Wikipedia match catches the common misspelling *folliculitus and recovers its canonical form folliculitis. Medical word search recovers pigmentosum from the medical dictionary. Frequent word search recovers misspellings of popular words, avoid sending them to the noisy Wikipedia search. Finally, Wikipedia search first map *ureatha to its closest alias ureathra in Wikipedia and then maps ureathra to the canonical form urethra.",
"cite_spans": [],
"ref_spans": [
{
"start": 24,
"end": 32,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "On the validation set, the process is unable to recover the word *preagnet (pregnant). We are able to recover most other errors on the validation set. Impact of error correction is evaluated in Section 6.2.1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error correction steps",
"sec_num": "3.2"
},
{
"text": "Pre-trained conditional generative language models have become the dominating paradigm for text generation and especially summarization, with recent models such as Pegasus (Zhang et al., 2020a) , BART (Lewis et al., 2020) , T5 (Raffel et al., 2020) , and PALM (Bi et al., 2020) achieving stateof-the-art results on standard benchmarks CNN-Dailymail (See et al., 2017) and XSUM (Narayan et al., 2018) . Recent work has also shown that these models achieve good performance in few-shot medical summarization settings (Goodwin et al., 2020) . Following (Goodwin et al., 2020) , we use Pegasus, BART, and T5 single systems as our baselines.",
"cite_spans": [
{
"start": 172,
"end": 193,
"text": "(Zhang et al., 2020a)",
"ref_id": "BIBREF19"
},
{
"start": 201,
"end": 221,
"text": "(Lewis et al., 2020)",
"ref_id": "BIBREF9"
},
{
"start": 227,
"end": 248,
"text": "(Raffel et al., 2020)",
"ref_id": "BIBREF15"
},
{
"start": 260,
"end": 277,
"text": "(Bi et al., 2020)",
"ref_id": "BIBREF2"
},
{
"start": 349,
"end": 367,
"text": "(See et al., 2017)",
"ref_id": "BIBREF16"
},
{
"start": 377,
"end": 399,
"text": "(Narayan et al., 2018)",
"ref_id": "BIBREF13"
},
{
"start": 515,
"end": 537,
"text": "(Goodwin et al., 2020)",
"ref_id": "BIBREF5"
},
{
"start": 550,
"end": 572,
"text": "(Goodwin et al., 2020)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Summarization with Pre-trained Conditional Generative Language Models",
"sec_num": "4"
},
{
"text": "\u2022 Pegasus (Zhang et al., 2020a ) is a conditional language model designed specifically for abstractive summarization and is pretrained with a self-supervised gap-sentencegeneration objective, where the model is pretrained to predict entire masked sentences from the document.",
"cite_spans": [
{
"start": 10,
"end": 30,
"text": "(Zhang et al., 2020a",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Summarization with Pre-trained Conditional Generative Language Models",
"sec_num": "4"
},
{
"text": "\u2022 BART (Lewis et al., 2020) is a model combining bi-directional and auto-regressive transformers, trained to both denoise and reconstruct corrupted texts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarization with Pre-trained Conditional Generative Language Models",
"sec_num": "4"
},
{
"text": "\u2022 T5 (Raffel et al., 2020) is pre-trained on multiple objectives, including masking, translation, classification, machine reading comprehension (MRC) and summarization, all formulated as conditional generation tasks.",
"cite_spans": [
{
"start": 5,
"end": 26,
"text": "(Raffel et al., 2020)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Summarization with Pre-trained Conditional Generative Language Models",
"sec_num": "4"
},
{
"text": "We use Pegasus-large, BART-large, and T5-base respectively in our experiments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarization with Pre-trained Conditional Generative Language Models",
"sec_num": "4"
},
{
"text": "Following previous work on reranking generative LM outputs (Mi et al., 2021) , we pick the best summary for each question using the following linear model from outputs of three heterogeneous generative LMs,",
"cite_spans": [
{
"start": 59,
"end": 76,
"text": "(Mi et al., 2021)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "T * = argmax T i \u03c8 i (T, T , S)w i (1)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "where T is output of a single system, T is the set of outputs of all single systems, and S is the input text. T * is the ensemble output, which is picked from single system outputs by highest score. The feature function \u03c8(T, T , S) is a function to estimate the quality of T using information from T and S. w i is a weight of \u03c8(T, T , S). In sequence generation tasks such as machine translation (Kumar and Byrne, 2004), \u03c8 is usually a combination of consensus and linguistic features and w i can be tuned by optimization algorithms such as MERT (Och, 2003) towards an automatic evaluation metric.",
"cite_spans": [
{
"start": 546,
"end": 557,
"text": "(Och, 2003)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "Our approach. We use a simple and coverageoriented approach for reranking, based on the size and characteristics of the validation data. We notice that the writing style of the validation set is different from the MeQSum data set which we use for training: in MeQSum 18.5% sentences start with \"What are the treatments for\", 14.6% start with \"Where can I find\", and 2.5% start with \"What are the causes of \". A model trained on MeQSum tends to generate these topic-based boilerplates that are not mentioned in the source text. But in the validation set, summaries do not have these boilerplate texts and resemble the content of the source text more closely, which inspires us to pick the output with high coverage of the source.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "We consider the validation set (50 sentences) too small for automatic tuning, so we design a minimal set of features and set the weights w i manually.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "Features. We use fidelity, length, consensus and wellformedness features:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "\u2022 Fidelity (w f ). We calculate the Rouge-2 score between the input and the prediction. A higher score indicate a high-coverage summary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "\u2022 Length (w l ). The length ratio between the prediction and the input. \u2022 Consensus (w c ). 1 if T shares any bigram with T \u2212 T , 0 otherwise.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "\u2022 Wellformedness (w w ). 1 if T has less than three subsentences and starts with one question marker, 0 otherwise.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "For our experiments on the validation set and Rouge-2 experiments on the test set, we set w f = 1, w l = 0.01, w c = 10, w w = 10. The idea is to select the summary that has highest coverage of the source that is a one sentence question, with at least one bi-gram in common with other summaries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "The choice to favor high coverage summary is based on this particular pair of training and validation data, rather than general ensemble principles for text generation. We switch the weights for w f and w l for length reranking experiments on the test set. Impact of reranking is evaluated in Section 6.2.2.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Output Reranking",
"sec_num": "5"
},
{
"text": "Our systems are based on the Transformers (Wolf et al., 2020) package. We finetune baseline models on the MeQSum (Ben Abacha and Demner-Fushman, 2019) dataset for 50 epochs, with batch size 8 and learning rate 2e-5 with the AdamW optimizer on Nvidia P100 GPUs. Finetuning is indispensable for this task: without finetuning, BART-large scores 0.06 Rouge-2 and 0.15 Table 3 : Results on the test set. EC: Input error correction; R1/2/L: Rouge-1/2/L; P: Precision, R: Recall; Best team: Best score among all teams; Scores in bold when our system achieves the best score.",
"cite_spans": [
{
"start": 42,
"end": 61,
"text": "(Wolf et al., 2020)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [
{
"start": 364,
"end": 371,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental settings",
"sec_num": "6.1"
},
{
"text": "Rouge-L on the validation set in preliminary experiments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental settings",
"sec_num": "6.1"
},
{
"text": "For experiments on the test set, models for ensemble are further finetuned for 50 epochs on the validation set. Models for error-corrected input are finetuned on an automatically corrected version of the validation set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental settings",
"sec_num": "6.1"
},
{
"text": "We report single and reranking system performance in Tables 1 and 2 respectively. Results are evaluated by Rouge (Lin, 2004) , which is based on n-gram or longest common sequence (LCS) matching of strings.",
"cite_spans": [
{
"start": 113,
"end": 124,
"text": "(Lin, 2004)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [
{
"start": 53,
"end": 67,
"text": "Tables 1 and 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Validation set experiments",
"sec_num": "6.2"
},
{
"text": "Among the pre-trained LMs in Table 1 , BART performs the best on the validation set. Comparing error-corrected (Pegasus/BART/T5 EC) and original (Pegasus/BART/T5) inputs, we note that error-corrected input significantly boosts the performance of Pegasus. In addition to corrected entity names, the fixed input also leads Pegasus to generate 5% longer output and results in a much higher Rouge-2 score in this small dataset. This trend is less significant on BART and T5, but adding error correction has a positive impact in general.",
"cite_spans": [],
"ref_spans": [
{
"start": 29,
"end": 36,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Single systems and error correction",
"sec_num": "6.2.1"
},
{
"text": "We compare the reranked systems against baselines, with or without error-corrected input in Table 2 . In both cases, reranking does not have significant effect on Rouge-2, but helps Rouge-L significantly. We suspect that reranking does improve word and style choice, but the room for increasing 2-gram matches is small on the validation set.",
"cite_spans": [],
"ref_spans": [
{
"start": 92,
"end": 99,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Reranking",
"sec_num": "6.2.2"
},
{
"text": "We run three sets of experiments on the test set and report results in Table 3 : single systems are the same systems tested on the validation set and ensembles are reranked outputs from systems further finetuned on the validation set.",
"cite_spans": [],
"ref_spans": [
{
"start": 71,
"end": 78,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "In addition to string-based Rouge (Lin, 2004) , test set results are also evaluated by pre-trained language model-based BERTScore (Zhang et al., 2020b) and HOLMS (Mrabet and Demner-Fushman, 2020 ) metrics:",
"cite_spans": [
{
"start": 34,
"end": 45,
"text": "(Lin, 2004)",
"ref_id": "BIBREF10"
},
{
"start": 130,
"end": 151,
"text": "(Zhang et al., 2020b)",
"ref_id": "BIBREF20"
},
{
"start": 162,
"end": 194,
"text": "(Mrabet and Demner-Fushman, 2020",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "\u2022 BERTScore (Zhang et al., 2020b) leverages the pre-trained contextual embeddings from BERT and matches words in candidate and reference sentences by cosine similarity, where matching is performed greedily for each word by choosing the most similar word in the other sentence.",
"cite_spans": [
{
"start": 12,
"end": 33,
"text": "(Zhang et al., 2020b)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "\u2022 HOLMS (Mrabet and Demner-Fushman, 2020) combines soft matching of contextual embeddings derived from pre-trained LMs and a string-based metric (Rouge-1 recall in practice).",
"cite_spans": [
{
"start": 8,
"end": 41,
"text": "(Mrabet and Demner-Fushman, 2020)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "String-based and pre-trained language modelbased metrics rank summaries differently. We discuss the impact of the choice of metrics in Section 6.4.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "We run two other experiments validating postprocessing and the UniLM language model (Dong et al., 2019) , they perform inferior to their respective baselines and are not reported in Table 3 .",
"cite_spans": [
{
"start": 84,
"end": 103,
"text": "(Dong et al., 2019)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [
{
"start": 182,
"end": 189,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "We notice in single system experiments that the characteristics of the test set is still different from the validation set: all systems suffer from low recall, which leads us to perform more aggressive length-based reranking.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "Length reranking. We experiment with a baseline approach that explicitly picks the longest output sentence by switching the weight of length and fidelity features in (1). The 3 systems in runs 4 and 5 are Pegasus and T5 finetuned on the validation set and the Pegasus system in run 3. Run 6 adds BART finetuned on the validation set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "We observe that this simple heuristic, together with further finetuning on the validation set, leads to significantly higher Rouge scores between runs 3 and 4 in Table 3 . This change also improves HOLMS and BERTScore, suggesting that recall / coverage-based sentence selection does correlate to summarization quality in this scenario. Rouge is further improved by adding BART to the combination between runs 5 and 6.",
"cite_spans": [],
"ref_spans": [
{
"start": 162,
"end": 169,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "Correcting input errors between runs 4 and 5 also helps Rouge significantly. BERTScore, which is based on word matching and utilizes BERT embeddings, is much less sensitive to small spelling errors and changes negatively. HOLMS changes positively as it has a Rouge component.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "The negative change of BERTScore also suggests that we should be more cautious applying input error correction to summarization: mistakes in error correction might not hurt string-based metrics (the word is often misspelled already), but they can change the meaning of the sentence and degrade summarization quality.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "Coverage reranking. In runs 7 and 8, we experiment with the the same setting as in Table 2 . 3 systems are Pegasus, BART, and T5 finetuned on the validation set. These runs achieve balanced Rouge precision and recall, and the highest Rouge-2 score across all runs. There are small improvement on all metrics, which is expected, as Rouge-2 is a better indicator of summarization coverage than length.",
"cite_spans": [],
"ref_spans": [
{
"start": 83,
"end": 90,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "According to BERT-based metrics, coveragebased reranking also leads to more steady improvement than length-based reranking. The overall improvement in all metrics suggests that coveragebased reranking does improve summarization quality in this task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Test set experiments",
"sec_num": "6.3"
},
{
"text": "In this shared task, we experimented with knowledge-based input error correction and coverage-oriented system reranking. These methods are effective in boosting string matching between the prediction and the reference summaries. According to Rouge metrics, our submissiong ranks first according to Rouge-1/2 metrics and ranks second according to the Rouge-L metric.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lessons learned",
"sec_num": "6.4"
},
{
"text": "According to BERT-based metrics, however, reranking has a smaller impact on summarization quality and error correction has little to no effect: we are about 1 point below the best submission according to BERTScore and HOLMS, which are shown to often have higher correlation with human judgement (Zhang et al., 2020b; Mrabet and Demner-Fushman, 2020) .",
"cite_spans": [
{
"start": 295,
"end": 316,
"text": "(Zhang et al., 2020b;",
"ref_id": "BIBREF20"
},
{
"start": 317,
"end": 349,
"text": "Mrabet and Demner-Fushman, 2020)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Lessons learned",
"sec_num": "6.4"
},
{
"text": "The discrepancy between the string-based and LM-based metrics makes the real improvement of summarization quality hard to measure. It is arguable that by focusing on misspellings and using coverage as surrogate for summarization quality, we might be optimizing more for the writing style and spelling, rather than the content of the summary. This shows the need of an efficient, optimizable summarization evaluation metric with high correlation with human judgement that our field agrees upon. We plan to look more into the choice of metric and optimization objectives for summarzation tasks in future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lessons learned",
"sec_num": "6.4"
},
{
"text": "We reported our experiments in MEDIQA 2021 shared task 1. We used knowledge-based error correction and coverage-oriented reranking improve summarization. Our system performed well on string-based Rouge metrics, but less so on BERTbased semantic metrics. We plan to investigate methods that improve summarization according to human judgement.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
},
{
"text": "https://books.google.com/ngrams/info",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/glutanimate/ hunspell-en-med-glut 3 https://docs.python.org/3/library/ difflib.html",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "On the summarization of consumer health questions",
"authors": [
{
"first": "Asma",
"middle": [],
"last": "Ben Abacha",
"suffix": ""
},
{
"first": "Dina",
"middle": [],
"last": "Demner-Fushman",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "2228--2234",
"other_ids": {
"DOI": [
"10.18653/v1/P19-1215"
]
},
"num": null,
"urls": [],
"raw_text": "Asma Ben Abacha and Dina Demner-Fushman. 2019. On the summarization of consumer health questions. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2228-2234, Florence, Italy.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Overview of the mediqa 2021 shared task on summarization in the medical domain",
"authors": [
{
"first": "Asma",
"middle": [],
"last": "Ben Abacha",
"suffix": ""
},
{
"first": "Yassine",
"middle": [],
"last": "Mrabet",
"suffix": ""
},
{
"first": "Yuhao",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Chaitanya",
"middle": [],
"last": "Shivade",
"suffix": ""
},
{
"first": "Curtis",
"middle": [],
"last": "Langlotz",
"suffix": ""
},
{
"first": "Dina",
"middle": [],
"last": "Demner-Fushman",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the 20th SIG-BioMed Workshop on Biomedical Language Processing, NAACL-BioNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Asma Ben Abacha, Yassine Mrabet, Yuhao Zhang, Chaitanya Shivade, Curtis Langlotz, and Dina Demner-Fushman. 2021. Overview of the mediqa 2021 shared task on summarization in the med- ical domain. In Proceedings of the 20th SIG- BioMed Workshop on Biomedical Language Pro- cessing, NAACL-BioNLP 2021.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "PALM: Pre-training an autoencod-ing&autoregressive language model for contextconditioned generation",
"authors": [
{
"first": "Bin",
"middle": [],
"last": "Bi",
"suffix": ""
},
{
"first": "Chenliang",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Chen",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Ming",
"middle": [],
"last": "Yan",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Songfang",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Fei",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Luo",
"middle": [],
"last": "Si",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "8681--8691",
"other_ids": {
"DOI": [
"10.18653/v1/2020.emnlp-main.700"
]
},
"num": null,
"urls": [],
"raw_text": "Bin Bi, Chenliang Li, Chen Wu, Ming Yan, Wei Wang, Songfang Huang, Fei Huang, and Luo Si. 2020. PALM: Pre-training an autoencod- ing&autoregressive language model for context- conditioned generation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Lan- guage Processing (EMNLP), pages 8681-8691.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Improving entity linking using surface form refinement",
"authors": [
{
"first": "Eric",
"middle": [],
"last": "Charton",
"suffix": ""
},
{
"first": "Marie-Jean",
"middle": [],
"last": "Meurs",
"suffix": ""
},
{
"first": "Ludovic",
"middle": [],
"last": "Jean-Louis",
"suffix": ""
},
{
"first": "Michel",
"middle": [],
"last": "Gagnon",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)",
"volume": "",
"issue": "",
"pages": "4609--4615",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eric Charton, Marie-Jean Meurs, Ludovic Jean-Louis, and Michel Gagnon. 2014. Improving entity link- ing using surface form refinement. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14), pages 4609- 4615, Reykjavik, Iceland.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Unified language model pre-training for natural language understanding and generation",
"authors": [
{
"first": "Li",
"middle": [],
"last": "Dong",
"suffix": ""
},
{
"first": "Nan",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Wenhui",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Furu",
"middle": [],
"last": "Wei",
"suffix": ""
},
{
"first": "Xiaodong",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Yu",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Jianfeng",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "Ming",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "Hsiao-Wuen",
"middle": [],
"last": "Hon",
"suffix": ""
}
],
"year": 2019,
"venue": "33rd Conference on Neural Information Processing Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xi- aodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. 2019. Unified language model pre-training for natural language understand- ing and generation. In 33rd Conference on Neural Information Processing Systems (NeurIPS 2019).",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Flight of the PEGASUS? comparing transformers on few-shot and zero-shot multidocument abstractive summarization",
"authors": [
{
"first": "Travis",
"middle": [],
"last": "Goodwin",
"suffix": ""
},
{
"first": "Max",
"middle": [],
"last": "Savery",
"suffix": ""
},
{
"first": "Dina",
"middle": [],
"last": "Demner-Fushman",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 28th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "5640--5646",
"other_ids": {
"DOI": [
"10.18653/v1/2020.coling-main.494"
]
},
"num": null,
"urls": [],
"raw_text": "Travis Goodwin, Max Savery, and Dina Demner- Fushman. 2020. Flight of the PEGASUS? compar- ing transformers on few-shot and zero-shot multi- document abstractive summarization. In Proceed- ings of the 28th International Conference on Com- putational Linguistics, pages 5640-5646, Barcelona, Spain (Online). International Committee on Compu- tational Linguistics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Xiaozhong Liu, and Sheng Yu. 2021. Biomedical question answering: A comprehensive review",
"authors": [
{
"first": "Qiao",
"middle": [],
"last": "Jin",
"suffix": ""
},
{
"first": "Zheng",
"middle": [],
"last": "Yuan",
"suffix": ""
},
{
"first": "Guangzhi",
"middle": [],
"last": "Xiong",
"suffix": ""
},
{
"first": "Qianlan",
"middle": [],
"last": "Yu",
"suffix": ""
},
{
"first": "Chuanqi",
"middle": [],
"last": "Tan",
"suffix": ""
},
{
"first": "Mosha",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Songfang",
"middle": [],
"last": "Huang",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2102.05281"
]
},
"num": null,
"urls": [],
"raw_text": "Qiao Jin, Zheng Yuan, Guangzhi Xiong, Qianlan Yu, Chuanqi Tan, Mosha Chen, Songfang Huang, Xi- aozhong Liu, and Sheng Yu. 2021. Biomedical ques- tion answering: A comprehensive review. arXiv preprint arXiv:2102.05281.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Minimum bayes-risk decoding for statistical machine translation",
"authors": [
{
"first": "Shankar",
"middle": [],
"last": "Kumar",
"suffix": ""
},
{
"first": "William",
"middle": [],
"last": "Byrne",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of HLT-NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shankar Kumar and William Byrne. 2004. Minimum bayes-risk decoding for statistical machine transla- tion. In Proceedings of HLT-NAACL.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Binary codes capable of correcting deletions, insertions, and reversals",
"authors": [
{
"first": "",
"middle": [],
"last": "Vladimir I Levenshtein",
"suffix": ""
}
],
"year": 1966,
"venue": "Soviet physics doklady",
"volume": "10",
"issue": "",
"pages": "707--710",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vladimir I Levenshtein. 1966. Binary codes capable of correcting deletions, insertions, and reversals. In Soviet physics doklady, volume 10, pages 707-710.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "BART: Denoising sequence-to-sequence pretraining for natural language generation, translation, and comprehension",
"authors": [
{
"first": "Mike",
"middle": [],
"last": "Lewis",
"suffix": ""
},
{
"first": "Yinhan",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Naman",
"middle": [],
"last": "Goyal ; Abdelrahman Mohamed",
"suffix": ""
},
{
"first": "Omer",
"middle": [],
"last": "Levy",
"suffix": ""
},
{
"first": "Veselin",
"middle": [],
"last": "Stoyanov",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "7871--7880",
"other_ids": {
"DOI": [
"10.18653/v1/2020.acl-main.703"
]
},
"num": null,
"urls": [],
"raw_text": "Mike Lewis, Yinhan Liu, Naman Goyal, Mar- jan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre- training for natural language generation, translation, and comprehension. In Proceedings of the 58th An- nual Meeting of the Association for Computational Linguistics, pages 7871-7880, Online.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "ROUGE: A package for automatic evaluation of summaries",
"authors": [
{
"first": "Chin-Yew",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 2004,
"venue": "Text Summarization Branches Out",
"volume": "",
"issue": "",
"pages": "74--81",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chin-Yew Lin. 2004. ROUGE: A package for auto- matic evaluation of summaries. In Text Summariza- tion Branches Out, pages 74-81, Barcelona, Spain.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Towards generalized models for beyond domain api task-oriented dialogue",
"authors": [
{
"first": "Haitao",
"middle": [],
"last": "Mi",
"suffix": ""
},
{
"first": "Qiyu",
"middle": [],
"last": "Ren",
"suffix": ""
},
{
"first": "Yinpei",
"middle": [],
"last": "Dai",
"suffix": ""
},
{
"first": "Yifan",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Jian",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Yongbin",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Jing",
"middle": [],
"last": "Zheng",
"suffix": ""
},
{
"first": "Peng",
"middle": [],
"last": "Xu",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the 9th Dialog System Technology Challenge",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Haitao Mi, Qiyu Ren, Yinpei Dai, Yifan He, Jian Sun, Yongbin Li, Jing Zheng, and Peng Xu. 2021. To- wards generalized models for beyond domain api task-oriented dialogue. In Proceedings of the 9th Dialog System Technology Challenge.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "HOLMS: Alternative summary evaluation with large language models",
"authors": [
{
"first": "Yassine",
"middle": [],
"last": "Mrabet",
"suffix": ""
},
{
"first": "Dina",
"middle": [],
"last": "Demner-Fushman",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 28th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "5679--5688",
"other_ids": {
"DOI": [
"10.18653/v1/2020.coling-main.498"
]
},
"num": null,
"urls": [],
"raw_text": "Yassine Mrabet and Dina Demner-Fushman. 2020. HOLMS: Alternative summary evaluation with large language models. In Proceedings of the 28th Inter- national Conference on Computational Linguistics, pages 5679-5688, Barcelona, Spain (Online).",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Don't give me the details, just the summary! Topic-aware convolutional neural networks for extreme summarization",
"authors": [
{
"first": "Shashi",
"middle": [],
"last": "Narayan",
"suffix": ""
},
{
"first": "Shay",
"middle": [
"B"
],
"last": "Cohen",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shashi Narayan, Shay B. Cohen, and Mirella Lapata. 2018. Don't give me the details, just the summary! Topic-aware convolutional neural networks for ex- treme summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Lan- guage Processing, Brussels, Belgium.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Minimum error rate training in statistical machine translation",
"authors": [
{
"first": "Franz Josef",
"middle": [],
"last": "Och",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "160--167",
"other_ids": {
"DOI": [
"10.3115/1075096.1075117"
]
},
"num": null,
"urls": [],
"raw_text": "Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proceedings of the 41st Annual Meeting of the Association for Computa- tional Linguistics, pages 160-167, Sapporo, Japan.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Exploring the limits of transfer learning with a unified text-totext transformer",
"authors": [
{
"first": "Colin",
"middle": [],
"last": "Raffel",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Shazeer",
"suffix": ""
},
{
"first": "Adam",
"middle": [],
"last": "Roberts",
"suffix": ""
},
{
"first": "Katherine",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Sharan",
"middle": [],
"last": "Narang",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Matena",
"suffix": ""
},
{
"first": "Yanqi",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Peter",
"middle": [
"J"
],
"last": "Liu",
"suffix": ""
}
],
"year": 2020,
"venue": "Journal of Machine Learning Research",
"volume": "21",
"issue": "140",
"pages": "1--67",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Colin Raffel, Noam Shazeer, Adam Roberts, Kather- ine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to- text transformer. Journal of Machine Learning Re- search, 21(140):1-67.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Get to the point: Summarization with pointergenerator networks",
"authors": [
{
"first": "Abigail",
"middle": [],
"last": "See",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Peter",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Liu",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Abigail See, Peter J. Liu, and Christopher D. Manning. 2017. Get to the point: Summarization with pointer- generator networks. CoRR, abs/1704.04368.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Matching health information seekers' queries to medical terms",
"authors": [
{
"first": "F",
"middle": [],
"last": "Lina",
"suffix": ""
},
{
"first": "Elise",
"middle": [],
"last": "Soualmia",
"suffix": ""
},
{
"first": "Zied",
"middle": [],
"last": "Prieur-Gaston",
"suffix": ""
},
{
"first": "Thierry",
"middle": [],
"last": "Moalla",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Lecroq",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "St\u00e9fan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Darmoni",
"suffix": ""
}
],
"year": 2012,
"venue": "BMC bioinformatics",
"volume": "13",
"issue": "14",
"pages": "1--15",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lina F Soualmia, Elise Prieur-Gaston, Zied Moalla, Thierry Lecroq, and St\u00e9fan J Darmoni. 2012. Match- ing health information seekers' queries to medical terms. BMC bioinformatics, 13(14):1-15.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Transformers: State-of-the-art natural language processing",
"authors": [
{
"first": "Thomas",
"middle": [],
"last": "Wolf",
"suffix": ""
},
{
"first": "Lysandre",
"middle": [],
"last": "Debut",
"suffix": ""
},
{
"first": "Victor",
"middle": [],
"last": "Sanh",
"suffix": ""
},
{
"first": "Julien",
"middle": [],
"last": "Chaumond",
"suffix": ""
},
{
"first": "Clement",
"middle": [],
"last": "Delangue",
"suffix": ""
},
{
"first": "Anthony",
"middle": [],
"last": "Moi",
"suffix": ""
},
{
"first": "Pierric",
"middle": [],
"last": "Cistac",
"suffix": ""
},
{
"first": "Tim",
"middle": [],
"last": "Rault",
"suffix": ""
},
{
"first": "Remi",
"middle": [],
"last": "Louf",
"suffix": ""
},
{
"first": "Morgan",
"middle": [],
"last": "Funtowicz",
"suffix": ""
},
{
"first": "Joe",
"middle": [],
"last": "Davison",
"suffix": ""
},
{
"first": "Sam",
"middle": [],
"last": "Shleifer",
"suffix": ""
},
{
"first": "Clara",
"middle": [],
"last": "Patrick Von Platen",
"suffix": ""
},
{
"first": "Yacine",
"middle": [],
"last": "Ma",
"suffix": ""
},
{
"first": "Julien",
"middle": [],
"last": "Jernite",
"suffix": ""
},
{
"first": "Canwen",
"middle": [],
"last": "Plu",
"suffix": ""
},
{
"first": "Teven",
"middle": [
"Le"
],
"last": "Xu",
"suffix": ""
},
{
"first": "Sylvain",
"middle": [],
"last": "Scao",
"suffix": ""
},
{
"first": "Mariama",
"middle": [],
"last": "Gugger",
"suffix": ""
},
{
"first": "Quentin",
"middle": [],
"last": "Drame",
"suffix": ""
},
{
"first": "Alexander",
"middle": [],
"last": "Lhoest",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Rush",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
"volume": "",
"issue": "",
"pages": "38--45",
"other_ids": {
"DOI": [
"10.18653/v1/2020.emnlp-demos.6"
]
},
"num": null,
"urls": [],
"raw_text": "Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier- ric Cistac, Tim Rault, Remi Louf, Morgan Funtow- icz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Trans- formers: State-of-the-art natural language process- ing. In Proceedings of the 2020 Conference on Em- pirical Methods in Natural Language Processing: System Demonstrations, pages 38-45, Online.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "PEGASUS: Pre-training with extracted gap-sentences for abstractive summarization",
"authors": [
{
"first": "Jingqing",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Yao",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "Mohammad",
"middle": [],
"last": "Saleh",
"suffix": ""
},
{
"first": "Peter",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 37th International Conference on Machine Learning",
"volume": "",
"issue": "",
"pages": "11328--11339",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Pe- ter Liu. 2020a. PEGASUS: Pre-training with ex- tracted gap-sentences for abstractive summarization. In Proceedings of the 37th International Conference on Machine Learning, pages 11328-11339.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Bertscore: Evaluating text generation with bert",
"authors": [
{
"first": "Tianyi",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Varsha",
"middle": [],
"last": "Kishore",
"suffix": ""
},
{
"first": "Felix",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Kilian",
"middle": [
"Q"
],
"last": "Weinberger",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Artzi",
"suffix": ""
}
],
"year": 2020,
"venue": "International Conference on Learning Representations",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020b. Bertscore: Evaluating text generation with bert. In Interna- tional Conference on Learning Representations.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Context-sensitive spelling correction of consumer-generated content on health care",
"authors": [
{
"first": "X",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "An",
"middle": [],
"last": "Zheng",
"suffix": ""
},
{
"first": "Jiaheng",
"middle": [],
"last": "Yin",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Xianyang",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Xu",
"suffix": ""
},
{
"first": "Wenqing",
"middle": [],
"last": "Cheng",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Xia",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 2015,
"venue": "JMIR Medical Informatics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "X. Zhou, An Zheng, Jiaheng Yin, R. Chen, Xi- anyang Zhao, Wei Xu, Wenqing Cheng, T. Xia, and S. Lin. 2015. Context-sensitive spelling correction of consumer-generated content on health care. JMIR Medical Informatics.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Question-summarization example and system architecture"
},
"FIGREF1": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Example of error correction"
},
"TABREF3": {
"text": "",
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null
}
}
}
} |