File size: 110,080 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 |
{
"paper_id": "I17-1030",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:38:30.334615Z"
},
"title": "Learning How to Simplify From Explicit Labeling of Complex-Simplified Text Pairs",
"authors": [
{
"first": "Fernando",
"middle": [],
"last": "Alva-Manchego",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {
"country": "UK"
}
},
"email": ""
},
{
"first": "Joachim",
"middle": [],
"last": "Bingel",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Copenhagen",
"location": {
"country": "Denmark"
}
},
"email": "bingel@di.ku.dk"
},
{
"first": "Gustavo",
"middle": [
"H"
],
"last": "Paetzold",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {
"country": "UK"
}
},
"email": "g.h.paetzold@sheffield.ac.uk"
},
{
"first": "Carolina",
"middle": [],
"last": "Scarton",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {
"country": "UK"
}
},
"email": "c.scarton@sheffield.ac.uk"
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {
"country": "UK"
}
},
"email": "l.specia@sheffield.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Current research in text simplification has been hampered by two central problems: (i) the small amount of high-quality parallel simplification data available, and (ii) the lack of explicit annotations of simplification operations, such as deletions or substitutions, on existing data. While the recently introduced Newsela corpus has alleviated the first problem, simplifications still need to be learned directly from parallel text using black-box, end-to-end approaches rather than from explicit annotations. These complex-simple parallel sentence pairs often differ to such a high degree that generalization becomes difficult. End-to-end models also make it hard to interpret what is actually learned from data. We propose a method that decomposes the task of TS into its sub-problems. We devise a way to automatically identify operations in a parallel corpus and introduce a sequence-labeling approach based on these annotations. Finally, we provide insights on the types of transformations that different approaches can model.",
"pdf_parse": {
"paper_id": "I17-1030",
"_pdf_hash": "",
"abstract": [
{
"text": "Current research in text simplification has been hampered by two central problems: (i) the small amount of high-quality parallel simplification data available, and (ii) the lack of explicit annotations of simplification operations, such as deletions or substitutions, on existing data. While the recently introduced Newsela corpus has alleviated the first problem, simplifications still need to be learned directly from parallel text using black-box, end-to-end approaches rather than from explicit annotations. These complex-simple parallel sentence pairs often differ to such a high degree that generalization becomes difficult. End-to-end models also make it hard to interpret what is actually learned from data. We propose a method that decomposes the task of TS into its sub-problems. We devise a way to automatically identify operations in a parallel corpus and introduce a sequence-labeling approach based on these annotations. Finally, we provide insights on the types of transformations that different approaches can model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Text Simplification (TS) is the task of reducing the complexity of a text without changing its meaning. Simplification can be applied at various linguistic levels, from lexical substitution to more global operations such as sentence splitting, paraphrasing or the deletion or reordering of entire clauses.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Existing corpora for TS generally come in one of two variants. The first focuses on very specific sub-problems, such as sentence compression FA and JB contributed equally to this paper. (Bingel and S\u00f8gaard, 2016) or the identification of difficult words (Paetzold and Specia, 2016a) , and typically encodes relevant simplification operations as discrete labels on tokens. The other variant includes more general, higher-level types of simplifications that often entail the rephrasing or re-structuring of sentences, with content added or removed. These \"natural\" simplifications are often created for end-users rather than for research purposes. Examples of the latter simplification resources include the Newsela (Xu et al., 2015) and Simple English Wikipedia corpora (Zhu et al., 2010; Coster and Kauchak, 2011b) . These resources generally encode interdependencies between different types of simplification better than single-purpose resources and may thus seem favorable for learning simplifications. However, the freedom given to editors and lack of explicit labels on the modifications performed makes generalization much more difficult, especially when existing resources are relatively small in comparison to corpora for other text-to-text problems like machine translation (MT). Nevertheless, these corpora have been extensively used to learn phrasebased statistical and neural models for end-to-end TS systems that bear resemblance to MT models (Specia, 2010; Zhu et al., 2010; Coster and Kauchak, 2011b; Wubben et al., 2012; Narayan and Gardent, 2014; Xu et al., 2016; Zhang and Lapata, 2017; Nisioi et al., 2017) .",
"cite_spans": [
{
"start": 186,
"end": 212,
"text": "(Bingel and S\u00f8gaard, 2016)",
"ref_id": "BIBREF2"
},
{
"start": 254,
"end": 282,
"text": "(Paetzold and Specia, 2016a)",
"ref_id": "BIBREF12"
},
{
"start": 714,
"end": 731,
"text": "(Xu et al., 2015)",
"ref_id": "BIBREF24"
},
{
"start": 769,
"end": 787,
"text": "(Zhu et al., 2010;",
"ref_id": "BIBREF28"
},
{
"start": 788,
"end": 814,
"text": "Coster and Kauchak, 2011b)",
"ref_id": "BIBREF6"
},
{
"start": 1455,
"end": 1469,
"text": "(Specia, 2010;",
"ref_id": "BIBREF20"
},
{
"start": 1470,
"end": 1487,
"text": "Zhu et al., 2010;",
"ref_id": "BIBREF28"
},
{
"start": 1488,
"end": 1514,
"text": "Coster and Kauchak, 2011b;",
"ref_id": "BIBREF6"
},
{
"start": 1515,
"end": 1535,
"text": "Wubben et al., 2012;",
"ref_id": "BIBREF23"
},
{
"start": 1536,
"end": 1562,
"text": "Narayan and Gardent, 2014;",
"ref_id": "BIBREF10"
},
{
"start": 1563,
"end": 1579,
"text": "Xu et al., 2016;",
"ref_id": "BIBREF25"
},
{
"start": 1580,
"end": 1603,
"text": "Zhang and Lapata, 2017;",
"ref_id": "BIBREF26"
},
{
"start": 1604,
"end": 1624,
"text": "Nisioi et al., 2017)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Adaptability and interpretability MT-style models are essentially black boxes that offer little or no control over the way in which a given input is modified. Additionally, in most cases the types of modifications that are actually learned are limited to paraphrasing of short sequences of words. We believe a middle ground is missing in terms of resources and approaches for TS, where models are learned from a more informed labeled dataset of natural simplifications, and can then be applied in a controlled way, e.g., in adaptive simplification scenarios that prioritize different ways of simplifying (e.g. compression or sentence splitting) depending on a particular user's needs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The only previous work on TS via explicitly predicting simplification operations is that by Bingel and S\u00f8gaard (2016) , who create training data from comparable text to label entire syntactic units and train a sequence labeling model to predict deletions and phrase substitutions in a complex sentence. Our approach is different in that it captures a larger variety of operations in a more global fashion, by using sentence-wide word alignments rather than surface heuristics. Furthermore, we use a more reliable (professionally created) corpus and our approach is more flexible as we do not rely on syntactic parse trees at test time.",
"cite_spans": [
{
"start": 92,
"end": 117,
"text": "Bingel and S\u00f8gaard (2016)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Contributions This paper introduces the following main contributions: (1) We provide an in-depth analysis on the potential and limitations of the dominant approach to TS: end-to-end MT-style models;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "(2) We devise a method to automatically identify specific simplification operations in aligned sentences from complexto-simple simplification corpora. This results in a corpus that can be used to study how human experts perform simplification tasks, as well as to train simplification models to address specific problems; and (3) We propose a sequence labeling model built from such a corpus to predict which simplification operations should be performed as a first step for a complete simplification pipeline. This approach is highly modular: once operations are identified, different methods can be applied to cover each simplification operation. We show that this operation-based TS approach is able to produce simpler texts than end-to-end models. The code for extracting the simplification operations is available at https:// github.com/ghpaetzold/massalign, while our sequence labeling model is released at https://github.com/jbingel/ ijcnlp2017_simplification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In what follows we give a brief description of previous work on statistical and neural models for TS. We first compare methods using versions of Simple English Wikipedia data (Zhu et al., 2010; Coster and Kauchak, 2011b) , before considering recent work that relies on the professionally edited Newsela corpus (Xu et al., 2015) . Zhu et al. (2010) propose a syntax-based translation model for TS that learns operations over the parse trees of the complex sentences. They outperform several baselines in terms of Flesch index. Coster and Kauchak (2011b) train a phrase-based machine translation (PBMT) system and obtain significant improvements in terms of BLEU (Papineni et al., 2002) over a baseline. Coster and Kauchak (2011a) extend a PBMT model to include phrase deletion and outperform Coster and Kauchak (2011b) . Wubben et al. (2012) also train a PBMT system for TS with a dissimilarity-based re-ranking heuristic, outperforming Zhu et al. (2010) in terms of BLEU. Narayan and Gardent (2014) built TS systems by combining discourse representation structures with a PBMT model, which outperforms previous approaches. Xu et al. (2016) modify a syntax-based MT system in order to use a new metric -SARI -for optimization and to include special rules for paraphrasing. Although their system does not outperform previous work in terms of BLEU, it achieves the best results according to SARI and human evaluation. Zhang et al. (2017) train a lexically constrained sequenceto-sequence neural network model for TS, based on the encoder-decoder architecture for MT. The system outperforms baseline systems (including a PBMT system) in terms of BLEU. Finally, Nisioi et al. (2017) propose a model for TS that is able to perform lexical replacements and content reduction. They use a neural encoder-decoder approach where they combine pre-trained (general domain and in-domain) word embeddings for the source and target sentences. They also perform beam search, finding the best beam size using either BLEU or SARI. Their best model outperforms previous PBMT-based approaches in terms of BLEU.",
"cite_spans": [
{
"start": 175,
"end": 193,
"text": "(Zhu et al., 2010;",
"ref_id": "BIBREF28"
},
{
"start": 194,
"end": 220,
"text": "Coster and Kauchak, 2011b)",
"ref_id": "BIBREF6"
},
{
"start": 310,
"end": 327,
"text": "(Xu et al., 2015)",
"ref_id": "BIBREF24"
},
{
"start": 330,
"end": 347,
"text": "Zhu et al. (2010)",
"ref_id": "BIBREF28"
},
{
"start": 661,
"end": 684,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF15"
},
{
"start": 702,
"end": 728,
"text": "Coster and Kauchak (2011a)",
"ref_id": "BIBREF5"
},
{
"start": 791,
"end": 817,
"text": "Coster and Kauchak (2011b)",
"ref_id": "BIBREF6"
},
{
"start": 820,
"end": 840,
"text": "Wubben et al. (2012)",
"ref_id": "BIBREF23"
},
{
"start": 936,
"end": 953,
"text": "Zhu et al. (2010)",
"ref_id": "BIBREF28"
},
{
"start": 972,
"end": 998,
"text": "Narayan and Gardent (2014)",
"ref_id": "BIBREF10"
},
{
"start": 1123,
"end": 1139,
"text": "Xu et al. (2016)",
"ref_id": "BIBREF25"
},
{
"start": 1657,
"end": 1677,
"text": "Nisioi et al. (2017)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Newsela corpus To the best of our knowledge, Zhang and Lapata (2017) is the only work that explores MT-based approaches on the Newsela corpus. They train an attention-based encoderdecoder model (Bahdanau et al., 2014) and use reinforcement learning with a reward policy combining SARI, BLEU and cosine similarity (to measure meaning preservation). Their approach shows improvements over a PBMT system in terms of BLEU and SARI, but no insights are given with respect to the transformations that are actually learned or how distant from the original sentences the simplifications are. They also experiment with the Simple Wikipedia corpus, yet do not outperform Narayan and Gardent (2014) on this data.",
"cite_spans": [
{
"start": 45,
"end": 68,
"text": "Zhang and Lapata (2017)",
"ref_id": "BIBREF26"
},
{
"start": 194,
"end": 217,
"text": "(Bahdanau et al., 2014)",
"ref_id": "BIBREF0"
},
{
"start": 661,
"end": 687,
"text": "Narayan and Gardent (2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Simple English Wikipedia",
"sec_num": null
},
{
"text": "The neural end-to-end model we implement as a baseline in this paper is equivalent to that in without the lexical constraints, while the statistical model is equivalent to the one in Coster and Kauchak (2011b) .",
"cite_spans": [
{
"start": 183,
"end": 209,
"text": "Coster and Kauchak (2011b)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Simple English Wikipedia",
"sec_num": null
},
{
"text": "In addition to requiring large amounts of training data, MT-based approaches to TS are limited because of their black-box way of addressing the problem. As we are going to show in this section, standard end-to-end systems without special adaptation to TS do not succeed in learning alternative formulations of the original text. With a few exceptions (by the neural model), they tend to repeat the original text. We conjecture that this is because, for most original-side material, TS corpora do not consistently enough offer alternative simplified formulations: in the majority of instances, most words are kept as in the original.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "To study the potential and limitations of end-toend translation models for TS, we build models using state-of-the-art MT-based approaches and the Newsela corpus, arguably the most reliable (professionally created) and realistic (aimed at a target audience rather than research) resource to date.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "The Newsela Corpus. 1 Newsela is a multicomparable corpus where each document comes in up to six levels of simplicity, from 0 (original) to 5 (simplest). In our experiments, we only use sentence pairs stemming from adjacent levels of simplicity within the same document. 2 Translation approaches require data aligned at the sentence level. Given the original Newsela 1 The Newsela Article Corpus was downloaded from https://newsela.com/data, version 2016-01-29. 2 The motivations for only using adjacent levels are (i) that we assume that these are not \"naturally\" created (i.e. an expert would not start from an original text and directly generate a level 5 text, but rather go from 0 to 1, 1 to 2, ..., 4 to 5), and (ii) that the high degree of linguistic and stylistic differences between non-adjacent levels makes learning even more complex. For example, the average edit distance for sentences in the 0-1 group is 0.19, while for sentences in the 0-5 group, it is 0.65. As far as the first reason is concerned, note that we could not find any publicly available simplification guidelines for the Newsela corpus. corpus, which only aligns different versions of the same document, we first align sentences using the algorithms described in (Paetzold and Specia, 2016b) . Their algorithms search for the best alignment path between the paragraphs and sentences of parallel documents based on TF-IDF cosine similarity and an incremental vicinity search range. They address limitations of previous strategies (Barzilay and Elhadad, 2003; Coster and Kauchak, 2011b; Smith et al., 2010; Xu et al., 2015; Bott and Saggion, 2011) by disregarding the need for (semi-) supervised training, allowing long-distance alignment skips, and capturing N-to-N alignments. The alignments produced are categorized as:",
"cite_spans": [
{
"start": 271,
"end": 272,
"text": "2",
"ref_id": null
},
{
"start": 462,
"end": 463,
"text": "2",
"ref_id": null
},
{
"start": 1243,
"end": 1271,
"text": "(Paetzold and Specia, 2016b)",
"ref_id": "BIBREF13"
},
{
"start": 1509,
"end": 1537,
"text": "(Barzilay and Elhadad, 2003;",
"ref_id": "BIBREF1"
},
{
"start": 1538,
"end": 1564,
"text": "Coster and Kauchak, 2011b;",
"ref_id": "BIBREF6"
},
{
"start": 1565,
"end": 1584,
"text": "Smith et al., 2010;",
"ref_id": "BIBREF18"
},
{
"start": 1585,
"end": 1601,
"text": "Xu et al., 2015;",
"ref_id": "BIBREF24"
},
{
"start": 1602,
"end": 1625,
"text": "Bott and Saggion, 2011)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "\u2022 Identical:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "The alignment is one-to-one and the sentences are exactly the same (96,909 pairs across all adjacent levels).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "\u2022 1-to-1: The alignment is one-to-one and the original-simplified sentences are different (130,790 pairs across all adjacent levels).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "\u2022 Split: The alignment is 1-to-N (42,545 pairs across all adjacent levels).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "\u2022 Join: The alignment is N-to-1 (7,962 pairs across all adjacent levels).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "Translation Models. We built two types of models using state-of-the-art MT-based approaches: a phrase-based statistical MT model using Moses (Koehn et al., 2007) , 3 and a neural MT model using Nematus (Sennrich et al., 2017) . 4 The Neural Text Simplification tool (NTS) made available by Nisioi et al. (2017) was also used for comparison. 5 For our translation-based experiments, we consider two combinations of sentence alignments, using (i) only one-to-one alignments (1-to-1) (130,970 sentence pairs), and (ii) all alignments (all), i.e., the entire sentence-aligned corpus with identical, 1-to-1, split and join alignments (278,206 sentence pairs). The first type of data (1to-1) is the focus of this paper (see \u00a74). The latter variant is included in the experiments for comparison, in particular to address the question whether more (but not necessarily better) data can aid dataintensive translation-based approaches. For all Simplification Quality. The first and second sections of Table 1 show the results of translationbased systems according to several metrics: similarity metrics commonly used in MT, comprising BLEU (Papineni et al., 2002) and TER (Snover et al., 2006 , minimum edit distance), as well a specific text simplification metric, SARI (Xu et al., 2016) . SARI measures how good the words added, deleted and kept by a simplification system are, after comparing the produced output to the original sentence and the simplification reference(s). It is similar to BLEU but rewards copying words from the original sentence. According to experiments performed by Xu et al. (2016) , SARI is the metric that best correlates with human judgments of simplicity.",
"cite_spans": [
{
"start": 141,
"end": 161,
"text": "(Koehn et al., 2007)",
"ref_id": "BIBREF9"
},
{
"start": 164,
"end": 165,
"text": "3",
"ref_id": null
},
{
"start": 202,
"end": 225,
"text": "(Sennrich et al., 2017)",
"ref_id": "BIBREF16"
},
{
"start": 228,
"end": 229,
"text": "4",
"ref_id": null
},
{
"start": 290,
"end": 310,
"text": "Nisioi et al. (2017)",
"ref_id": "BIBREF11"
},
{
"start": 341,
"end": 342,
"text": "5",
"ref_id": null
},
{
"start": 1130,
"end": 1153,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF15"
},
{
"start": 1162,
"end": 1182,
"text": "(Snover et al., 2006",
"ref_id": "BIBREF19"
},
{
"start": 1261,
"end": 1278,
"text": "(Xu et al., 2016)",
"ref_id": "BIBREF25"
},
{
"start": 1582,
"end": 1598,
"text": "Xu et al. (2016)",
"ref_id": "BIBREF25"
}
],
"ref_spans": [
{
"start": 991,
"end": 998,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "For both \"all\" and \"1-to-1\" variants, the BLEU and TER scores between hypotheses and references are worse for Nematus, showing that a baseline neural model tends to be more aggressive and potentially generate noisier modifications than Moses equivalents. To measure how strongly the various approaches modify the input sentences, these scores are also reported between the generated simplifications and the original inputs. Again, these metrics are worse for Nematus-based models, showing that they indeed perform more modifications on the sentences. Moses in turn is very conservative, keeping 90-93% of the test sentences exactly in their original version. SARI shows low scores for all systems. NTS is also conservative in the \"all\" variant (attested by the high BLEU score between hypotheses and original sentences). For \"1-to-1\", NTS produces more simplifications, diverging more from the original sentences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "Sentence-level Operations. Interestingly, even though Moses and Nematus are trained on the same data, they differ substantially with respect to what they can learn. This is demonstrated by an automatic inspection we conducted on the simplifications produced by both systems trained over all types of sentence alignments, i.e. including sentence splits and joins. Table 2 reports the count and proportion of instances in the test set representing types of sentence-level transformation between the original and simplified sentence. It can be noted that Moses is much more conservative than Nematus and simply tends to copy the original as the output (\"Identical\" cases). However, as the majority (57%) of aligned sentences in the professional Newsela simplifications are edited, we do not consider copying a valid \"simplification\" in most cases. Note also that Moses displays an excessively high BLEU score between the original and hypothesis sentences (98.77), while the similarity between the original and reference sentences is much lower (71.57).",
"cite_spans": [],
"ref_spans": [
{
"start": 363,
"end": 370,
"text": "Table 2",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "Manually inspecting some of the simplifications made, we find that when it comes to sentence splits, both MT-based simplifiers seem to be able to perform this type of transformation in an accu- rate way. However, the proportion of such cases is very low (0.05% and 1.27% for Moses and Nematus, respectively) compared to the proportion in the gold data (13.5%) of the sentence pairs contain at least one split. Moses only joins sentences in four cases, but these are all spurious instances where a period is incorrectly removed. Nematus is more successful at learning this type of operation. In most cases, it discards entire clauses that contain less relevant content. For example, it simplifies the sentence \"Lincoln often cried in public and recited sad poetry, according to Joshusa Wolf Shenk, who wrote a book called Lincoln's Melancholy\" to \"Lincoln often cried in public and recited sad poetry\". We also find a few examples where the content that is not discarded is rewritten to some extent, mostly for grammaticality. The Nematus simplification of \"Frank was what the instructors called a 'rock star'; he emerged as a leader who worked hard to keep the group together\" onto \"Frank was a leader who worked hard to keep the group together \" is a good example of that.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "When it comes to 1-to-1 transformations, which can include a number of different operations (see \u00a74), most transformations made by Nematus consist of segment deletions, some of which are paired with localized segment rewritings. As for Moses, most 1-to-1 outputs are identical to the original except for a few spurious typographic and punctuation changes. Because of that, Nematus simplifications are in average four tokens shorter than both complex originals and Moses simplifications.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "A strong limitation of both models is their inability to address lexical complexity, performing very few lexical replacements. Most of the sentences that are lexically simplified have only one word replaced by another that does not preserve its original meaning. Take, for example, the word clears in the sentence \"It clears the way for troops on the ground with its huge bullets\", which was replaced by gathers by Nematus, and the word agribusiness, which was replaced by offering by Moses in sentence \"Older brother Nate has taken college courses on livestock raising and agribusiness\". Some of these issues become more evident in the human evaluation we performed comparing both end-to-end systems to our proposed approach ( \u00a75.2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via End-to-End Models",
"sec_num": "3"
},
{
"text": "Our approach to TS differs from translation-based models by explicitly predicting a set of operations to be applied at different positions in a complex sentence. Concretely, we tackle simplification as a sequence labeling problem, predicting operations at the token level and applying them downstream. As there are no high-quality and largescale resources from which such operation sequences could be learned, we first generate training data as explained below. 6",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Simplification via Sequence Labeling",
"sec_num": "4"
},
{
"text": "Given 1-to-1 sentence pairs, our method for data generation identifies deletions, additions, substitutions, rewrites (replacing or adding non-content words), and reorderings performed between sentences pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generating Training Data",
"sec_num": "4.1"
},
{
"text": "Automatic operation annotation. The annotation process uses the following set of operation labels: DELETE (D), REPLACE (R), and MOVE (M) in the original (source) sentence; ADD (A) in the simplified sentence; and REWRITE (RW) in both. 7 We first generate word alignments between the original and simplified sentences using the aligner by Sultan et al. (2014) . Based on these alignments, we perform a word-level annotation for labels DELETE and REPLACE. Our heuristics are that if two words are aligned and are not an exact match, then the corresponding label is REPLACE. If a word in the original sentence is not aligned, it must be a DELETE, and if a word in the simplified sentence is not aligned, it is an ADD. In any other case, the word receives label C (COPY) or O (not part of a simplification operation) in the original or simplified sentence, respectively. For details, see Algorithm 1 in the supplementary material. Figure 1 presents an example for our automatic labeling approach. We consider REWRITE labels as special cases of REPLACE where the words involved are isolated (not in a group of same operation labels) and belong to a list of non-content words.",
"cite_spans": [
{
"start": 234,
"end": 235,
"text": "7",
"ref_id": null
},
{
"start": 337,
"end": 357,
"text": "Sultan et al. (2014)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [
{
"start": 926,
"end": 935,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Generating Training Data",
"sec_num": "4.1"
},
{
"text": "Finally, we label reorderings (MOVE) by determining if the relative index of a word (considering preceding or following deletions and additions) in the original sentence changes in the simplified one (Algorithm 2). See Figure 2 for an example. Words or phrases that are kept, replaced or rewritten, may be subject to reorderings, such that a token may have more than one label (e.g. REPLACE and MOVE). For that, we extend the set of operations by the compound operations REPLACE+MOVE (RM) and REWRITE+MOVE (RWM).",
"cite_spans": [],
"ref_spans": [
{
"start": 219,
"end": 227,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Generating Training Data",
"sec_num": "4.1"
},
{
"text": "Evaluation of automatic labels. To test our algorithms, we compare their output to manual annotations for 100 sentences from level pair 0-1 of the Newsela corpus. The manual annotations were performed by four proficient English speakers. For 30 of those sentences, we calculated the pairwise inter-annotator agreement between annotators, yielding an average kappa value of 0.57. We obtain an accuracy of 0.92 for all labels, and a micro-averaged F 1 score of 0.70 for all positive labels (i.e. excluding 'C' and 'O'). Table 3 presents details on the performance of our annotation algorithms over the identified operations. Of the positive labels, the algorithms annotate most accurately additions and deletions. According to the confusion matrix in Table 4 , the relatively low ability of capturing replacements is due to labeling them as deletions. This is mainly caused by word miss-alignments and by parser errors that our heuristics cannot recover from. The same logic applies for labels REPLACE+MOVE and REWRITE+MOVE. We are also able to capture most MOVEments (high recall), but our reordering heuristic still requires improvement. We refer to these automatically generated labels as silver labels. As we describe in the next sections, the corpus annotated with these labels will be used to train our sequence labeling approach, eliminating the need for costly human-annotated data (i.e. gold labels). As a second way of evaluating the quality of our automatic labeling, we use these silver labels in a semi-oracle trial where we apply the actual simplification operations as given in the annotated corpus. In other words, we simply take the automatic labels as true and use the A D M R RM RW RWM C O A 240 0 0 0 0 2 0 0 19 D 15 333 8 4 5 1 1 4 0 M 0 1 22 0 0 0 0 1 0 R 0 33 0 28 6 0 0 4 0 RM 0 8 0 0 4 0 0 0 0 RW 3 31 4 7 2 4 0 6 0 RWM 0 6 0 0 0 0 0 0 0 C 0 24 98 1 1 1 0 1807 0 O 105 0 0 0 0 9 0 0 1998 Table 4 : Confusion matrix of true (rows) and automatically annotated (columns) operations on the manually annotated data.",
"cite_spans": [],
"ref_spans": [
{
"start": 749,
"end": 756,
"text": "Table 4",
"ref_id": null
},
{
"start": 1685,
"end": 1955,
"text": "A D M R RM RW RWM C O A 240 0 0 0 0 2 0 0 19 D 15 333 8 4 5 1 1 4 0 M 0 1 22 0 0 0 0 1 0 R 0 33 0 28 6 0 0 4 0 RM 0 8 0 0 4 0 0 0 0 RW 3 31 4 7 2 4 0 6 0 RWM 0 6 0 0 0 0 0 0 0 C 0 24 98 1 1 1 0 1807 0 O",
"ref_id": "TABREF2"
},
{
"start": 1979,
"end": 1986,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generating Training Data",
"sec_num": "4.1"
},
{
"text": "alignments between original and simplified words to apply the actual operations. This is what we refer to as silver operations in Table 1 . Using the automatic labeling would lead to much more accurate and less conservative simplifications than all translation-based approaches: it achieves the highest SARI and BLEU scores, and the lowest rate of copied input sentences among all systems tested using the 1-to-1 alignments. Therefore, the challenges now are (i) to predict such labels ( \u00a75.1), and (ii) to devise high-performing TS modules to apply simplification operations for each type of label ( \u00a74.2).",
"cite_spans": [],
"ref_spans": [
{
"start": 130,
"end": 137,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Automatically Annotated",
"sec_num": null
},
{
"text": "For our experiments ( \u00a75), we consider two of the operations that our algorithms can identify with high precision: DELETE and REPLACE. 8 Applying deletions is straightforward and amounts to simply omitting the respective token when generating the hypothesis sentence. For the REPLACE operation, we use the supervised Lexical Simplification approach of Paetzold and Specia (2017) . Their simplifier generates candidate substitutions for target words using parallel complex-to-simple corpora and retrofitted context-aware word embedding models, selects the ones that fit the context of the target word through the unsupervised boundary ranking approach, then ranks candidates using a supervised neural ranking model trained over manually annotated simplifications. It also performs a final confidence check step: the target is only replaced by the highest ranking candidate if the trigram probability of two words preceding the target is higher for the candidate.",
"cite_spans": [
{
"start": 352,
"end": 378,
"text": "Paetzold and Specia (2017)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Application of Operations",
"sec_num": "4.2"
},
{
"text": "Based on the automatic annotation procedure outlined above, we generate sequence annotations of 1-to-1 simplification operations in the Newsela corpus. On this data, we explore the questions (i) whether we can predict simplification operations to be performed on unseen data, and (ii) to what degree the prediction of these operations allows us to generate good simplifications.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "5"
},
{
"text": "To predict simplification operations for each input word, we train a bidirectional recurrent neural network, with an initial embedding layer of size 300 and two hidden LSTM (Long-Short Term Memory) layers of size 100. The training is done using Keras (Chollet, 2015) , with a batch size of 64, categorical cross-entropy loss and a dropout rate of 0.2 after the hidden layers. We optimize the model with Adagrad (Duchi et al., 2011) . We monitor the tagging accuracy on held-out development data and employ early stopping when the development loss increases. We repeat this process ten times with random initializations and select the best model based on development set accuracy. Table 5 shows that the LSTM model does not predict the silver labels very well. In particular, the model is relatively conservative with respect to the prediction of simplification operations, and tends to overpredict the majority class (i.e., to copy a token). 9 DELETE is the operation that our model predicts best. Table 6 shows the relative confusion of predicted operations versus the silver labels, and confirms that the main error type of our system is to keep a token rather than performing some simplification operation on it. We also see a tendency for other operations to be predicted as deletions.",
"cite_spans": [
{
"start": 251,
"end": 266,
"text": "(Chollet, 2015)",
"ref_id": null
},
{
"start": 411,
"end": 431,
"text": "(Duchi et al., 2011)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [
{
"start": 680,
"end": 687,
"text": "Table 5",
"ref_id": "TABREF6"
},
{
"start": 998,
"end": 1005,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "Prediction of Simplification Operations",
"sec_num": "5.1"
},
{
"text": "The results in the lower part of Table 1 (\"Predicted operations (1-to-1)\"), however, show that even though the operation predictions are far from the silver labels, our system is able to generate simple output by only applying the DELETE and Table 6 : Confusion matrix of true (rows) and predicted (columns) operations on the test data.",
"cite_spans": [],
"ref_spans": [
{
"start": 33,
"end": 40,
"text": "Table 1",
"ref_id": null
},
{
"start": 242,
"end": 249,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "Prediction of Simplification Operations",
"sec_num": "5.1"
},
{
"text": "REPLACE operations. In particular, our method achieves a better SARI score than all the baseline systems on the 1-to-1 alignments. As we consider the extrinsic evaluation of the final TS results to be more indicative of the quality of our model than its intrinsic evaluation in the sequence labeling task, we view this as a positive result.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Prediction of Simplification Operations",
"sec_num": "5.1"
},
{
"text": "We finally conduct a human evaluation of 100 simplifications produced by five simplifiers:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "\u2022 The experts' Reference simplification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "\u2022 The Moses simplifier (1-to-1).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "\u2022 The Nematus simplifier (1-to-1).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "\u2022 The NTS simplifier (1-to-1).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "\u2022 Our Sequence Labeling (SL) simplifier.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "Human evaluators (four NLP experts) are given the original sentence and the simplification in each of the above versions, and are asked to judge each of them with respect to their grammaticality (G), meaning preservation (M) and simplicity (S), using a Likert scale between 1 (worst) and 5 (best) for each aspect. We define \"simplicity\" as the extent to which the sentence was simpler than the original and thus easier to understand. A control set of 20 sentences is evaluated by all annotators in order to compute inter-annotator agreement.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human Evaluation",
"sec_num": "5.2"
},
{
"text": "Reference 5.00\u00b10.0 4.45\u00b10.9 2.70\u00b11.3 SL 4.16\u00b11.0 3.91\u00b11.1 1.66\u00b10.9 Nematus 4.49\u00b10.9 3.99\u00b11.2 1.46\u00b10.9 Moses 4.98\u00b10.2 4.99\u00b10.1 1.14\u00b10.4 NTS 4.75\u00b10.6 4.08\u00b11.26 1.53\u00b11.0",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "G M S",
"sec_num": null
},
{
"text": "Fleiss' Kappa 0.372 0.457 0.342 Table 7 : Average scores and standard deviation for grammaticality (G), meaning preservation (M) and simplicity (S) for the systems evaluated. The last row shows the inter-annotator agreement scores in terms of Fleiss' Kappa. Table 7 illustrates the average scores and standard deviations obtained by each system according to each criterion. As expected, the Moses simplifier obtains the highest grammaticality and meaning preservation scores, but the lowest simplicity scores, given that it tends to merely reproduce the input. Although Nematus and NTS manage to obtain slightly higher simplification scores, they still average very close to the lower end of the simplicity scale. Our SL approach, in turn, shows significantly higher simplicity scores than the other systems (according to a t-test with p < 0.05). Its less conservative edits, however, may in some cases come at the cost of lower scores for grammaticality and meaning preservation. The last row in Table 7 shows the values of inter-annotator agreement in terms of Fleiss' Kappa for each evaluation aspect. Table 8 exemplifies some of the sentences for which our system was rated better and worse than the baselines. It is important to mention that, although the first two reference simplifications in Table 8 feature only minor punctuation changes, only 2,538 references (0.8%) in the dataset are of this type.",
"cite_spans": [],
"ref_spans": [
{
"start": 32,
"end": 39,
"text": "Table 7",
"ref_id": null
},
{
"start": 258,
"end": 265,
"text": "Table 7",
"ref_id": null
},
{
"start": 997,
"end": 1004,
"text": "Table 7",
"ref_id": null
},
{
"start": 1105,
"end": 1112,
"text": "Table 8",
"ref_id": null
},
{
"start": 1300,
"end": 1307,
"text": "Table 8",
"ref_id": null
}
],
"eq_spans": [],
"section": "G M S",
"sec_num": null
},
{
"text": "We presented a novel approach to sentence simplification that uses automatically labeled training data from a large simplification corpus. Based on this annotated corpus, we devise a sequence labeling approach to text simplification that predicts simplification operations for individual words in the original sentence. Specific modules are then triggered to deal with each predicted operation. SL better than other Moses, Nematus and NTS O Kyarra Garrett has learned how to take blood pressure and perform CPR -and she is not even out of high school yet. R Kyarra Garrett has learned how to take blood pressure and perform CPR, and she is not even out of high school yet. M Kyarra Garrett has learned how to take blood pressure and perform CPR -and she is not even out of high school yet. N UNK Garrett loves out to take blood pressure and perform, and she is not even out of high school yet. T Chance Garrett has learned how to take blood pressure. L Kyarra Garrett has learned how to take blood pressure and perform CPR.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "O in her mind she stops at particular locations to pick up the correct cookie crumbs. R in her mind, she stops at particular locations to pick up the correct cookie crumbs. M in her mind she stops at particular locations to pick up the correct cookie crumbs. N she stops at particular locations to pick up the correct cookie UNK. T in her mind she stops at particular locations to pick up the correct cookie momentum. L in her mind she stops at particular areas to pick up cookie crumbs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "SL worse than Moses, Nematus or NTS O despite the limitations, Palestinian cooking is not without its fans. R despite the limitations, Palestinian cooking has its fans. M despite the limitations, Palestinian cooking is not without its fans. N Palestinian cooking is not without its fans. T even Palestinian cooking is not without its fans. L despite the limitations, Palestinian cooking is not without its fans.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "O \"we always thought there has to be a more efficient way of doing this,\" Zach Fiene said. R he said he always thought there had to be a better way of doing it. M \"we always thought there has to be a more efficient way of doing this,\" Zach Fiene said. N \"we always thought there has to be a more efficient way of doing this,\" said Zach Ghani, who is the 18-year-old said. T Zach Fiene said there has to be a more efficient way of doing this. L \"we always thought there has to be more efficient way doing this said. Table 8 : Example including original (O) and reference (R) sentences from the Newsela Article Corpus, and outputs generated by Moses (M), Nematus (N), NTS (T) and our sequence labeling approach (L).",
"cite_spans": [],
"ref_spans": [
{
"start": 515,
"end": 522,
"text": "Table 8",
"ref_id": null
}
],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "The experiments reported here cover only deletions and lexical substitutions as operations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "Our approach has several theoretical advantages over end-to-end translation models, including easier interpretability of the types of simplification learned, as well as the possibility for late decoding for adaptive simplification. In practical terms, we showed that our system outperforms translationbased approaches on a number of metrics and overcomes the problems of excessive repetition of the original content.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "According to human evaluation, our system achieves higher simplicity scores than the baseline systems, although this comes at the cost of slightly lower meaning preservation and grammaticality. We hypothesize that some of the problematic cases stem from not realizing the addition operation. In general, our approach will likely profit from good models for the remaining operations, especially those that can also operate on spans of several tokens, making research on such models a natural direction for further work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Further Work",
"sec_num": "6"
},
{
"text": "We follow instructions from http://www.statmt. org/moses/?n=Moses.Baseline4 We use a vocabulary size of 30, 000 and the same parameters as inSennrich et al. (2016).5 We use the same configurations asNisioi et al. (2017).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "For the experiments with the proposed TS approach, only 1-to-1 alignments are suitable. It is indeed not realistic to expect that complex operations that involve significant structural changes (e.g., splitting or joining sentences) could be modeled using sequence labeling approaches. For such complex operations, we believe explicitly representing the sentences' syntactic structures and learning abstract syntactic transformation rules (e.g. as inWoodsend and Lapata (2011) orFeblowitz and Kauchak (2013)) would be more advisable. However, we note that, as previously shown, translationbased end-to-end approaches also fail to learn such complex operations.7 Target-side annotations serve for analysis; they are ignored in our experiments as they are unavailable at test time.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We focus on this subset of operations since we currently lack good models to apply to the remaining operations. ADD, for example, would presume access to an external resource such as a knowledge base that would serve as a basis for inferring added content (which is oftentimes background information, for example an explanation that a certain person has a certain function). The results we obtain can thus be viewed as a lower bound on the simplification quality that can be expected from a model that integrates other operations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "By weighting the loss function by the ground truth class support at each timestep, we were able to alleviate the effect of a predominant majority class to some degree.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was partly supported by the EC project SIMPATICO (H2020-EURO-6-2015, grant number 692819).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Neural machine translation by jointly learning to align and translate",
"authors": [
{
"first": "Dzmitry",
"middle": [],
"last": "Bahdanau",
"suffix": ""
},
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "Yoshua",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2014,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. CoRR, abs/1409.0473.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Sentence alignment for monolingual comparable corpora",
"authors": [
{
"first": "Regina",
"middle": [],
"last": "Barzilay",
"suffix": ""
},
{
"first": "Noemie",
"middle": [],
"last": "Elhadad",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of EMNLP",
"volume": "",
"issue": "",
"pages": "25--32",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Regina Barzilay and Noemie Elhadad. 2003. Sentence alignment for monolingual comparable corpora. In Proceedings of EMNLP, pages 25-32.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Text Simplification as Tree Labeling",
"authors": [
{
"first": "Joachim",
"middle": [],
"last": "Bingel",
"suffix": ""
},
{
"first": "Anders",
"middle": [],
"last": "S\u00f8gaard",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "337--343",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joachim Bingel and Anders S\u00f8gaard. 2016. Text Sim- plification as Tree Labeling. In Proceedings of ACL, pages 337-343.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "An unsupervised alignment algorithm for text simplification corpus construction",
"authors": [
{
"first": "Stefan",
"middle": [],
"last": "Bott",
"suffix": ""
},
{
"first": "Horacio",
"middle": [],
"last": "Saggion",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of MTTG",
"volume": "",
"issue": "",
"pages": "20--26",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stefan Bott and Horacio Saggion. 2011. An unsu- pervised alignment algorithm for text simplifica- tion corpus construction. In Proceedings of MTTG, pages 20-26.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Learning to simplify sentences using wikipedia",
"authors": [
{
"first": "William",
"middle": [],
"last": "Coster",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Kauchak",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of the Workshop on Monolingual Text-To-Text Generation",
"volume": "",
"issue": "",
"pages": "1--9",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "William Coster and David Kauchak. 2011a. Learning to simplify sentences using wikipedia. In Proceed- ings of the Workshop on Monolingual Text-To-Text Generation, pages 1-9.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Simple English Wikipedia: a new text simplification task",
"authors": [
{
"first": "William",
"middle": [],
"last": "Coster",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Kauchak",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "665--669",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "William Coster and David Kauchak. 2011b. Simple English Wikipedia: a new text simplification task. In Proceedings of ACL, pages 665-669.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Adaptive subgradient methods for online learning and stochastic optimization",
"authors": [
{
"first": "John",
"middle": [],
"last": "Duchi",
"suffix": ""
},
{
"first": "Elad",
"middle": [],
"last": "Hazan",
"suffix": ""
},
{
"first": "Yoram",
"middle": [],
"last": "Singer",
"suffix": ""
}
],
"year": 2011,
"venue": "Journal of Machine Learning Research",
"volume": "12",
"issue": "",
"pages": "2121--2159",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John Duchi, Elad Hazan, and Yoram Singer. 2011. Adaptive subgradient methods for online learning and stochastic optimization. Journal of Machine Learning Research, 12:2121-2159.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Sentence simplification as tree transduction",
"authors": [
{
"first": "Dan",
"middle": [],
"last": "Feblowitz",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Kauchak",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the PITR Workshop",
"volume": "",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dan Feblowitz and David Kauchak. 2013. Sentence simplification as tree transduction. In Proceedings of the PITR Workshop, pages 1-10.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Moses: Open source toolkit for statistical machine translation",
"authors": [
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "Hieu",
"middle": [],
"last": "Hoang",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Birch",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Marcello",
"middle": [],
"last": "Federico",
"suffix": ""
},
{
"first": "Nicola",
"middle": [],
"last": "Bertoldi",
"suffix": ""
},
{
"first": "Brooke",
"middle": [],
"last": "Cowan",
"suffix": ""
},
{
"first": "Wade",
"middle": [],
"last": "Shen",
"suffix": ""
},
{
"first": "Christine",
"middle": [],
"last": "Moran",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Zens",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Dyer",
"suffix": ""
},
{
"first": "Ond\u0159ej",
"middle": [],
"last": "Bojar",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Constantin",
"suffix": ""
},
{
"first": "Evan",
"middle": [],
"last": "Herbst",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of ACL -Demonstration Session",
"volume": "",
"issue": "",
"pages": "177--180",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ond\u0159ej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proceedings of ACL -Demonstration Session, pages 177-180.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Hybrid simplification using deep semantics and machine translation",
"authors": [
{
"first": "Shashi",
"middle": [],
"last": "Narayan",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Gardent",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "435--445",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shashi Narayan and Claire Gardent. 2014. Hybrid sim- plification using deep semantics and machine trans- lation. In Proceedings of ACL, pages 435-445.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Exploring neural text simplification models",
"authors": [
{
"first": "Sergiu",
"middle": [],
"last": "Nisioi",
"suffix": ""
},
{
"first": "Simone",
"middle": [
"Paolo"
],
"last": "Sanja\u0161tajner",
"suffix": ""
},
{
"first": "Liviu",
"middle": [
"P"
],
"last": "Ponzetto",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Dinu",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "85--91",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sergiu Nisioi, Sanja\u0160tajner, Simone Paolo Ponzetto, and Liviu P. Dinu. 2017. Exploring neural text sim- plification models. In Proceedings of ACL, pages 85-91.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Semeval 2016 task 11: Complex word identification",
"authors": [
{
"first": "Gustavo",
"middle": [
"H"
],
"last": "Paetzold",
"suffix": ""
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 10th SemEval",
"volume": "",
"issue": "",
"pages": "560--569",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gustavo H. Paetzold and Lucia Specia. 2016a. Se- meval 2016 task 11: Complex word identification. In Proceedings of the 10th SemEval, pages 560-569.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Vicinity-driven paragraph and sentence alignment for comparable corpora",
"authors": [
{
"first": "Gustavo",
"middle": [
"Henrique"
],
"last": "Paetzold",
"suffix": ""
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
}
],
"year": 2016,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gustavo Henrique Paetzold and Lucia Specia. 2016b. Vicinity-driven paragraph and sentence alignment for comparable corpora. CoRR, abs/1612.04113.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Lexical simplification with neural ranking. Proceedings of EACL",
"authors": [
{
"first": "Gustavo",
"middle": [
"Henrique"
],
"last": "Paetzold",
"suffix": ""
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "34--40",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gustavo Henrique Paetzold and Lucia Specia. 2017. Lexical simplification with neural ranking. Pro- ceedings of EACL, pages 34-40.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "BLEU: a method for automatic evaluation of machine translation",
"authors": [
{
"first": "Kishore",
"middle": [],
"last": "Papineni",
"suffix": ""
},
{
"first": "Salim",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "Todd",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "Wei-Jing",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "311--318",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. In Proceedings of ACL, pages 311-318.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Nematus: a toolkit for neural machine translation",
"authors": [
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
},
{
"first": "Orhan",
"middle": [],
"last": "Firat",
"suffix": ""
},
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Birch",
"suffix": ""
},
{
"first": "Barry",
"middle": [],
"last": "Haddow",
"suffix": ""
},
{
"first": "Julian",
"middle": [],
"last": "Hitschler",
"suffix": ""
},
{
"first": "Marcin",
"middle": [],
"last": "Junczys-Dowmunt",
"suffix": ""
},
{
"first": "Samuel",
"middle": [],
"last": "L\u00e4ubli",
"suffix": ""
},
{
"first": "Antonio",
"middle": [],
"last": "Valerio Miceli",
"suffix": ""
},
{
"first": "Jozef",
"middle": [],
"last": "Barone",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Mokry",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Nadejde",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the Software Demonstrations of EACL",
"volume": "",
"issue": "",
"pages": "65--68",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rico Sennrich, Orhan Firat, Kyunghyun Cho, Alexan- dra Birch, Barry Haddow, Julian Hitschler, Marcin Junczys-Dowmunt, Samuel L\u00e4ubli, Antonio Valerio Miceli Barone, Jozef Mokry, and Maria Nadejde. 2017. Nematus: a toolkit for neural machine trans- lation. In Proceedings of the Software Demonstra- tions of EACL, pages 65-68, Valencia, Spain. ACL.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Edinburgh neural machine translation systems for wmt 16",
"authors": [
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
},
{
"first": "Barry",
"middle": [],
"last": "Haddow",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Birch",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of WMT",
"volume": "",
"issue": "",
"pages": "371--376",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Edinburgh neural machine translation sys- tems for wmt 16. In Proceedings of WMT, pages 371-376.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Extracting parallel sentences from comparable corpora using document level alignment",
"authors": [
{
"first": "R",
"middle": [],
"last": "Jason",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Smith",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Quirk",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of NAACL",
"volume": "",
"issue": "",
"pages": "403--411",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jason R Smith, Chris Quirk, and Kristina Toutanova. 2010. Extracting parallel sentences from compara- ble corpora using document level alignment. In Pro- ceedings of NAACL, pages 403-411.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "A study of translation edit rate with targeted human annotation",
"authors": [
{
"first": "Matthew",
"middle": [],
"last": "Snover",
"suffix": ""
},
{
"first": "Bonnie",
"middle": [],
"last": "Dorr",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Schwartz",
"suffix": ""
},
{
"first": "Linnea",
"middle": [],
"last": "Micciulla",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Makhoul",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of AMTA",
"volume": "",
"issue": "",
"pages": "223--231",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew Snover, Bonnie Dorr, Richard Schwartz, Lin- nea Micciulla, and John Makhoul. 2006. A study of translation edit rate with targeted human annotation. In Proceedings of AMTA, pages 223-231.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Translating from complex to simplified sentences",
"authors": [
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
}
],
"year": 2010,
"venue": "Computational Processing of the Portuguese Language",
"volume": "",
"issue": "",
"pages": "30--39",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lucia Specia. 2010. Translating from complex to sim- plified sentences. In Computational Processing of the Portuguese Language, pages 30-39. Springer.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Back to basics for monolingual alignment: Exploiting word similarity and contextual evidence",
"authors": [
{
"first": "Md",
"middle": [],
"last": "Sultan",
"suffix": ""
},
{
"first": "Steven",
"middle": [],
"last": "Bethard",
"suffix": ""
},
{
"first": "Tamara",
"middle": [],
"last": "Sumner",
"suffix": ""
}
],
"year": 2014,
"venue": "TACL",
"volume": "2",
"issue": "",
"pages": "219--230",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Md Sultan, Steven Bethard, and Tamara Sumner. 2014. Back to basics for monolingual alignment: Exploit- ing word similarity and contextual evidence. TACL, 2:219-230.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Learning to simplify sentences with quasi-synchronous grammar and integer programming",
"authors": [
{
"first": "Kristian",
"middle": [],
"last": "Woodsend",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of EMNLP",
"volume": "",
"issue": "",
"pages": "409--420",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kristian Woodsend and Mirella Lapata. 2011. Learn- ing to simplify sentences with quasi-synchronous grammar and integer programming. In Proceedings of EMNLP, pages 409-420.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Sentence simplification by monolingual machine translation",
"authors": [
{
"first": "",
"middle": [],
"last": "Sander Wubben",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Van Den",
"suffix": ""
},
{
"first": "Emiel",
"middle": [],
"last": "Bosch",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Krahmer",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "1015--1024",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sander Wubben, Antal van den Bosch, and Emiel Krahmer. 2012. Sentence simplification by mono- lingual machine translation. In Proceedings of ACL, pages 1015-1024.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Problems in current text simplification research: New data can help",
"authors": [
{
"first": "Wei",
"middle": [],
"last": "Xu",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Courtney",
"middle": [],
"last": "Napoles",
"suffix": ""
}
],
"year": 2015,
"venue": "TACL",
"volume": "3",
"issue": "",
"pages": "283--297",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wei Xu, Chris Callison-Burch, and Courtney Napoles. 2015. Problems in current text simplification re- search: New data can help. TACL, 3:283-297.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Optimizing statistical machine translation for text simplification",
"authors": [
{
"first": "Wei",
"middle": [],
"last": "Xu",
"suffix": ""
},
{
"first": "Courtney",
"middle": [],
"last": "Napoles",
"suffix": ""
},
{
"first": "Ellie",
"middle": [],
"last": "Pavlick",
"suffix": ""
},
{
"first": "Quanze",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
}
],
"year": 2016,
"venue": "TACL",
"volume": "4",
"issue": "",
"pages": "401--415",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch. 2016. Optimizing statistical machine translation for text simplification. TACL, 4:401-415.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Sentence simplification with deep reinforcement learning",
"authors": [
{
"first": "Xingxing",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of EMNLP",
"volume": "",
"issue": "",
"pages": "595--605",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xingxing Zhang and Mirella Lapata. 2017. Sentence simplification with deep reinforcement learning. In Proceedings of EMNLP, pages 595-605.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "A constrained sequenceto-sequence neural model for sentence simplification",
"authors": [
{
"first": "Yaoyuan",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Zhenxu",
"middle": [],
"last": "Ye",
"suffix": ""
},
{
"first": "Yansong",
"middle": [],
"last": "Feng",
"suffix": ""
},
{
"first": "Dongyan",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "Rui",
"middle": [],
"last": "Yan",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yaoyuan Zhang, Zhenxu Ye, Yansong Feng, Dongyan Zhao, and Rui Yan. 2017. A constrained sequence- to-sequence neural model for sentence simplifica- tion. CoRR, abs/1704.02312.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "A monolingual tree-based translation model for sentence simplification",
"authors": [
{
"first": "Zhemin",
"middle": [],
"last": "Zhu",
"suffix": ""
},
{
"first": "Delphine",
"middle": [],
"last": "Bernhard",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of COL-ING",
"volume": "",
"issue": "",
"pages": "1353--1361",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhemin Zhu, Delphine Bernhard, and Iryna Gurevych. 2010. A monolingual tree-based translation model for sentence simplification. In Proceedings of COL- ING, pages 1353-1361.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Example of automatic labeling based on word alignments between an original (top) and a simplified (bottom) sentence in the Newsela corpus. Unaligned words on the original side receive label 'D' (DELETE), while words that are aligned to a different form receive 'R' (REPLACE). Aligned words without an explicit label receive a 'C' label (COPY). Sentences are from the Newsela Article Corpus.",
"uris": null,
"type_str": "figure",
"num": null
},
"FIGREF1": {
"text": "Example of automatic annotation for label MOVE ('M'). Sentences are from the Newsela Article Corpus.",
"uris": null,
"type_str": "figure",
"num": null
},
"TABREF2": {
"text": "Count and proportion of instances affected by each type of simplification transformation performed by Moses and Nematus.",
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null
},
"TABREF4": {
"text": "Per-label performance of automatic annotation of operations.",
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null
},
"TABREF6": {
"text": "Per-label performance of automatic operation prediction with the LSTM model.",
"type_str": "table",
"html": null,
"content": "<table><tr><td/><td>Predicted</td><td/></tr><tr><td/><td colspan=\"2\">D M R RM RW RWM C</td></tr><tr><td>D</td><td>.49 .06 .07 .00 .00</td><td>.00 .38</td></tr><tr><td>M</td><td>.41 .16 .05 .00 .00</td><td>.00 .38</td></tr><tr><td>R</td><td>.23 .05 .34 .00 .00</td><td>.00 .38</td></tr><tr><td colspan=\"2\">RM .32 .09 .21 .00 .00</td><td>.00 .38</td></tr><tr><td colspan=\"2\">RW .38 .00 .00 .00 .07</td><td>.00 .54</td></tr><tr><td colspan=\"2\">RWM .62 .03 .00 .00 .04</td><td>.00 .32</td></tr><tr><td>C</td><td>.33 .09 .06 .00 .00</td><td>.00 .51</td></tr></table>",
"num": null
}
}
}
} |