File size: 101,402 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 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:12:22.054794Z"
},
"title": "A Consolidated Dataset for Knowledge-based Question Generation using Predicate Mapping of Linked Data",
"authors": [
{
"first": "Johanna",
"middle": [],
"last": "Melly",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "HEIG-VD / HES-SO Route de Vevey",
"location": {
"addrLine": "105a Route de Cheseaux 1, 521 Route de Cheseaux 1",
"postCode": "521 1618, 1401, 1401",
"settlement": "Ch\u00e2tel-Saint-Denis, Yverdon-les-Bains, Yverdon-les-Bains",
"region": "CP, CP",
"country": "Switzerland Switzerland Switzerland"
}
},
"email": "johanna@melly.me"
},
{
"first": "Gabriel",
"middle": [],
"last": "Luthier",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "HEIG-VD / HES-SO Route de Vevey",
"location": {
"addrLine": "105a Route de Cheseaux 1, 521 Route de Cheseaux 1",
"postCode": "521 1618, 1401, 1401",
"settlement": "Ch\u00e2tel-Saint-Denis, Yverdon-les-Bains, Yverdon-les-Bains",
"region": "CP, CP",
"country": "Switzerland Switzerland Switzerland"
}
},
"email": "gabriel.luthier@heig-vd.chandrei.popescu-belis@heig-vd.ch"
},
{
"first": "Andrei",
"middle": [],
"last": "Popescu-Belis",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "HEIG-VD / HES-SO Route de Vevey",
"location": {
"addrLine": "105a Route de Cheseaux 1, 521 Route de Cheseaux 1",
"postCode": "521 1618, 1401, 1401",
"settlement": "Ch\u00e2tel-Saint-Denis, Yverdon-les-Bains, Yverdon-les-Bains",
"region": "CP, CP",
"country": "Switzerland Switzerland Switzerland"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "In this paper, we present the ForwardQuestions data set, made of human-generated questions related to knowledge triples. This data set results from the conversion and merger of the existing SimpleDBPediaQA and SimpleQuestionsWikidata data sets, including the mapping of predicates from DBPedia to Wikidata, and the selection of 'forward' questions as opposed to 'backward' ones. The new data set can be used to generate novel questions given an unseen Wikidata triple, by replacing the subjects of existing questions with the new one and then selecting the best candidate questions using semantic and syntactic criteria. Evaluation results indicate that the question generation method using ForwardQuestions improves the quality of questions by about 20% with respect to a baseline not using ranking criteria.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "In this paper, we present the ForwardQuestions data set, made of human-generated questions related to knowledge triples. This data set results from the conversion and merger of the existing SimpleDBPediaQA and SimpleQuestionsWikidata data sets, including the mapping of predicates from DBPedia to Wikidata, and the selection of 'forward' questions as opposed to 'backward' ones. The new data set can be used to generate novel questions given an unseen Wikidata triple, by replacing the subjects of existing questions with the new one and then selecting the best candidate questions using semantic and syntactic criteria. Evaluation results indicate that the question generation method using ForwardQuestions improves the quality of questions by about 20% with respect to a baseline not using ranking criteria.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Question generation from linked data is a promising approach for producing large corpora of questions and answers. A primary use of these corpora is for training and evaluating question answering systems (Duan et al., 2017) , while other uses are for education (Pham et al., 2018) , tutoring (Su et al., 2019) , or entertainment. Automatic question generation can be based on texts or on large repositories of linked data. In the latter case, an initial set of human-generated questions is often necessary to generate new ones, but such data sets are strongly related to specific linked data formats, and are difficult to port to new repositories. In this paper, we present the ForwardQuestions corpus of human-generated questions associated to knowledge triples from the Wikidata knowledge base. We constructed this corpus by converting and merging two partially overlapping corpora of questions, SimpleDBPediaQA and SimpleQues-tionsWikidata, which were separately derived from subsets of the SimpleQuestions corpus. These three data sets are respectively based on DBpedia, Wikidata and Freebase, but the latter resource is no longer available. Specifically, we enriched SimpleQuestionsWikidata with a substantial number of questions from SimpleDBPediaQA, by converting DBpedia predicates to Wikidata ones and keeping only the 'forward' questions, given our final goal of quiz generation. 1 The overlap between these two resources is only of 32%, showing that the resulting data set has considerable novelty. As a result, we make available, under the Creative Commons Attribution license (BY), the ForwardQuestions corpus of 38k questions related to 94 different Wikidata predicates. 2 Furthermore, we show how ForwardQuestions can be used to generate new questions from previously unseen triples, by replacing the subjects of existing questions with new ones, and then ranking candidate questions on semantic and syntactic criteria. The questions can be used, for instance, in a chatbot that generates quizzes on any topic indicated by a user, thanks to a strategy for selecting relevant triples from Wikidata. The evaluation results with human subjects who rate the quality of the questions show that the best questions generated by our method reach about 80% approval, of which 10 points are due to the question ranking method. The paper is organized as follows. In Section 2, we review related work and present the SimpleQuestions, Simple-DBpediaQA and SimpleQuestionsWikidata resources. In Section 3 we explain how the latter two data sets were converted and merged into the new ForwardQuestions data set. In Section 4, we describe our template-based question generation method and the semantic and syntactic ranking strategies, used in a chatbot presented briefly in Section 5. In Section 6, we define the evaluation protocol and quantify the improvements brought by our resource and question generation method.",
"cite_spans": [
{
"start": 204,
"end": 223,
"text": "(Duan et al., 2017)",
"ref_id": "BIBREF7"
},
{
"start": 261,
"end": 280,
"text": "(Pham et al., 2018)",
"ref_id": "BIBREF14"
},
{
"start": 292,
"end": 309,
"text": "(Su et al., 2019)",
"ref_id": "BIBREF18"
},
{
"start": 1390,
"end": 1391,
"text": "1",
"ref_id": null
},
{
"start": 1685,
"end": 1686,
"text": "2",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Question answering (QA) has been extensively researched in the past. Many methods use textual documents to find answers, while others consider knowledge bases, such as large sets of knowledge triples (subject, predicate, object). QA over knowledge bases requires data sets with questions and their answers, for training and evaluation. For instance, the data sets used for the QALD evaluations (Unger et al., 2016) typically include hundreds of questions, most of which can be answered based on a single triple, while others require a combination of triples. In the past, triple stores such as Freebase (with around 40 million entities) or DBpedia (an order of magnitude smaller) have been used to design QA systems (Bast and Haussmann, 2015) . The termination of the Freebase repository raised the question of resource conversion to DBpedia, or to the more recent Wikidata triple store, 3 which is a knowledge graph derived from Wikipedia infoboxes and allows data querying with SPARQL (Malyshev et al., 2018) . The main challenge remains however the generation of questions from triples, which is a costly process that has been partially automated in the past, as we briefly review hereafter.",
"cite_spans": [
{
"start": 394,
"end": 414,
"text": "(Unger et al., 2016)",
"ref_id": "BIBREF20"
},
{
"start": 716,
"end": 742,
"text": "(Bast and Haussmann, 2015)",
"ref_id": "BIBREF1"
},
{
"start": 987,
"end": 1010,
"text": "(Malyshev et al., 2018)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Relation to Previous Work",
"sec_num": "2."
},
{
"text": "Existing methods for question generation start either from textual data or from knowledge triples. Heilman and Smith (2010) defined rules for syntactic transformation of declarative sentences into questions, which were then ranked by a logistic regression model, reaching an acceptance rate of about 50% for the 20% top-ranked questions. Chali and Hasan (2015) used named entity and predicate-argument information to generate questions, but evaluated them only automatically. They used LDA to estimate topic relevance, and syntactic tree kernels for grammaticality judgments. A rule-based approach to generate questions from relative subordinate sentences extracted from Wikipedia was proposed by Khullar et al. (2018) . This method generated better questions than Heilman and Smith, but relied crucially on the availability of relative pronouns and adverbs. More recent models attempt to generate questions from sentences using deep neural networks, e.g. starting from a sentence and the intended answer word (Sun et al., 2018; Zhao et al., 2018) . Currently, their accuracy on long sentences such as those from Wikipedia is sufficient for quiz generation, especially since they were only evaluated by quantitative comparisons to the SQuAD data set (Rajpurkar et al., 2016) ). Recent improvements aim at predicting the question type from the answer and then add this prediction to the neural generator (Zhou et al., 2019) . Serban et al. (2016) proposed two methods for question generation from Freebase. The neural network approach used TransE multi-relational embeddings (Bordes et al., 2013) and leveraged conditional language generation models. They generated a corpus of 30 million questions based on Freebase triples, which were evaluated with the BLEU metric and partly with human judges. Their template-based baseline model scored only slightly below, but is applicable also when TransE embeddings are not available -hence, it is the starting point of our present proposal.",
"cite_spans": [
{
"start": 99,
"end": 123,
"text": "Heilman and Smith (2010)",
"ref_id": "BIBREF9"
},
{
"start": 338,
"end": 360,
"text": "Chali and Hasan (2015)",
"ref_id": "BIBREF4"
},
{
"start": 697,
"end": 718,
"text": "Khullar et al. (2018)",
"ref_id": "BIBREF11"
},
{
"start": 1010,
"end": 1028,
"text": "(Sun et al., 2018;",
"ref_id": "BIBREF19"
},
{
"start": 1029,
"end": 1047,
"text": "Zhao et al., 2018)",
"ref_id": "BIBREF21"
},
{
"start": 1250,
"end": 1274,
"text": "(Rajpurkar et al., 2016)",
"ref_id": "BIBREF16"
},
{
"start": 1403,
"end": 1422,
"text": "(Zhou et al., 2019)",
"ref_id": "BIBREF22"
},
{
"start": 1425,
"end": 1445,
"text": "Serban et al. (2016)",
"ref_id": "BIBREF17"
},
{
"start": 1574,
"end": 1595,
"text": "(Bordes et al., 2013)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Automatic Question Generation",
"sec_num": "2.1."
},
{
"text": "The SimpleQuestions dataset (Bordes et al., 2015) 4 features 108,442 questions in English obtained through a crowdsourcing platform. Each question is accompanied by the knowledge triple from Freebase on which it is based, which 3 www.wikidata.org 4 Part of the bAbI evaluation tasks from Facebook Research:",
"cite_spans": [
{
"start": 28,
"end": 49,
"text": "(Bordes et al., 2015)",
"ref_id": "BIBREF3"
},
{
"start": 50,
"end": 51,
"text": "4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Human-generated Questions from Triples",
"sec_num": "2.2."
},
{
"text": "research.fb.com/downloads/babi/.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Human-generated Questions from Triples",
"sec_num": "2.2."
},
{
"text": "also provides its answer. For instance, one question is \"What does Jimmy Neutron do?\", and the triple ('Jimmy Neutron', 'fictional character occupation', 'inventor') indicates that the answer is \"inventor\". An important distinction introduced by SimpleQuestions, coming from the observation of human-generated questions, is between forward and backward questions. A forward question bears on the object of a triple, while a backward one bears on its subject, and is often formulated using passive voice. For instance, from the triple ('The Dishwasher: Dead Samurai', 'publisher', 'Xbox Game Studios'), someone generated the question \"What company published The Dishwasher: Dead Samurai?\", which is a forward one. However, from the triple ('Rampage', 'publisher', 'Midway Games'), someone wrote the question \"What game is published by Midway Games?\", which is a backward one. One reason to consider this distinction is that predicates do not appear in both active and passive forms in the triple store, so questions are allowed to bear on the subject or or object of a triple. Due to the termination of Freebase, the triples of a subset of questions from SimpleQuestions have been converted to DBpedia triples, resulting in the SimpleDBpediaQA data set (Azmy et al., 2018) . 5 Two formatted questions from this data set are presented in Table 1 . 'Query' is the original question formulated over a Freebase triple, whose former predicate URL is given under 'Freebase Predicate'. 'Subject' points to the URL of the concept on DBpedia. There are three subfields under 'predicate list': the DBpedia URL of the predicate, the direction of the question (forward or backward), and a constraint on the expected answer type for backward questions. A subset of SimpleQuestions different from the one above has been converted to Wikidata triples, resulting in the SimpleQuestionsWikidata set (Diefenbach et al., 2017) . 6 The resource is available as a text document, formatted as shown in Table 2 . Each line has four tab-separated fields, containing the Wikidata identifier of the triple's subject, predicate, and object, and the question itself. The predicates of forward questions have Wikidata identifiers prefixed with 'P', e.g. 'P413' refers to the Wikidata property at www.wikidata.org/wiki/Property: P413 with the English label \"position played on team / speciality\". Backward questions are indicated by predicates whose initial letter was changed from 'P' to 'R', as in the third example from Table 2 , where 'R509' indicates the fact that the 'P509' property (\"cause of death\", www. wikidata.org/wiki/Property:P509) holds between the object and the subject and not vice-versa, and that the actual triple in Wikidata is (Q6371569, R509, Q12152), \"Karl Anton Rickenbacher died of myocardial infarction.\" Finally, a smaller set of about 700 questions collected from users over Wikidata triples is also available as the WDAquaCore0Wikidata set (Diefenbach et al., 2017 ",
"cite_spans": [
{
"start": 1252,
"end": 1271,
"text": "(Azmy et al., 2018)",
"ref_id": "BIBREF0"
},
{
"start": 1274,
"end": 1275,
"text": "5",
"ref_id": null
},
{
"start": 1881,
"end": 1906,
"text": "(Diefenbach et al., 2017)",
"ref_id": "BIBREF6"
},
{
"start": 1909,
"end": 1910,
"text": "6",
"ref_id": null
},
{
"start": 2940,
"end": 2964,
"text": "(Diefenbach et al., 2017",
"ref_id": "BIBREF6"
}
],
"ref_spans": [
{
"start": 1336,
"end": 1343,
"text": "Table 1",
"ref_id": "TABREF1"
},
{
"start": 1979,
"end": 1986,
"text": "Table 2",
"ref_id": "TABREF2"
},
{
"start": 2492,
"end": 2499,
"text": "Table 2",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Human-generated Questions from Triples",
"sec_num": "2.2."
},
{
"text": "From the 108,442 entries in SimpleQuestions, 43,086 were included in SimpleDBpediaQA, while 49,202 were included in SimpleQuestionsWikidata. There is therefore a potential to select more of the original questions for inclusion in ForwardQuestions. Questions in SimpleDBpediaQA are not accompanied by the object of their triple, which means that their correct answer cannot be verified directly from the data set, unlike those from SimpleQuestionsWikidata, as it appears when comparing Table 1 with Table 2 . This is not a major problem, nevertheless, because: (1) in general, the correct answer may not be unique even if the question is based on a single triple (e.g. \"who are the children of Barack Obama?\"), so the underlying triple store is still needed to verify the answer;",
"cite_spans": [],
"ref_spans": [
{
"start": 485,
"end": 505,
"text": "Table 1 with Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Comparison of SimpleDBpediaQA and SimpleQuestionsWikidata",
"sec_num": "2.3."
},
{
"text": "(2) for our intended use, the questions from the database are only used as templates to generate new questions from new triples (see Section 4), therefore the objects of the original triples are never needed. Qualitatively, the questions in SimpleDBpediaQA cover a smaller range of predicates than those in SimpleQuestions-Wikidata, and contain fewer questions per triple. The latter set uses Wikidata predicates, which are often more finegrained than the DBpedia ones (for instance distinguishing 'father' and 'mother' where DBpedia has only 'parent').",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of SimpleDBpediaQA and SimpleQuestionsWikidata",
"sec_num": "2.3."
},
{
"text": "Another qualitative observation is that SimpleDBpediaQA contains a somewhat larger proportion of triples that are not useful for question generation, as they correspond to various numeric identifiers of entities in 3rd party repositories.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of SimpleDBpediaQA and SimpleQuestionsWikidata",
"sec_num": "2.3."
},
{
"text": "DBpedia or Wikidata triples represent only small subsets of the knowledge embodied in Wikipedia, which is why it may seem that generating questions directly from Wikipedia sentences could lead to more varied questions (Heilman and Smith, 2010; Chali and Hasan, 2015) . However, our pilot experiments in this direction pointed to strong limitations. For instance, we considered identifying patterns such as verb + named entity in sentences from Wikipedia, and then reversing them to build a question, e.g. from \"World War II ended in 1945\" we aimed to derive \"When did World War II end?\" However, several difficulties appeared: (1) the VB+NE pattern also applies to relative clauses (e.g. \"Billie Joe Armstrong took two years to write American Idiot\") from which questions cannot be easily generated;",
"cite_spans": [
{
"start": 218,
"end": 243,
"text": "(Heilman and Smith, 2010;",
"ref_id": "BIBREF9"
},
{
"start": 244,
"end": 266,
"text": "Chali and Hasan, 2015)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Motivation for ForwardQuestions",
"sec_num": "3.1."
},
{
"text": "(2) the interrogative word is hard to predict;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Motivation for ForwardQuestions",
"sec_num": "3.1."
},
{
"text": "(3) pronouns lead to unintelligible questions; (4) answers should not be limited to named entities. Therefore, we turned to the use of knowledge triples, following the template-based baseline proposed by Serban et al. (2016) . Triples enable a straightforward generation method: transform the triple (subject, predicate, object) into a question bearing on the 'predicate' property of the 'subject', knowing that 'object' one of the correct answers. For instance, from ('Harry Potter', 'mother', 'Lily Potter') one can construct \"Who is the mother of Harry Potter?\". Note that 'subject' and 'object' do not necessarily have these grammatical functions in the sentence from which the triple was generated, as these functions depend on the form of the predicate. In the above example, the natural formulation \"Lily Potter is the mother of Harry Potter\" actually reverses these roles.",
"cite_spans": [
{
"start": 204,
"end": 224,
"text": "Serban et al. (2016)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Motivation for ForwardQuestions",
"sec_num": "3.1."
},
{
"text": "It appeared however that, in general, the specific wordings describing the subject, the predicate, and the expected type of answer are difficult to generate correctly. For instance, from ('Harry Potter', 'composer', 'John Williams'), the derived question \"Who is the composer of Harry Potter?\" is incorrectly formulated -a correct version is \"Which composer wrote the music for the film Harry Potter?\". This is why we use template-based generation from questions written by humans in the ForwardQuestions data set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Motivation for ForwardQuestions",
"sec_num": "3.1."
},
{
"text": "SimpleDBpediaQA and SimpleQuestionsWikidata are both subsets of SimpleQuestions. Hence, they have a certain amount of overlapping questions, but also some that are specific to each set. Therefore, merging the two subsets results in a larger one, named ForwardQuestions. Given their different formats, we decided to convert them to a new format, which preserves all the information from both data sets. This format also includes a template derived from each question, which can be used for question generation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "The main added value of the resource is the conversion of DBpedia predicates to Wikidata ones, resulting in a resource that is enriched with respect to both of its sources, although it still cannot recover all original SimpleQuestions items based on Freebase predicates, as not all of them have mappings in Wikidata. We do not include backward questions, because they are not convenient for generating new questions. Indeed, they typically accept a much larger number of possible answers than forward ones, and may therefore appear as either too open or too easy. Indeed, asking about a property of a subject makes a good question, as subjects have a limited number of properties. However, asking which subjects have a given property is generally not a good question because the same property can potentially apply to a very large number of subjects. For instance, \"In what country is Geneva?\" is a good question, while \"What city is in Switzerland?\" is not, although both are based on ('Geneva', 'country', 'Switzerland'). Given our goal of quiz generation, we exclude backward questions, of which there are 14,632 in SimpleDBpediaQA (34%) and 12,420 in SimpleQuestions-Wikidata (25%). We now describe the mapping process for the SimpleQues-tionsWikidata entires, and discuss afterwards the differences with SimpleDBpediaQA. We process each (subject, predicate, object, question) line as follows. We first exclude predicate starting with an 'R' (backwards question). Then, we query the Wikidata API to find the English labels of the subject, predicate, and object, and exclude questions for which the subject or the predicate cannot be found. Next, we build a template from each question, for question generation. We identify the position of the subject in the question, and replace it with the string '<placeholder>'. As different referring expressions were sometimes used for subjects, we allow for some flexibility when matching subject labels. For instance, we replace dashes, apostrophes and non-ASCII characters with white spaces, to increase the number of matches. Still, due to misspellings, simplifications, confusion of subject or object, or insertion of external knowledge about the subject, no match can be identified for about 4% of the questions, which are excluded. A similar conversion was performed for SimpleDBpe-diaQA entries, but this required a mapping of DBpedia predicates to Wikidata ones, which we explain in the next subsection. The subjects were also mapped to their Wikidata equivalent, using requests to the APIs and matching the English labels of the entities (as stated above, objects are missing in this case). As a result, each item appears in ForwardQuestions as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "\u2022 Question: full text and template based on it;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "\u2022 Subject: label (English words) and Wikidata code;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "\u2022 Predicate: label and Wikidata code;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "\u2022 Object: if available, label and Wikidata code.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Construction of the Data Set",
"sec_num": "3.2."
},
{
"text": "Predicates from DBpedia appearing in SimpleDBpediaQA questions must be mapped to Wikidata ones before inclusion in ForwardQuestions. For instance, 'playerPosition' from DBpedia must be mapped to 'position played on team / speciality' (P413) in Wikidata. For some of the 6,236 Wikidata predicates, their equivalent in DBpedia is specified, but this happens only for 177 predicates out of the 365 ones appearing in SimpleDBpediaQA, leaving 188 predicates with no known DBpedia equivalents. We mapped these 188 remaining predicates using two approaches. Firstly, we looked for partial matches of the DBpedia labels with those from an online list of 1,872 Wikidata predicates with labels. 8 For example, for DBPedia's 'populationTotal' predicate, we could easily find the equivalent Wikidata predicate 'population'. Secondly, for nonmatched labels, we performed a manual word-based search on Wikidata and selected the closest matching predicate. The final mapping of predicates is provided with the For-wardQuestions data set in the mapping.json file of the Github repository (footnote 2). Each entry includes the DBpedia name and the matching Wikidata code, e.g. ('primeMinister', 'p6'). The first 177 predicates are those with explicit DBpedia equivalents in Wikidata, while the following 188 ones are those we mapped. In fact, we mapped many more predicates than those actually appearing in the selected questions, in anticipation of future needs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Converting DBpedia Predicates to Wikidata",
"sec_num": "3.3."
},
{
"text": "To sum up, we merged the forward questions from SimpleDBpediaQA and SimpleQuestionsWikidata, discarded backwards ones, removed duplicates (32% of the SimpleDBpediaQA), converted DBpedia predicates to Wikidata ones, and generated question templates by replacing subjects with <placeholder>. The ForwardQuestions data set contains 38,480 questions, having in total 94 different predicates. The various filtering operations, especially backward question removal and subject matching, have led us to keep only about 35% of the original SimpleQuestions-Wikidata entries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "3.4."
},
{
"text": "The most frequent predicate in ForwardQuestions is 'genre', which appears in more than 8,000 questions. Its meaning is quite general (akin to 'type' or 'category') and it can appear in triples concerning movies, books, music albums, artists, etc. 9 The next predicates by decreasing frequency are 'place of birth', 'country of citizenship', 'sex or gender', and 'position played on team / speciality'. The full list with frequencies is provided with the data set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "3.4."
},
{
"text": "The ForwardQuestions is intended to help with the generation of new questions, from knowledge triples not included in the set. We propose a method inspired from the rulebased baseline from Serban et al. (2016) , with the following differences. Their data set used Freebase, but we use Wikidata as the underlying triple store: our observations show that these predicates are often more precise, and specify sufficiently the type of the expected answer. For this reason, we created for each item in ForwardQuestions a template with a generic placeholder for the subject, unlike Serban et al. 2016who use type-specific placeholders such as <location placeholder>, which strongly reduces the number of questions available for generation. It is still an open question whether the size of ForwardQuestions allows the training of deep learning models; for the time being, we use the following template-based generation method. We generate a sample set of questions using 20 randomly selected templates among all those having the same predicate as the given triple, by replacing the placeholder of the question with the subject of the triple. 10 We then rank the questions using semantic similarity (4.1) and a language model (4.2).",
"cite_spans": [
{
"start": 189,
"end": 209,
"text": "Serban et al. (2016)",
"ref_id": "BIBREF17"
},
{
"start": 1135,
"end": 1137,
"text": "10",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Question Generation from Triples",
"sec_num": "4."
},
{
"text": "The main issue to address can be illustrated with the following example. If we use a template such as \"What kind of music does <placeholder> play?\", derived from ('John Duffey', 'genre', 'bluegrass music'), but we want to generate a question based on the new triple ('Claude Monet', 'genre', 'portrait'), then we obtain the question \"What kind of music does Claude Monet play?\", with the expected correct answer being 'portrait'. The question is incorrect because the rendering of the predicate 'genre' in the initial question is too specific and incompatible with the sense of the new triple. Alternatively, the reference to the subject in the template can also be too specific, e.g. if we use the template \"What genre is the tv program <placeholder>?\" with the triple above, we obtain the incorrect question \"What genre is the tv program Claude Monet?\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question Generation from Triples",
"sec_num": "4."
},
{
"text": "To avoid the issues exemplified above, we use semantic similarity between the word vectors provided by the word2vec library (Mikolov et al., 2013) . 11 We compare the 9 For instance, the triple ('John Duffey', 'genre', 'bluegrass music') has the associated question \"What kind of music does John Duffey play?\" 10 We filter out any parentheses from the subject's description, e.g. 'The Danton (1983 film)' is reduced to 'The Danton'.",
"cite_spans": [
{
"start": 124,
"end": 146,
"text": "(Mikolov et al., 2013)",
"ref_id": "BIBREF13"
},
{
"start": 149,
"end": 151,
"text": "11",
"ref_id": null
},
{
"start": 310,
"end": 312,
"text": "10",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Ranking with semantic similarity",
"sec_num": "4.1."
},
{
"text": "11 As implemented in the Gensim package (radimrehurek.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Ranking with semantic similarity",
"sec_num": "4.1."
},
{
"text": "com/gensim/models/word2vec.html) with Google's pre-trained model. As the templates are very short, and for average of the word vectors from the opening Wikipedia paragraph of the subject with the average of word vectors of the question template using cosine similarity. We also compute the similarity between the words of the template and those from the opening Wikipedia paragraph of the object, and retain the maximum of the two similarities as the semantic compatibility score of the question and the triple.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Ranking with semantic similarity",
"sec_num": "4.1."
},
{
"text": "We observed that some ungrammatical questions obtained high semantic compatibility scores. Our second goal is thus to filter them out, using the KenLM language modeling software (Heafield et al., 2013) 12 with a language model for English provided by Zamia. 13 The perplexity score of the language model for the full question provides an estimate of the well-formedness of the question, i.e. a syntactic fluency score. Therefore, for a given triple, we combine the semantic and fluency scores, giving more weight to the first one, and select the question which has the highest average score.",
"cite_spans": [
{
"start": 178,
"end": 201,
"text": "(Heafield et al., 2013)",
"ref_id": "BIBREF8"
},
{
"start": 258,
"end": 260,
"text": "13",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Ranking with a language model",
"sec_num": "4.2."
},
{
"text": "The method for generating questions from arbitrary triples, using ForwardQuestions, can be used to build a quiz chatbot which prompts the user to select a topic. This topic is matched to a Wikipedia page, from which we find a reasonable number of interesting triples, from which questions can be generated as explained above. The chatbot proposes the questions one by one to the user, and compares their answers to the expected ones. 14 Figure 1 : Overview of the quiz generation chatbot.",
"cite_spans": [],
"ref_spans": [
{
"start": 437,
"end": 445,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Use of Questions for a Quiz Chatbot",
"sec_num": "5."
},
{
"text": "The Wikidata entry of the topic indicated by the user may contain a large triple set, but all of them have the topic itself as the subject. This reduces the diversity of questions, as all of them will bear on some property of the subject.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Selection of Triples for a Given Topic",
"sec_num": "5.1."
},
{
"text": "For instance, if the topic is 'Queen', the British rock band (Q15862), then all questions from triples on their Wikidata page will have 'Queen' as the subject: \"When was Queen founded?\", \"What country is Queen from?\", \"What is the music genre of Queen?\", etc.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Selection of Triples for a Given Topic",
"sec_num": "5.1."
},
{
"text": "To increase diversity, we select additional triples from entities that are related to the main one, as follows. First, we select a list of Wikidata predicates that tend to connect the subject to meaningful objects, such as 'has part' (P527). Using the Wikidata entries of these objects, we select additional triples. For instance, from ('Queen', 'has part', 'Freddie Mercury') we infer that 'Freddie Mercury' is a related entity, and find the triple ('Freddie Mercury', 'religion', 'Zoroastrianism') which allows us to build the question: \"What was Freddie Mercury's religion?\" for the topic 'Queen'. We identified about 50 such predicates that allow the extension of the topic. However, some notions such as 'Rock Music' or 'Cooking' have few properties on their Wikidata pages, which is why we also use a second strategy, relying on the Wikipedia page of the subject. We use the \"See also\" or \"Related topics\" section to retrieve related subjects, but these sections are not always present and contain only a small list of subjects, of variable relevance. Therefore, we consider all hyperlinks from the page to other Wikipedia entries, and select the 20 first random subjects whose Wikidata entries contain at least 25 triples (so that each related subject has sufficient substance).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Selection of Triples for a Given Topic",
"sec_num": "5.1."
},
{
"text": "In addition, we merge the objects of triples that have the same subject and predicate, to obtain the list of all possible correct answers. For instance, from ('Barack Obama', 'child', 'Malia Obama') and ('Barack Obama', 'child', 'Sasha Obama'), we obtain a single triple for question generation, with a list of two acceptable answers. Finally, for any topic selected by the user, we randomly select 10 subjects found with the first method, and 20 found with the second one, with the aim of obtaining about 100 knowledge triples from which questions are generated.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Selection of Triples for a Given Topic",
"sec_num": "5.1."
},
{
"text": "We implemented a chatbot demo with the following components. Actions on Google 15 is the front-end proposed by Google to create apps for its Google Home smart speaker. Dialogflow 16 , which is connected to Actions, enables the design of simple dialogue models. The backend, running on one of our servers, is coded in Python with the Flask web development framework. As we found that our question generation is too slow to run in real time (taking several minutes on a mid-range computer, especially due to querying Wikipedia pages), we generated questions offline for several subjects (\"Olympic Games\", \"Politics of the United States\", \"Rock music\", \"Super Mario Bros.\", \"Switzerland\", \"The Legend of Zelda\", and \"World War II\"). The chatbot proposes to the user three randomly selected topics, among which one must be chosen. Sample questions (Q) and their correct answers (A) for \"World War II\" are: ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Implementation of a Chatbot Prototype",
"sec_num": "5.2."
},
{
"text": "\u2022 Q:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Implementation of a Chatbot Prototype",
"sec_num": "5.2."
},
{
"text": "The following evaluation protocol is targeted at the quality of the questions and their correct answer(s), and not at the usability of the chatbot, which depends also on the dialogue model and the speech recognition system. 17 For each triple and question, we asked human judges to rate the following quality aspects:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of the Questions",
"sec_num": "6."
},
{
"text": "1. Triple",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of the Questions",
"sec_num": "6."
},
{
"text": "\u2022 Importance of predicate and object. How interesting are the predicate and the object? For instance, for the triple (\"North America\", \"located in time zone\", {\"Hawaii-Aleutian Time Zone\", . . .}), the predicate and its value do not appear to be interesting.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of the Questions",
"sec_num": "6."
},
{
"text": "(a) Specification of the subject. For instance, in the question \"Who was responsible for the music in the film Super Mario Bros.?\", the specification of the subject is incorrect given that Super Mario Bros. is a video game, not a film.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "(b) Specification of the object. For instance, in \"Which city in Scotland did J. R. R. Tolkien come from?\", the specification of the expect answer (the triple's object) is wrong because Tolkien is from Birmingham, which is not in Scotland.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "(c) Formulation of the question. Is the question understandable and well-formed in English? This includes spelling mistakes. For instance, \"Who was the published the game Harry Potter?\" is poorly formulated.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "(d) Correctness of the expected answer.. For instance, for \"Who was the film Harry Potter and the Deathly Hallows based on the story by?\", the expected answer is 'Steve Kloves' (author of the screenplay), but one may estimate that 'J. K. Rowling' should be the correct answer, as she is the author of the original book.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "3. Overall: is this a good item for a quiz?",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "We asked four persons not familiar with the project to perform the following comparison. Given a knowledge triple, we show them the best question found by our method and the worst one (also according to our method) from a random subset of 20 questions generated for the triple. The human must rate each of the questions, without knowing their origin, on a five-point scale for each criterion. The goal is thus to measure the improvement brought by our method, with respect to a rather poor question, but still much stronger than the worst of all questions. For a set of 105 triples and 210 questions, we obtained 472 ratings. Table 3 : Average scores of the best question and of a random poor question on a five-point scale.",
"cite_spans": [],
"ref_spans": [
{
"start": 626,
"end": 633,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Question",
"sec_num": "2."
},
{
"text": "The results are presented in Table 3 . On all but one dimension, the best question shows clear improvement with respect to the poor one. The best questions score below poor ones regarding \"formulation\", but both scores are in fact rather high. With an overall quality of 3.35 out of 5, the questions are satisfactory, but there is also potential for progress. The largest improvement brought by our method (1 point out of 5) is for the specification of the subject, which is excellent for the best questions. The specification of the object (expected answer) is also improved. This was indeed one of our main goals, given that user-generated questions often include specifiers which become incorrect when the subject is replaced with another one. The improvement of the expected answer is quite similar to the one for the specification of the object, as these two elements are closely related. Finally, the relevance of the predicate + object is quite high, showing that the triple selection method is effective.",
"cite_spans": [],
"ref_spans": [
{
"start": 29,
"end": 36,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Score",
"sec_num": null
},
{
"text": "In this paper, we presented the ForwardQuestion data set, which we make available under the same CC-BY licence. The data set results from the conversion and combination of the SimpleDBpediaQA and SimpleQuestionsWikidata datasets, in particular by mapping predicates from Freebase to Wikidata. The 38,480 questions of the data set are accompanied by templates where the subject is replaced by a placeholder, in preparation for question generation that can be used in a quiz chatbot. The difficulties of triple conversion and predicate mapping strongly point to the need for interoperable semantic annotation in the realm of knowledge-based question generation. In future work on quiz generation, we aim to improve the relevance of the triples selected for a topic, as well as the diversity of the questions. While the size of the data set remains modest for use with deep learning generation methods, the triples could be used in conjunction with a pre-trained language model such as GPT-2 (Radford et al., 2019) or CTRL (Keskar et al., 2019) , to serve as adaptation data for neural question generation conditioned on the triples.",
"cite_spans": [
{
"start": 1020,
"end": 1041,
"text": "(Keskar et al., 2019)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7."
},
{
"text": "github.com/johannamelly/ForwardQuestions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "github.com/castorini/SimpleDBpediaQA 6 Data set available at github.com/askplatypus/ wikidata-simplequestions.7 github.com/WDAqua/WDAquaCore0Questions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Found atquarry.wmflabs.org/run/45013/ output/1/json. The gathering of all predicates in one list simply facilitated our search.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "simplicity reasons, we did not experiment with more elaborate paragraph representation models.12 github.com/kpu/kenlm 13 goofy.zamia.org/zamia-speech/lm/, large model 14 In other words, there are no follow-up questions, unlike Su et al.'s (2019) system, based on an ontology with RDF triples restricted to the dialogue domain.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The complete chatbot could be evaluated with the recent set of 17 metrics made available byCs\u00e1ky et al. (2019).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank the University of Applied Sciences of Western Switzerland (HES-SO) for the PLACAT grant that partly supported this work (AGP n. 82681), as well as the evaluators for the time they devoted to rating the questions submitted to them.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Farewell Freebase: Migrating the SimpleQuestions dataset to DBpedia",
"authors": [
{
"first": "M",
"middle": [],
"last": "Azmy",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Shi",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Ilyas",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 27th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "2093--2103",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Azmy, M., Shi, P., Lin, J., and Ilyas, I. (2018). Farewell Freebase: Migrating the SimpleQuestions dataset to DB- pedia. In Proceedings of the 27th International Confer- ence on Computational Linguistics, pages 2093-2103, Santa Fe, New Mexico, USA. Association for Compu- tational Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "More accurate question answering on freebase",
"authors": [
{
"first": "H",
"middle": [],
"last": "Bast",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Haussmann",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the Conference of Knowledge Management (CIKM'15)",
"volume": "",
"issue": "",
"pages": "1431--1440",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bast, H. and Haussmann, E. (2015). More accurate ques- tion answering on freebase. In Proceedings of the Con- ference of Knowledge Management (CIKM'15), pages 1431-1440, Melbourne, Australia.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Translating embeddings for modeling multi-relational data",
"authors": [
{
"first": "A",
"middle": [],
"last": "Bordes",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Usunier",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Garcia-Duran",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Weston",
"suffix": ""
},
{
"first": "Yakhnenko",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2013,
"venue": "Advances in Neural Information Processing Systems (NeurIPS)",
"volume": "",
"issue": "",
"pages": "2787--2795",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bordes, A., Usunier, N., Garcia-Duran, A., Weston, J., and Yakhnenko, O. (2013). Translating embeddings for modeling multi-relational data. In Advances in Neural Information Processing Systems (NeurIPS), pages 2787- 2795.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Large-scale simple question answering with memory networks",
"authors": [
{
"first": "A",
"middle": [],
"last": "Bordes",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Usunier",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Chopra",
"suffix": ""
},
{
"first": "Weston",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bordes, A., Usunier, N., Chopra, S., and Weston, J. (2015). Large-scale simple question answering with memory networks. CoRR, abs/1506.02075.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Towards topicto-question generation",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Chali",
"suffix": ""
},
{
"first": "S",
"middle": [
"A"
],
"last": "Hasan",
"suffix": ""
}
],
"year": 2015,
"venue": "Computational Linguistics",
"volume": "41",
"issue": "1",
"pages": "1--20",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chali, Y. and Hasan, S. A. (2015). Towards topic- to-question generation. Computational Linguistics, 41(1):1-20.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Improving neural conversational models with entropy-based data filtering",
"authors": [
{
"first": "R",
"middle": [],
"last": "Cs\u00e1ky",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Purgai",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Recski",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "5650--5669",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cs\u00e1ky, R., Purgai, P., and Recski, G. (2019). Improving neural conversational models with entropy-based data filtering. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5650-5669, Florence, Italy. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Question answering benchmarks for Wikidata",
"authors": [
{
"first": "D",
"middle": [],
"last": "Diefenbach",
"suffix": ""
},
{
"first": "T",
"middle": [
"P"
],
"last": "Tanon",
"suffix": ""
},
{
"first": "K",
"middle": [
"D"
],
"last": "Singh",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Maret",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the ISWC 2017 Posters & Demonstrations and Industry Tracks co-located with 16th International Semantic Web Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Diefenbach, D., Tanon, T. P., Singh, K. D., and Maret, P. (2017). Question answering benchmarks for Wikidata. In Proceedings of the ISWC 2017 Posters & Demonstra- tions and Industry Tracks co-located with 16th Interna- tional Semantic Web Conference, Vienna, Austria.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Question generation for question answering",
"authors": [
{
"first": "N",
"middle": [],
"last": "Duan",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Tang",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Zhou",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "866--874",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Duan, N., Tang, D., Chen, P., and Zhou, M. (2017). Ques- tion generation for question answering. In Proceedings of the 2017 Conference on Empirical Methods in Natu- ral Language Processing, pages 866-874, Copenhagen, Denmark. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Scalable modified Kneser-Ney language model estimation",
"authors": [
{
"first": "K",
"middle": [],
"last": "Heafield",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Pouzyrevsky",
"suffix": ""
},
{
"first": "J",
"middle": [
"H"
],
"last": "Clark",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (ACL 2013)",
"volume": "",
"issue": "",
"pages": "690--696",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Heafield, K., Pouzyrevsky, I., Clark, J. H., and Koehn, P. (2013). Scalable modified Kneser-Ney language model estimation. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (ACL 2013), pages 690-696, Sofia, Bulgaria.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Good question! Statistical ranking for question generation",
"authors": [
{
"first": "M",
"middle": [],
"last": "Heilman",
"suffix": ""
},
{
"first": "N",
"middle": [
"A"
],
"last": "Smith",
"suffix": ""
}
],
"year": 2010,
"venue": "Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "609--617",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Heilman, M. and Smith, N. A. (2010). Good question! Sta- tistical ranking for question generation. In Human Lan- guage Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Compu- tational Linguistics, pages 609-617, Los Angeles, CA, USA, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "CTRL: a conditional transformer language model for controllable generation",
"authors": [
{
"first": "N",
"middle": [
"S"
],
"last": "Keskar",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Mccann",
"suffix": ""
},
{
"first": "L",
"middle": [
"R"
],
"last": "Varshney",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Xiong",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Socher",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Keskar, N. S., McCann, B., Varshney, L. R., Xiong, C., and Socher, R. (2019). CTRL: a conditional transformer lan- guage model for controllable generation. arXiv cs.CL, 1909.05858.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Automatic question generation using relative pronouns and adverbs",
"authors": [
{
"first": "P",
"middle": [],
"last": "Khullar",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Rachna",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Hase",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Shrivastava",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of ACL 2018, Student Research Workshop",
"volume": "",
"issue": "",
"pages": "153--158",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Khullar, P., Rachna, K., Hase, M., and Shrivastava, M. (2018). Automatic question generation using relative pronouns and adverbs. In Proceedings of ACL 2018, Stu- dent Research Workshop, pages 153-158, Melbourne, Australia. Association for Computational Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Getting the most out of Wikidata: Semantic technology usage in Wikipedia's knowledge graph",
"authors": [
{
"first": "S",
"middle": [],
"last": "Malyshev",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Kr\u00f6tzsch",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Gonz\u00e1lez",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Gonsior",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Bielefeldt",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 17th International Semantic Web Conference (ISWC 2018)",
"volume": "11137",
"issue": "",
"pages": "376--394",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Malyshev, S., Kr\u00f6tzsch, M., Gonz\u00e1lez, L., Gonsior, J., and Bielefeldt, A. (2018). Getting the most out of Wikidata: Semantic technology usage in Wikipedia's knowledge graph. In Proceedings of the 17th International Seman- tic Web Conference (ISWC 2018), LNCS volume 11137, pages 376-394, Monterey, CA, USA.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Distributed representations of words and phrases and their compositionality",
"authors": [
{
"first": "T",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Sutskever",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "G",
"middle": [
"S"
],
"last": "Corrado",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2013,
"venue": "Advances in Neural Information Processing Systems (NIPS)",
"volume": "",
"issue": "",
"pages": "3111--3119",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., and Dean, J. (2013). Distributed representations of words and phrases and their compositionality. In Advances in Neural Information Processing Systems (NIPS), pages 3111-3119.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Chatbot as an intelligent personal assistant for mobile language learning",
"authors": [
{
"first": "X",
"middle": [
"L"
],
"last": "Pham",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Pham",
"suffix": ""
},
{
"first": "Q",
"middle": [
"M"
],
"last": "Nguyen",
"suffix": ""
},
{
"first": "T",
"middle": [
"H"
],
"last": "Nguyen",
"suffix": ""
},
{
"first": "T",
"middle": [
"T H"
],
"last": "Cao",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 2018 2nd International Conference on Education and E-Learning",
"volume": "",
"issue": "",
"pages": "16--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pham, X. L., Pham, T., Nguyen, Q. M., Nguyen, T. H., and Cao, T. T. H. (2018). Chatbot as an intelligent personal assistant for mobile language learning. In Proceedings of the 2018 2nd International Conference on Education and E-Learning (ICEEL 2018), page 16-21, Bali, Indonesia. Association for Computing Machinery.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Language models are unsupervised multitask learners",
"authors": [
{
"first": "A",
"middle": [],
"last": "Radford",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Child",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Luan",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Amodei",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Sutskever",
"suffix": ""
}
],
"year": 2019,
"venue": "OpenAI Blog",
"volume": "",
"issue": "8",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., and Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI Blog, 1(8).",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "SQuAD: 100,000+ questions for machine comprehension of text",
"authors": [
{
"first": "P",
"middle": [],
"last": "Rajpurkar",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Lopyrev",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Liang",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "2383--2392",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rajpurkar, P., Zhang, J., Lopyrev, K., and Liang, P. (2016). SQuAD: 100,000+ questions for machine comprehen- sion of text. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2383-2392, Austin, Texas. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Generating factoid questions with recurrent neural networks: The 30M factoid question-answer corpus",
"authors": [
{
"first": "I",
"middle": [
"V"
],
"last": "Serban",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Garc\u00eda-Dur\u00e1n",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Gulcehre",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Ahn",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Chandar",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Courville",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "588--598",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Serban, I. V., Garc\u00eda-Dur\u00e1n, A., Gulcehre, C., Ahn, S., Chandar, S., Courville, A., and Bengio, Y. (2016). Gen- erating factoid questions with recurrent neural networks: The 30M factoid question-answer corpus. In Proceed- ings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 588-598, Berlin, Germany. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Follow-up question generation using neural tensor network-based domain ontology population in an interview coaching system",
"authors": [
{
"first": "M.-H",
"middle": [],
"last": "Su",
"suffix": ""
},
{
"first": "C.-H",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Chang",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of INTERSPEECH",
"volume": "",
"issue": "",
"pages": "4185--4189",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Su, M.-H., Wu, C.-H., and Chang, Y. (2019). Follow-up question generation using neural tensor network-based domain ontology population in an interview coaching system. In Proceedings of INTERSPEECH, pages 4185- 4189, Graz, Austria.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Answer-focused and position-aware neural question generation",
"authors": [
{
"first": "X",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Lyu",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Ma",
"suffix": ""
},
{
"first": "Wang",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "3930--3939",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sun, X., Liu, J., Lyu, Y., He, W., Ma, Y., and Wang, S. (2018). Answer-focused and position-aware neural question generation. In Proceedings of the 2018 Con- ference on Empirical Methods in Natural Language Pro- cessing, pages 3930-3939, Brussels, Belgium. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Sixth open challenge on question answering over linked data (QALD-6)",
"authors": [
{
"first": "C",
"middle": [],
"last": "Unger",
"suffix": ""
},
{
"first": "A.-C",
"middle": [
"N"
],
"last": "Ngomo",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Cabrio",
"suffix": ""
}
],
"year": 2016,
"venue": "Semantic Web Challenges",
"volume": "641",
"issue": "",
"pages": "171--177",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Unger, C., Ngomo, A.-C. N., and Cabrio, E. (2016). Sixth open challenge on question answering over linked data (QALD-6). In Semantic Web Challenges, CCIS 641, pages 171-177, Berlin. Springer-Verlag.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Paragraphlevel neural question generation with maxout pointer and gated self-attention networks",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Ni",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Ding",
"suffix": ""
},
{
"first": "Q",
"middle": [],
"last": "Ke",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "3901--3910",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhao, Y., Ni, X., Ding, Y., and Ke, Q. (2018). Paragraph- level neural question generation with maxout pointer and gated self-attention networks. In Proceedings of the 2018 Conference on Empirical Methods in Natural Lan- guage Processing, pages 3901-3910, Brussels, Belgium. Association for Computational Linguistics.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Question-type driven question generation",
"authors": [
{
"first": "W",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Wu",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
"volume": "",
"issue": "",
"pages": "6032--6037",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhou, W., Zhang, M., and Wu, Y. (2019). Question-type driven question generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 6032-6037, Hong Kong, China. Association for Compu- tational Linguistics.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"num": null,
"html": null,
"type_str": "table",
"content": "<table><tr><td>Direction</td><td>forward</td></tr><tr><td>Constraint</td><td>null</td></tr><tr><td>ID</td><td>00042</td></tr><tr><td>Query</td><td>which home is an example of italianate architecture?</td></tr><tr><td>Subject</td><td/></tr><tr><td>Predicate List</td><td/></tr><tr><td>Predicate</td><td>http://dbpedia.org/ontology/architecturalStyle</td></tr><tr><td>Direction</td><td>backward</td></tr><tr><td>Constraint</td><td>http://dbpedia.org/ontology/ArchitecturalStructure</td></tr></table>",
"text": "). 7 ID 00035 Query what is the place of birth of sam edwards? Subject http://dbpedia.org/resource/Sam_Edwards_(physicist) Freebase Predicate www.freebase.com/people/person/place_of_birth Predicate List Predicate http://dbpedia.org/ontology/birthPlace ://dbpedia.org/resource/Italianate_architecture Freebase Predicate www.freebase.com/architecture/architectural_style/examples"
},
"TABREF1": {
"num": null,
"html": null,
"type_str": "table",
"content": "<table><tr><td>Subject</td><td>Pred. Object</td><td>Question</td></tr><tr><td colspan=\"3\">Q2747238 P413 Q5059480 what position does</td></tr><tr><td/><td/><td>carlos gomez play?</td></tr><tr><td colspan=\"2\">Q1176417 P136 Q37073</td><td>what type of mu-</td></tr><tr><td/><td/><td>sic does david ruf-</td></tr><tr><td/><td/><td>fin play</td></tr><tr><td>Q12152</td><td colspan=\"2\">R509 Q6371569 which swiss con-</td></tr><tr><td/><td/><td>ductor's cause of</td></tr><tr><td/><td/><td>death is myocardial</td></tr><tr><td/><td/><td>infarction?</td></tr></table>",
"text": "Examples of SimpleDBpediaQA entries: a forward and a backward question."
},
"TABREF2": {
"num": null,
"html": null,
"type_str": "table",
"content": "<table/>",
"text": ""
},
"TABREF3": {
"num": null,
"html": null,
"type_str": "table",
"content": "<table><tr><td>\u2022 Q: Who was the developer for A6M Zero?</td></tr><tr><td>A: Mitsubishi Heavy Industries</td></tr><tr><td>\u2022 Q: What was the cause of death for Adolf Hitler?</td></tr><tr><td>A: shot to the head, suicide by shooting</td></tr></table>",
"text": "Which country was involved in the Eastern Front? 15 console.actions.google.com 16 dialogflow.com A: Nazi Germany, Soviet Union, . . . \u2022 Q: Who was one of the major figures in the Attack on Pearl Harbor? A: Husband Edward Kimmel, Mitsuo Fuchida, . . ."
}
}
}
} |