File size: 101,025 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
{
    "paper_id": "P17-1015",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:18:56.790075Z"
    },
    "title": "Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems",
    "authors": [
        {
            "first": "Wang",
            "middle": [],
            "last": "Ling",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Oxford",
                "location": {}
            },
            "email": "lingwang@google.com"
        },
        {
            "first": "Dani",
            "middle": [],
            "last": "Yogatama",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Oxford",
                "location": {}
            },
            "email": "dyogatama@google.com"
        },
        {
            "first": "Chris",
            "middle": [],
            "last": "Dyer",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Oxford",
                "location": {}
            },
            "email": "cdyer@google.com"
        },
        {
            "first": "Phil",
            "middle": [],
            "last": "Blunsom",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Oxford",
                "location": {}
            },
            "email": "pblunsom@google.com"
        },
        {
            "first": "\u2666",
            "middle": [
                "\u2660"
            ],
            "last": "Deepmind",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Oxford",
                "location": {}
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Solving algebraic word problems requires executing a series of arithmetic operations-a program-to obtain a final answer. However, since programs can be arbitrarily complicated, inducing them directly from question-answer pairs is a formidable challenge. To make this task more feasible, we solve these problems by generating answer rationales, sequences of natural language and human-readable mathematical expressions that derive the final answer through a series of small steps. Although rationales do not explicitly specify programs, they provide a scaffolding for their structure via intermediate milestones. To evaluate our approach, we have created a new 100,000-sample dataset of questions, answers and rationales. Experimental results show that indirect supervision of program learning via answer rationales is a promising strategy for inducing arithmetic programs.",
    "pdf_parse": {
        "paper_id": "P17-1015",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Solving algebraic word problems requires executing a series of arithmetic operations-a program-to obtain a final answer. However, since programs can be arbitrarily complicated, inducing them directly from question-answer pairs is a formidable challenge. To make this task more feasible, we solve these problems by generating answer rationales, sequences of natural language and human-readable mathematical expressions that derive the final answer through a series of small steps. Although rationales do not explicitly specify programs, they provide a scaffolding for their structure via intermediate milestones. To evaluate our approach, we have created a new 100,000-sample dataset of questions, answers and rationales. Experimental results show that indirect supervision of program learning via answer rationales is a promising strategy for inducing arithmetic programs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Behaving intelligently often requires mathematical reasoning. Shopkeepers calculate change, tax, and sale prices; agriculturists calculate the proper amounts of fertilizers, pesticides, and water for their crops; and managers analyze productivity. Even determining whether you have enough money to pay for a list of items requires applying addition, multiplication, and comparison. Solving these tasks is challenging as it involves recognizing how goals, entities, and quantities in the real-world map onto a mathematical formalization, computing the solution, and mapping the solution back onto the world. As a proxy for the richness of the real world, a series of papers have used natural language specifications of algebraic word problems, and solved these by either learning to fill in templates that can be solved with equation solvers (Hosseini et al., 2014; or inferring and modeling operation sequences (programs) that lead to the final answer (Roy and Roth, 2015) .",
                "cite_spans": [
                    {
                        "start": 841,
                        "end": 864,
                        "text": "(Hosseini et al., 2014;",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 952,
                        "end": 972,
                        "text": "(Roy and Roth, 2015)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we learn to solve algebraic word problems by inducing and modeling programs that generate not only the answer, but an answer rationale, a natural language explanation interspersed with algebraic expressions justifying the overall solution. Such rationales are what examiners require from students in order to demonstrate understanding of the problem solution; they play the very same role in our task. Not only do natural language rationales enhance model interpretability, but they provide a coarse guide to the structure of the arithmetic programs that must be executed. In fact the learner we propose (which relies on a heuristic search; \u00a74) fails to solve this task without modeling the rationales-the search space is too unconstrained.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "This work is thus related to models that can explain or rationalize their decisions (Hendricks et al., 2016; Harrison et al., 2017) . However, the use of rationales in this work is quite different from the role they play in most prior work, where interpretation models are trained to generate plausible sounding (but not necessarily accurate) posthoc descriptions of the decision making process they used. In this work, the rationale is generated as a latent variable that gives rise to the answer-it is thus a more faithful representation of the steps used in computing the answer.",
                "cite_spans": [
                    {
                        "start": 84,
                        "end": 108,
                        "text": "(Hendricks et al., 2016;",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 109,
                        "end": 131,
                        "text": "Harrison et al., 2017)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "This paper makes three contributions. First, we have created a new dataset with more than 100,000 algebraic word problems that includes both answers and natural language answer rationales ( \u00a72).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Problem 1: Question: Two trains running in opposite directions cross a man standing on the platform in 27 seconds and 17 seconds respectively and they cross each other in 23 seconds. The ratio of their speeds is: Options: A) 3/7 B) 3/2 C) 3/88 D) 3/8 E) 2/2 Rationale: Let the speeds of the two trains be x m/sec and y m/sec respectively. Then, length of the first train = 27x meters, and length of the second train = 17 y meters. (27x + 17y) / (x + y) = 23 \u2192 27x + 17y = 23x + 23y \u2192 4x = 6y \u2192 x/y = 3/2. Correct Option: B Problem 2: Question: From a pack of 52 cards, two cards are drawn together at random. What is the probability of both the cards being kings? Options: A) 2/1223 B) 1/122 C) 1/221 D) 3/1253 E) 2/153 Rationale: Let s be the sample space. Then n(s) = 52C2 = 1326 E = event of getting 2 kings out of 4 n(E) = 4C2 = 6 P(E) = 6/1326 = 1/221 Answer is C Correct Option: C Problem 3: Question: For which of the following does p(a)\u2212p(b) = p(ab) for all values of a and b? Options:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 1 illustrates three representative instances",
                "sec_num": null
            },
            {
                "text": "A) p(x) = x 2 , B) p(x) = x/2, C) p(x) = x + 5, D) p(x) = 2x1, E) p(x) = |x|",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 1 illustrates three representative instances",
                "sec_num": null
            },
            {
                "text": "Rationale: To solve this easiest way is just put the value and see that if it equals or not. with option A. p(a) = a 2 and p(b from the dataset. Second, we propose a sequence to sequence model that generates a sequence of instructions that, when executed, generates the rationale; only after this is the answer chosen ( \u00a73). Since the target program is not given in the training data (most obviously, its specific form will depend on the operations that are supported by the program interpreter); the third contribution is thus a technique for inferring programs that generate a rationale and, ultimately, the answer. Even constrained by a text rationale, the search space of possible programs is quite large, and we employ a heuristic search to find plausible next steps to guide the search for programs ( \u00a74). Empirically, we are able to show that state-of-the-art sequence to sequence models are unable to perform above chance on this task, but that our model doubles the accuracy of the baseline ( \u00a76).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 1 illustrates three representative instances",
                "sec_num": null
            },
            {
                "text": ") = b 2 so L.H.S = a 2 \u2212 b 2 and R.H.S = (a \u2212 b) 2 \u2192 a 2 + b 2 \u2212 2ab. so L.H.S not equal to R.H.S with option B. p(a) = a/2 and p(b) = b/2 L.H.S = a/2 \u2212 b/2 \u2192 1/2(a \u2212 b) R.H.S = (a \u2212 b)/2 so L",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 1 illustrates three representative instances",
                "sec_num": null
            },
            {
                "text": "We built a dataset with 100,000 problems with the annotations shown in Figure 1 . Each question is decomposed in four parts, two inputs and two outputs: the description of the problem, which we will denote as the question, and the possible (multiple choice) answer options, denoted as options. Our goal is to generate the description of the rationale used to reach the correct answer, denoted as rationale and the correct option label. Problem 1 illustrates an example of an algebra problem, which must be translated into an expression (i.e., (27x + 17y)/(x + y) = 23) and then the desired quantity (x/y) solved for. Problem 2 is an example that could be solved by multi-step arithmetic operations proposed in (Roy and Roth, 2015) . Finally, Problem 3 describes a problem that is solved by testing each of the options, which has not been addressed in the past.",
                "cite_spans": [
                    {
                        "start": 710,
                        "end": 730,
                        "text": "(Roy and Roth, 2015)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 71,
                        "end": 79,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "2"
            },
            {
                "text": "We first collect a set of 34,202 seed problems that consist of multiple option math questions covering a broad range of topics and difficulty levels. Examples of exams with such problems include the GMAT (Graduate Management Admission Test) and GRE (General Test). Many websites contain example math questions in such exams, where the answer is supported by a rationale.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Construction",
                "sec_num": "2.1"
            },
            {
                "text": "Next, we turned to crowdsourcing to generate new questions. We create a task where users are presented with a set of 5 questions from our seed dataset. Then, we ask the Turker to choose one of the questions and write a similar question. We also force the answers and rationale to differ from the original question in order to avoid paraphrases of the original question. Once again, we manually check a subset of the jobs for each Turker for quality control. The type of questions generated using this method vary. Some turkers propose small changes in the values of the questions (e.g., changing the equality p(a)p(b) = p(ab) in Problem 3 to a different equality is a valid question, as long as the rationale and options are rewritten to reflect the change). We designate these as replica problems as the natural language used in the question and rationales tend to be only minimally unaltered. Others propose new problems in the same topic where the generated questions tend to differ more radically from existing ones. Some Turkers also copy math problems available on the web, and we define in the instructions that this is not allowed, as it will generate multiple copies of the same problem in the dataset if two or more Turkers copy from the same resource. These Turkers can be detected by checking the nearest neighbours within the collected datasets as problems obtained from online resources are frequently submitted by more than one Turker. Using this method, we obtained 70,318 additional questions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Construction",
                "sec_num": "2.1"
            },
            {
                "text": "Descriptive statistics of the dataset is shown in Figure 1 . In total, we collected 104,519 problems (34,202 seed problems and 70,318 crowdsourced problems). We removed 500 problems as heldout set (250 for development and 250 for testing). As replicas of the heldout problems may be present in the training set, these were removed manually by listing for each heldout instance the closest problems in the training set in terms of character-based Levenstein distance. After filtering, 100,949 problems remained in the training set. We also show the average number of tokens (total number of tokens in the question, options and rationale) and the vocabulary size of the questions and rationales. Finally, we provide the same statistics exclusively for tokens that are numeric values and tokens that are not. Figure 2 shows the distribution of examples based on the total number of tokens. We can see that most examples consist of 30 to 500 tokens, but there are also extremely long examples with more than 1000 tokens in our dataset.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 50,
                        "end": 58,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    },
                    {
                        "start": 806,
                        "end": 814,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Statistics",
                "sec_num": "2.2"
            },
            {
                "text": "Generating rationales for math problems is challenging as it requires models that learn to perform math operations at a finer granularity as each step within the solution must be explained. 17y)/(x + y) = 23 must be solved to obtain the answer. In previous work , this could be done by feeding the equation into an expression solver to obtain x/y = 3/2. However, this would skip the intermediate steps 27x + 17y = 23x + 23y and 4x = 6y, which must also be generated in our problem. We propose a model that jointly learns to generate the text in the rationale, and to perform the math operations required to solve the problem. This is done by generating a program, containing both instructions that generate output and instructions that simply generate intermediate values used by following instructions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Model",
                "sec_num": "3"
            },
            {
                "text": "In traditional sequence to sequence models (Sutskever et al., 2014; Bahdanau et al., 2014) , the goal is to predict the output sequence y = y 1 , . . . , y |y| from the input sequence x = x 1 , . . . , x |x| , with lengths |y| and |x|.",
                "cite_spans": [
                    {
                        "start": 43,
                        "end": 67,
                        "text": "(Sutskever et al., 2014;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 68,
                        "end": 90,
                        "text": "Bahdanau et al., 2014)",
                        "ref_id": "BIBREF1"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "In our particular problem, we are given the problem and the set of options, and wish to predict the rationale and the correct option. We set x as the sequence of words in the problem, concatenated with words in each of the options separated by a special tag. Note that knowledge about the possible options is required as some problems are solved by the process of elimination or by testing each of the options (e.g. Problem 3). We wish to generate y, which is the sequence of words in the rationale. We also append the correct option as the last word in y, which is interpreted as the chosen option. For example, y in Problem 1 is \"Let the . . . = 3/2 . EOR B EOS \", whereas in Problem 2 it is \"Let s be . . . Answer is C EOR C EOS \", where \" EOS \" is the end of sentence symbol and \" EOR \" is the end of rationale symbol.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "i x z v r From Id(\"Let\") Let y1 a Id(\"s\") s y2 pack",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Id(\"be\") be y3 of Id(\"the\") the y4 52",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Id(\"sample\") sample y5 cards Id(\"space\") space y6 , Id(\".\") . y7 two Id(\"\\n\") \\n y8 cards Id(\"Then\") Then y9 are",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Id(\"n\") n y10 drawn Id(\"(\") ( y11 together Id(\"s\") s y12 at Id(\")\") ) y13 random Id(\"=\") = y14 . Str to Float(x5) 52 m1 What",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Float to Str(m1) 52 y15 is Id(\"C\") C y16 the Id(\"2\") 2 y17 probability Id(\"=\") = y18 of Str to Float(y17) 2 m2 both Choose(m1,m2) 1326 m3 cards",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Float to Str(m3) 1326 y19 being Id(\"E\") E y20 kings Id(\"=\") = y21 ?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "Id(\"event\") event ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "y22 <O> Id(\"of\") of y23 A) Id(\"getting\") getting y24 2/1223 Id(\"2\") 2 y25 <O> Id(\"kings\") kings y26 B) Id(\"out\") out y27 1/122 Id(\"of\") of y28 . . . . . . . . . . . . . . . |z| Id(\" EOS \") EOS y |y|",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problem Definition",
                "sec_num": "3.1"
            },
            {
                "text": "We wish to generate a latent sequence of program instructions, z = z 1 , . . . , z |z| , with length |z|, that will generate y when executed. We express z as a program that can access x, y, and the memory buffer m. Upon finishing execution we expect that the sequence of output tokens to be placed in the output vector y. Table 2 illustrates an example of a sequence of instructions that would generate an excerpt from Problem 2, where columns x, z, v, and r denote the input sequence, the instruction sequence (program), the values of executing the instruction, and where each value v i is written (i.e., either to the output or to the memory). In this example, instructions from indexes 1 to 14 simply fill each position with the observed output y 1 , . . . , y 14 with a string, where the Id operation simply returns its parame-ter without applying any operation. As such, running this operation is analogous to generating a word by sampling from a softmax over a vocabulary. However, instruction z 15 reads the input word x 5 , 52, and applies the operation Str to Float, which converts the word 52 into a floating point number, and the same is done for instruction z 20 , which reads a previously generated output word y 17 . Unlike, instructions z 1 , . . . , z 14 , these operations write to the external memory m, which stores intermediate values. A more sophisticated instruction-which shows some of the power of our model-is z 21 = Choose(m 1 , m 2 ) \u2192 m 3 which evaluates m 1 m 2 and stores the result in m 3 . This process repeats until the model generates the end-of-sentence symbol. The last token of the program as said previously must generate the correct option value, from \"A\" to \"E\". By training a model to generate instructions that can manipulate existing tokens, the model benefits from the additional expressiveness needed to solve math problems within the generation process. In total we define 22 different operations, 13 of which are frequently used operations when solving math problems. These are: Id, Add, Subtract, Multiply, Divide, Power, Log, Sqrt, Sine, Cosine, Tangent, Factorial, and Choose (number of combinations). We also provide 2 operations to convert between Radians and Degrees, as these are needed for the sine, cosine and tangent operations. There are 6 operations that convert floating point numbers into strings and vice-versa. These include the Str to Float and Float to Str operations described previously, as well as operations which convert between floating point numbers and fractions, since in many math problems the answers are in the form \"3/4\". For the same reason, an operation to convert between a floating point number and number grouped in thousands is also used (e.g. 1000000 to \"1,000,000\" or \"1.000.000\"). Finally, we define an operation (Check) that given the input string, searches through the list of options and returns a string with the option index in {\"A\", \"B\", \"C\", \"D\", \"E\"}. If the input value does not match any of the options, or more than one option contains that value, it cannot be applied. For instance, in Problem 2, once the correct probability \"1/221\" is generated, by applying the check operation to this number we can obtain correct option \"C\". Figure 3 : Illustration of the generation process of a single instruction tuple at timestamp i.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 322,
                        "end": 329,
                        "text": "Table 2",
                        "ref_id": "TABREF2"
                    },
                    {
                        "start": 3228,
                        "end": 3236,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Generating Programs to Generate Rationales",
                "sec_num": "3.2"
            },
            {
                "text": "h i softmax o i r i softmax r i q i,j=1 softmax q ij softmax copy input a ij q i,j+1 copy output h i+1 j < argc(o i )? v i execute",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating Programs to Generate Rationales",
                "sec_num": "3.2"
            },
            {
                "text": "In our model, programs consist of sequences of instructions, z. We turn now to how we model each z i , conditional on the text program specification, and the program's history. The instruction z i is a tuple consisting of an operation (o i ), an ordered sequence of its arguments (a i ), and a decision about where its results will be placed (r i ) (is it appended in the output y or in a memory buffer m?), and the result of applying the operation to its arguments (",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "v i ). That is, z i = (o i , r i , a i , v i ).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "Formally, o i is an element of the pre-specified set of operations O, which contains, for example add, div, Str to Float, etc. The number of arguments required by o i is given by argc(o i ), e.g., argc(add) = 2 and argc(log) = 1. The arguments are a i = a i,1 , . . . , a i,argc(o i ) . An instruction will generate a return value v i upon execution, which will either be placed in the output y or hidden. This decision is controlled by r i . We define the instruction probability as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "p(o i , a i , r i ,v i | z <i , x, y, m) = p(o i | z <i , x) \u00d7 p(r i | z <i , x, o i )\u00d7 argc(o i ) j=1 p(a i,j | z <i , x, o i , m, y)\u00d7 [v i = apply(o i , a)],",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "where [p] evaluates to 1 if p is true and 0 otherwise, and apply(f, x) evaluates the operation f with arguments x. Note that the apply function is not learned, but pre-defined.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "The network used to generate an instruction at a given timestamp i is illustrated in Figure 3 . We first use the recurrent state In order to predict r i , we generate a new hidden state r i , which is a function of the current program context h i , and an embedding of the current predicted operation, o i . As the output can either be placed in the memory m or the output y, we compute the probability p(r",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 85,
                        "end": 93,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "h i to generate p(o i | z <i , x) = softmax o i \u2208O (h i ),",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "i = OUTPUT | z <i , x, o i ) = \u03c3(r i \u2022 w r + b r ),",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "where \u03c3 is the logistic sigmoid function. If r i = OUTPUT, v i is appended to the output y; otherwise it is appended to the memory m.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "Once we generate r i , we must predict a i , the argc(o i )-length sequence of arguments that operation o i requires. The jth argument a i,j can be either generated from a softmax over the vocabulary, copied from the input vector x, or copied from previously generated values in the output y or memory m. This decision is modeled using a latent predictor network (Ling et al., 2016) , where the control over which method used to generate a i,j is governed by a latent variable q i,j \u2208 {SOFTMAX, COPY-INPUT, COPY-OUTPUT}. Similar to when predicting r i , in order to make this choice, we also generate a new hidden state for each argument slot j, denoted by q i,j with an LSTM, feeding the previous argument in at each time step, and initializing it with r i and by reading the predicted value of the output r i .",
                "cite_spans": [
                    {
                        "start": 363,
                        "end": 382,
                        "text": "(Ling et al., 2016)",
                        "ref_id": "BIBREF12"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 If q i,j = SOFTMAX, a i,j is generated by sampling from a softmax over the vocabulary Y,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "p(a i,j | q i,j = SOFTMAX) = softmax a i,j \u2208Y (q i,j ).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "This corresponds to a case where a string is used as argument (e.g. y 1 =\"Let\").",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 If q i,j = COPY-INPUT, a i,j is obtained by copying an element from the input vector with a pointer network (Vinyals et al., 2015) over input words x 1 , . . . , x |x| , represented by their encoder LSTM state u 1 , . . . , u |x| . As such, we compute the probability distribution over input words as:",
                "cite_spans": [
                    {
                        "start": 110,
                        "end": 132,
                        "text": "(Vinyals et al., 2015)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "p(a i,j | q i,j =COPY-INPUT) = (1) softmax a i,j \u2208x 1 ,...,x |x| f (u a i,j , q i,j )",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "Function f computes the affinity of each token x a i,j and the current output context q i,j . A common implementation of f , which we follow, is to apply a linear projection from [u a i,j ; q i,j ] into a fixed size vector (where [u; v] is vector concatenation), followed by a tanh and a linear projection into a single value.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "\u2022 If q i,j = COPY-OUTPUT, the model copies from either the output y or the memory m. This is equivalent to finding the instruction z i , where the value was generated. Once again, we define a pointer network that points to the output instructions and define the distribution over previously generated instructions as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "p(a i,j | q i,j =COPY-OUTPUT) = softmax a i,j \u2208z 1 ,...,z i\u22121 f (h a i,j , q i,j )",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "Here, the affinity is computed using the decoder state h a i,j and the current state q i,j .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "Finally, we embed the argument a i,j 1 and the state q i,j to generate the next state q i,j+1 . Once all arguments for o i are generated, the operation is executed to obtain v i . Then, the embedding of v i , the final state of the instruction q i,|a i | and the previous state h i are used to generate the state at the next timestamp h i+1 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Generating and Executing Instructions",
                "sec_num": "3.3"
            },
            {
                "text": "The set of instructions z that will generate y is unobserved. Thus, given x we optimize the marginal probability function:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "p(y | x) = z\u2208Z p(y | z)p(z | x) = z\u2208Z(y) p(z | x),",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "where p(y | z) is the Kronecker delta function \u03b4 e(z),y , which is 1 if the execution of z, denoted as e(z), generates y and 0 otherwise. Thus, we can redefine p(y|x), the marginal over all programs Z, as a marginal over programs that would generate y, defined as Z(y). As marginalizing over z \u2208 Z(y) is intractable, we approximate the marginal by generating samples from our model. Denote the set of samples that are generated by\u1e90(y). We maximize z \u2208\u1e90(y)p(z|x).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "However, generating programs that generate y is not trivial, as randomly sampling from the RNN distribution over instructions at each timestamp is unlikely to generate a sequence z \u2208 Z(y).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "This is analogous to the question answering work in Liang et al. (2016) , where the query that generates the correct answer must be found during inference, and training proved to be difficult without supervision. In Roy and Roth (2015) this problem is also addressed by adding prior knowledge to constrain the exponential space.",
                "cite_spans": [
                    {
                        "start": 52,
                        "end": 71,
                        "text": "Liang et al. (2016)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 216,
                        "end": 235,
                        "text": "Roy and Roth (2015)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "In our work, we leverage the fact that we are generating rationales, where there is a sense of progression within the rationale. That is, we assume that the rationale solves the problem step by step. For instance, in Problem 2, the rationale first describes the number of combinations of two cards in a deck of 52 cards, then describes the number of combinations of two kings, and finally computes the probability of drawing two kings. Thus, while generating the final answer without the rationale requires a long sequence of latent instructions, generating each of the tokens of the rationale requires far less operations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "More formally, given the sequence z 1 , . . . , z i\u22121 generated so far, and the possible values for z i given by the network, denoted Z i , we wish to filter Z i to Z i (y k ), which denotes a set of possible options that contain at least one path capable of generating the next token at index k. Finding the set Z i (y k ) is achieved by testing all combinations of instructions that are possible with at most one level of indirection, and keeping those that can generate y k . This means that the model can only generate one intermediate value in memory (not including the operations that convert strings into floating point values and vice-versa).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "Decoding. During decoding we find the most likely sequence of instructions z given x, which can be performed with a stack-based decoder. However, it is important to refer that each generated instruction",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "z i = (o i , r i , a i,1 , . . . , a i,|a i | , v i )",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "must be executed to obtain v i . To avoid generating unexecutable code-e.g., log(0)-each hypothesis instruction is executed and removed if an error occurs. Finally, once the \" EOR \" tag is generated, we only allow instructions that would generate one of the option \"A\" to \"E\" to be generated, which guarantees that one of the options is chosen.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "for each instruction z i , the model needs to compute the probability distribution between all the attendable units c conditioned on the previous state h i\u22121 . For the attention model and input copy mechanisms, c = x 0,i\u22121 and for the output copy mechanism c = z. These operations generally involve an exponential number of matrix multiplications as the size of c and z grows. For instance, during the computation of the probabilities for the input copy mechanism in Equation 1, the affinity function f between the current context q and a given input u k is generally implemented by projecting u and q into a single vector followed by a non-linearity, which is projected into a single affinity value. Thus, for each possible input u, 3 matrix multiplications must be performed. Furthermore, for RNN unrolling, parameters and intermediate outputs for these operations must be replicated for each timestamp. Thus, as z becomes larger the attention and copy mechanisms quickly become a memory bottleneck as the computation graph becomes too large to fit on the GPU. In contrast, the sequence-to-sequence model proposed in (Sutskever et al., 2014) , does not suffer from these issues as each timestamp is dependent only on the previous state h i\u22121 .",
                "cite_spans": [
                    {
                        "start": 1119,
                        "end": 1143,
                        "text": "(Sutskever et al., 2014)",
                        "ref_id": "BIBREF20"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "To deal with this, we use a training method we call staged back-propagation which saves memory by considering slices of K tokens in z, rather than the full sequence. That is, to train on a minibatch where |z| = 300 with K = 100, we would actually train on 3 mini-batches, where the first batch would optimize for the first z 1:100 , the second for z 101:200 and the third for z 201:300 . The advantage of this method is that memory intensive operations, such as attention and the copy mechanism, only need to be unrolled for K steps, and K can be adjusted so that the computation graph fits in memory.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "However, unlike truncated back-propagation for language modeling, where context outside the scope of K is ignored, sequence-to-sequence models require global context. Thus, the sequence of states h is still built for the whole sequence z. Afterwards, we obtain a slice h j:j+K , and compute the attention vector. 2 Finally, the prediction of the instruction is conditioned on the LSTM state and the attention vector.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Inducing Programs while Learning",
                "sec_num": "4"
            },
            {
                "text": "We apply our model to the task of generating rationales for solutions to math problems, evaluating it on both the quality of the rationale and the ability of the model to obtain correct answers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "6"
            },
            {
                "text": "As the baseline we use the attention-based sequence to sequence model proposed by Bahdanau et al. (2014) , and proposed augmentations, allowing it to copy from the input (Ling et al., 2016) and from the output (Merity et al., 2016) .",
                "cite_spans": [
                    {
                        "start": 82,
                        "end": 104,
                        "text": "Bahdanau et al. (2014)",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 170,
                        "end": 189,
                        "text": "(Ling et al., 2016)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 210,
                        "end": 231,
                        "text": "(Merity et al., 2016)",
                        "ref_id": "BIBREF13"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Baselines",
                "sec_num": "6.1"
            },
            {
                "text": "We used a two-layer LSTM with a hidden size of H = 200, and word embeddings with size 200. The number of levels that the graph G is expanded during sampling D is set to 5. Decoding is performed with a beam of 200. As for the vocabulary of the softmax and embeddings, we keep the most frequent 20,000 word types, and replace the rest of the words with an unknown token. During training, the model only learns to predict a word as an unknown token, when there is no other alternative to generate the word.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Hyperparameters",
                "sec_num": "6.2"
            },
            {
                "text": "The evaluation of the rationales is performed with average sentence level perplexity and BLEU-4 (Papineni et al., 2002) . When a model cannot generate a token for perplexity computation, we predict unknown token. This benefits the baselines as they are less expressive. As the perplexity of our model is dependent on the latent program that is generated, we force decode our model to generate the rationale, while maximizing the probability of the program. This is analogous to the method used to obtain sample programs described in Section 4, but we choose the most likely instructions at each timestamp instead of sampling. Finally, the correctness of the answer is evaluated by computing the percentage of the questions, where the chosen option matches the correct one.",
                "cite_spans": [
                    {
                        "start": 96,
                        "end": 119,
                        "text": "(Papineni et al., 2002)",
                        "ref_id": "BIBREF15"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation Metrics",
                "sec_num": "6.3"
            },
            {
                "text": "The test set results, evaluated on perplexity, BLEU, and accuracy, are presented in Table 3 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 84,
                        "end": 91,
                        "text": "Table 3",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "6.4"
            },
            {
                "text": "Perplexity. In terms of perplexity, we observe that the regular sequence to sequence model fares poorly on this dataset, as the model requires the generation of many values that tend to be sparse. Adding an input copy mechanism greatly improves the perplexity as it allows the generation process to use values that were mentioned in the question. The output copying mechanism improves perplexity slightly over the input copy mechanism, as many values are repeated after their first occurrence. For instance, in Problem 2, the value \"1326\" is used twice, so even though the model cannot generate it easily in the first occurrence, the second one can simply be generated by copying the first one. We can observe that our model yields significant improvements over the baselines, demonstrating that the ability to generate new values by algebraic manipulation is essential in this task. An example of a program that is inferred is shown in Figure 4 . The graph was generated by finding the most likely program z that generates y. Each node isolates a value in x, m, or y, where arrows indicate an operation executed with the outgoing nodes as arguments and incoming node as the return of the operation. For simplicity, operations that copy or convert values (e.g. from string to float) were not included, but nodes that were copied/converted share the same color. Examples of tokens where our model can obtain the perplexity reduction are the values \"0.025\", \"0.023\", \"0.002\" and finally the answer \"E\" , as these cannot be copied from the input or output.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 937,
                        "end": 945,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "6.4"
            },
            {
                "text": "BLEU. We observe that the regular sequence to sequence model achieves a low BLEU score. In fact, due to the high perplexities the model generates very short rationales, which frequently consist of segments similar to \"Answer should be D\", as most rationales end with similar statements. By applying the copy mechanism the BLEU score improves substantially, as the model can define the variables that are used in the rationale. Interestingly, the output copy mechanism adds no further improvement in the perplexity evaluation. This is because during decoding all values that can be copied from the output are values that could have been generated by the model either from the softmax or the input copy mechanism. As such, adding an output copying mechanism adds little to the expressiveness of the model during decoding. Finally, our model can achieve the highest BLEU score as it has the mechanism to generate the intermediate and final values in the rationale.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "6.4"
            },
            {
                "text": "Accuracy. In terms of accuracy, we see that all baseline models obtain values close to chance (20%), indicating that they are completely unable to solve the problem. In contrast, we see that our model can solve problems at a rate that is significantly higher than chance, demonstrating the value of our program-driven approach, and its ability to learn to generate programs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "6.4"
            },
            {
                "text": "In general, the problems we solve correctly correspond to simple problems that can be solved in one or two operations. Examples include questions such as \"Billy cut up each cake into 10 slices, and ended up with 120 slices altogether. How many cakes did she cut up? A) 9 B) 7 C) 12 D) 14 E) 16\", which can be solved in a single step. In this case, our model predicts \"120 / 10 = 12 cakes. Answer is C\" as the rationale, which is reasonable.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "6.4"
            },
            {
                "text": "While we show that our model can outperform the models built up to date, generating complex rationales as those shown in Figure 1 correctly is still an unsolved problem, as each additional step adds complexity to the problem both during inference and decoding. Yet, this is the first result showing that it is possible to solve math problems in such a manner, and we believe this modeling approach and dataset will drive work on this problem.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 121,
                        "end": 129,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Discussion.",
                "sec_num": "6.5"
            },
            {
                "text": "Extensive efforts have been made in the domain of math problem solving (Hosseini et al., 2014; Roy and Roth, 2015) , which aim at obtaining the correct answer to a given math problem. Other work has focused on learning to map math expressions into formal languages (Roy et al., 2016) . We aim to generate natural language rationales, where the bindings between variables and the problem solving approach are mixed into a single generative model that attempts to solve the problem while explaining the approach taken.",
                "cite_spans": [
                    {
                        "start": 71,
                        "end": 94,
                        "text": "(Hosseini et al., 2014;",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 95,
                        "end": 114,
                        "text": "Roy and Roth, 2015)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 265,
                        "end": 283,
                        "text": "(Roy et al., 2016)",
                        "ref_id": "BIBREF19"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "7"
            },
            {
                "text": "Our approach is strongly tied with the work on sequence to sequence transduction using the encoder-decoder paradigm (Sutskever et al., 2014;  Bottle R contains 250 capsules and costs $ 6.25 . Bottle T contains 130 capsules and costs $ 2.99 . What is the difference between the cost per capsule for bottle R and the cost per capsule for bottle T ?",
                "cite_spans": [
                    {
                        "start": 116,
                        "end": 140,
                        "text": "(Sutskever et al., 2014;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 141,
                        "end": 141,
                        "text": "",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "7"
            },
            {
                "text": "(A) $ 0. Figure 4 : Illustration of the most likely latent program inferred by our algorithm to explain a held-out question-rationale pair. Bahdanau et al., 2014; Kalchbrenner and Blunsom, 2013) , and inherits ideas from the extensive literature on semantic parsing (Jones et al., 2012; Berant et al., 2013; Andreas et al., 2013; Quirk et al., 2015; Liang et al., 2016; Neelakantan et al., 2016) and program generation (Reed and de Freitas, 2016; Graves et al., 2016) , namely, the usage of an external memory, the application of different operators over values in the memory and the copying of stored values into the output sequence.",
                "cite_spans": [
                    {
                        "start": 140,
                        "end": 162,
                        "text": "Bahdanau et al., 2014;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 163,
                        "end": 194,
                        "text": "Kalchbrenner and Blunsom, 2013)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 266,
                        "end": 286,
                        "text": "(Jones et al., 2012;",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 287,
                        "end": 307,
                        "text": "Berant et al., 2013;",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 308,
                        "end": 329,
                        "text": "Andreas et al., 2013;",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 330,
                        "end": 349,
                        "text": "Quirk et al., 2015;",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 350,
                        "end": 369,
                        "text": "Liang et al., 2016;",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 370,
                        "end": 395,
                        "text": "Neelakantan et al., 2016)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 447,
                        "end": 467,
                        "text": "Graves et al., 2016)",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 9,
                        "end": 17,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "7"
            },
            {
                "text": "Providing textual explanations for classification decisions has begun to receive attention, as part of increased interest in creating models whose decisions can be interpreted. Lei et al. (2016) , jointly modeled both a classification decision, and the selection of the most relevant subsection of a document for making the classification decision. Hendricks et al. (2016) generate textual explanations for visual classification problems, but in contrast to our model, they first generate an answer, and then, conditional on the answer, generate an explanation. This effectively creates a post-hoc justification for a classification decision rather than a program for deducing an answer. These papers, like ours, have jointly modeled rationales and answer predictions; however, we are the first to use rationales to guide program induction.",
                "cite_spans": [
                    {
                        "start": 177,
                        "end": 194,
                        "text": "Lei et al. (2016)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "7"
            },
            {
                "text": "In this work, we addressed the problem of generating rationales for math problems, where the task is to not only obtain the correct answer of the problem, but also generate a description of the method used to solve the problem. To this end, we collect 100,000 question and rationale pairs, and propose a model that can generate natural language and perform arithmetic operations in the same decoding process. Experiments show that our method outperforms existing neural models, in both the fluency of the rationales that are generated and the ability to solve the problem.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "8"
            },
            {
                "text": "The embeddings of a given argument ai,j and the return value vi are obtained with a lookup table embedding and two flags indicating whether it is a string and whether it is a float. Furthermore, if the the value is a float we also add its numeric value as a feature.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Staged Back-propagationAs it is shown inFigure 2, math rationales with more than 200 tokens are not uncommon, and with additional intermediate instructions, the size z can easily exceed 400. This poses a practical challenge for training the model. For both the attention and copy mechanisms,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "This modeling strategy is sometimes known as late fusion, as the attention vector is not used for state propagation, it is incorporated \"later\".",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Semantic parsing as machine translation",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Andreas",
                        "suffix": ""
                    },
                    {
                        "first": "Andreas",
                        "middle": [],
                        "last": "Vlachos",
                        "suffix": ""
                    },
                    {
                        "first": "Stephen",
                        "middle": [],
                        "last": "Clark",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jacob Andreas, Andreas Vlachos, and Stephen Clark. 2013. Semantic parsing as machine translation. In Proc. of ACL.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "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 Ben- gio. 2014. Neural machine translation by jointly learning to align and translate. arXiv 1409.0473.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Semantic parsing on freebase from question-answer pairs",
                "authors": [
                    {
                        "first": "Jonathan",
                        "middle": [],
                        "last": "Berant",
                        "suffix": ""
                    },
                    {
                        "first": "Andrew",
                        "middle": [],
                        "last": "Chou",
                        "suffix": ""
                    },
                    {
                        "first": "Roy",
                        "middle": [],
                        "last": "Frostig",
                        "suffix": ""
                    },
                    {
                        "first": "Percy",
                        "middle": [],
                        "last": "Liang",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. 2013. Semantic parsing on freebase from question-answer pairs. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Hybrid computing using a neural network with dynamic external memory",
                "authors": [
                    {
                        "first": "Alex",
                        "middle": [],
                        "last": "Graves",
                        "suffix": ""
                    },
                    {
                        "first": "Greg",
                        "middle": [],
                        "last": "Wayne",
                        "suffix": ""
                    },
                    {
                        "first": "Malcolm",
                        "middle": [],
                        "last": "Reynolds",
                        "suffix": ""
                    },
                    {
                        "first": "Tim",
                        "middle": [],
                        "last": "Harley",
                        "suffix": ""
                    },
                    {
                        "first": "Ivo",
                        "middle": [],
                        "last": "Danihelka",
                        "suffix": ""
                    },
                    {
                        "first": "Agnieszka",
                        "middle": [],
                        "last": "Grabska-Barwiska",
                        "suffix": ""
                    },
                    {
                        "first": "Sergio",
                        "middle": [
                            "Gmez"
                        ],
                        "last": "Colmenarejo",
                        "suffix": ""
                    },
                    {
                        "first": "Edward",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    },
                    {
                        "first": "Tiago",
                        "middle": [],
                        "last": "Ramalho",
                        "suffix": ""
                    },
                    {
                        "first": "John",
                        "middle": [],
                        "last": "Agapiou",
                        "suffix": ""
                    },
                    {
                        "first": "Adri",
                        "middle": [
                            "Puigdomnech"
                        ],
                        "last": "Badia",
                        "suffix": ""
                    },
                    {
                        "first": "Karl",
                        "middle": [
                            "Moritz"
                        ],
                        "last": "Hermann",
                        "suffix": ""
                    },
                    {
                        "first": "Yori",
                        "middle": [],
                        "last": "Zwols",
                        "suffix": ""
                    },
                    {
                        "first": "Georg",
                        "middle": [],
                        "last": "Ostrovski",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Koray Kavukcuoglu, and Demis Hassabis",
                "volume": "538",
                "issue": "",
                "pages": "471--476",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alex Graves, Greg Wayne, Malcolm Reynolds, Tim Harley, Ivo Danihelka, Agnieszka Grabska- Barwiska, Sergio Gmez Colmenarejo, Edward Grefenstette, Tiago Ramalho, John Agapiou, Adri Puigdomnech Badia, Karl Moritz Hermann, Yori Zwols, Georg Ostrovski, Adam Cain, Helen King, Christopher Summerfield, Phil Blunsom, Koray Kavukcuoglu, and Demis Hassabis. 2016. Hybrid computing using a neural network with dynamic external memory. Nature 538(7626):471- 476.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Rationalization: A neural machine translation approach to generating natural language explanations",
                "authors": [
                    {
                        "first": "Brent",
                        "middle": [],
                        "last": "Harrison",
                        "suffix": ""
                    },
                    {
                        "first": "Upol",
                        "middle": [],
                        "last": "Ehsan",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [
                            "O"
                        ],
                        "last": "Riedl",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Brent Harrison, Upol Ehsan, and Mark O. Riedl. 2017. Rationalization: A neural machine translation ap- proach to generating natural language explanations. CoRR abs/1702.07826.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Generating visual explanations",
                "authors": [
                    {
                        "first": "Lisa",
                        "middle": [
                            "Anne"
                        ],
                        "last": "Hendricks",
                        "suffix": ""
                    },
                    {
                        "first": "Zeynep",
                        "middle": [],
                        "last": "Akata",
                        "suffix": ""
                    },
                    {
                        "first": "Marcus",
                        "middle": [],
                        "last": "Rohrbach",
                        "suffix": ""
                    },
                    {
                        "first": "Jeff",
                        "middle": [],
                        "last": "Donahue",
                        "suffix": ""
                    },
                    {
                        "first": "Bernt",
                        "middle": [],
                        "last": "Schiele",
                        "suffix": ""
                    },
                    {
                        "first": "Trevor",
                        "middle": [],
                        "last": "Darrell",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. ECCV",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lisa Anne Hendricks, Zeynep Akata, Marcus Rohrbach, Jeff Donahue, Bernt Schiele, and Trevor Darrell. 2016. Generating visual explanations. In Proc. ECCV.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Learning to solve arithmetic word problems with verb categorization",
                "authors": [
                    {
                        "first": "Mohammad Javad",
                        "middle": [],
                        "last": "Hosseini",
                        "suffix": ""
                    },
                    {
                        "first": "Hannaneh",
                        "middle": [],
                        "last": "Hajishirzi",
                        "suffix": ""
                    },
                    {
                        "first": "Oren",
                        "middle": [],
                        "last": "Etzioni",
                        "suffix": ""
                    },
                    {
                        "first": "Nate",
                        "middle": [],
                        "last": "Kushman",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. 2014. Learning to solve arithmetic word problems with verb catego- rization. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Semantic parsing with bayesian tree transducers",
                "authors": [
                    {
                        "first": "Keeley",
                        "middle": [],
                        "last": "Bevan",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Jones",
                        "suffix": ""
                    },
                    {
                        "first": "Sharon",
                        "middle": [],
                        "last": "Johnson",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Goldwater",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Bevan Keeley Jones, Mark Johnson, and Sharon Gold- water. 2012. Semantic parsing with bayesian tree transducers. In Proc. of ACL.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Recurrent continuous translation models",
                "authors": [
                    {
                        "first": "Nal",
                        "middle": [],
                        "last": "Kalchbrenner",
                        "suffix": ""
                    },
                    {
                        "first": "Phil",
                        "middle": [],
                        "last": "Blunsom",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nal Kalchbrenner and Phil Blunsom. 2013. Recurrent continuous translation models. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Learning to automatically solve algebra word problems",
                "authors": [
                    {
                        "first": "Nate",
                        "middle": [],
                        "last": "Kushman",
                        "suffix": ""
                    },
                    {
                        "first": "Yoav",
                        "middle": [],
                        "last": "Artzi",
                        "suffix": ""
                    },
                    {
                        "first": "Luke",
                        "middle": [],
                        "last": "Zettlemoyer",
                        "suffix": ""
                    },
                    {
                        "first": "Regina",
                        "middle": [],
                        "last": "Barzilay",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nate Kushman, Yoav Artzi, Luke Zettlemoyer, and Regina Barzilay. 2014. Learning to automatically solve algebra word problems. In Proc. of ACL.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Rationalizing neural predictions",
                "authors": [
                    {
                        "first": "Tao",
                        "middle": [],
                        "last": "Lei",
                        "suffix": ""
                    },
                    {
                        "first": "Regina",
                        "middle": [],
                        "last": "Barzilay",
                        "suffix": ""
                    },
                    {
                        "first": "Tommi",
                        "middle": [],
                        "last": "Jaakkola",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2016. Rationalizing neural predictions. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Neural symbolic machines: Learning semantic parsers on freebase with weak supervision",
                "authors": [
                    {
                        "first": "Chen",
                        "middle": [],
                        "last": "Liang",
                        "suffix": ""
                    },
                    {
                        "first": "Jonathan",
                        "middle": [],
                        "last": "Berant",
                        "suffix": ""
                    },
                    {
                        "first": "Quoc",
                        "middle": [],
                        "last": "Le",
                        "suffix": ""
                    },
                    {
                        "first": "Kenneth",
                        "middle": [
                            "D"
                        ],
                        "last": "Forbus",
                        "suffix": ""
                    },
                    {
                        "first": "Ni",
                        "middle": [],
                        "last": "Lao",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chen Liang, Jonathan Berant, Quoc Le, Kenneth D. Forbus, and Ni Lao. 2016. Neural symbolic ma- chines: Learning semantic parsers on freebase with weak supervision. arXiv 1611.00020.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Latent predictor networks for code generation",
                "authors": [
                    {
                        "first": "Wang",
                        "middle": [],
                        "last": "Ling",
                        "suffix": ""
                    },
                    {
                        "first": "Edward",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    },
                    {
                        "first": "Karl",
                        "middle": [
                            "Moritz"
                        ],
                        "last": "Hermann",
                        "suffix": ""
                    },
                    {
                        "first": "Tom\u00e1s",
                        "middle": [],
                        "last": "Kocisk\u00fd",
                        "suffix": ""
                    },
                    {
                        "first": "Andrew",
                        "middle": [],
                        "last": "Senior",
                        "suffix": ""
                    },
                    {
                        "first": "Fumin",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Phil",
                        "middle": [],
                        "last": "Blunsom",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wang Ling, Edward Grefenstette, Karl Moritz Her- mann, Tom\u00e1s Kocisk\u00fd, Andrew Senior, Fumin Wang, and Phil Blunsom. 2016. Latent predictor networks for code generation. In Proc. of ACL.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Pointer sentinel mixture models",
                "authors": [
                    {
                        "first": "Stephen",
                        "middle": [],
                        "last": "Merity",
                        "suffix": ""
                    },
                    {
                        "first": "Caiming",
                        "middle": [],
                        "last": "Xiong",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Bradbury",
                        "suffix": ""
                    },
                    {
                        "first": "Richard",
                        "middle": [],
                        "last": "Socher",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. 2016. Pointer sentinel mixture models. arXiv 1609.07843.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Neural programmer: Inducing latent programs with gradient descent",
                "authors": [
                    {
                        "first": "Arvind",
                        "middle": [],
                        "last": "Neelakantan",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Quoc",
                        "suffix": ""
                    },
                    {
                        "first": "Ilya",
                        "middle": [],
                        "last": "Le",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sutskever",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. ICLR",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Arvind Neelakantan, Quoc V. Le, and Ilya Sutskever. 2016. Neural programmer: Inducing latent pro- grams with gradient descent. In Proc. ICLR.",
                "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": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. Bleu: A method for automatic eval- uation of machine translation. In Proc. of ACL.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Language to code: Learning semantic parsers for if-this-then-that recipes",
                "authors": [
                    {
                        "first": "Chris",
                        "middle": [],
                        "last": "Quirk",
                        "suffix": ""
                    },
                    {
                        "first": "Raymond",
                        "middle": [],
                        "last": "Mooney",
                        "suffix": ""
                    },
                    {
                        "first": "Michel",
                        "middle": [],
                        "last": "Galley",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proc. of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chris Quirk, Raymond Mooney, and Michel Galley. 2015. Language to code: Learning semantic parsers for if-this-then-that recipes. In Proc. of ACL.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Neural programmer-interpreters",
                "authors": [
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Scott",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. of ICLR",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Scott E. Reed and Nando de Freitas. 2016. Neural programmer-interpreters. In Proc. of ICLR.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Solving general arithmetic word problems",
                "authors": [
                    {
                        "first": "Subhro",
                        "middle": [],
                        "last": "Roy",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Roth",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Subhro Roy and Dan Roth. 2015. Solving general arithmetic word problems. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Equation parsing: Mapping sentences to grounded equations",
                "authors": [
                    {
                        "first": "Subhro",
                        "middle": [],
                        "last": "Roy",
                        "suffix": ""
                    },
                    {
                        "first": "Shyam",
                        "middle": [],
                        "last": "Upadhyay",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Roth",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. of EMNLP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Subhro Roy, Shyam Upadhyay, and Dan Roth. 2016. Equation parsing: Mapping sentences to grounded equations. In Proc. of EMNLP.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Sequence to sequence learning with neural networks. arXiv 1409",
                "authors": [
                    {
                        "first": "Ilya",
                        "middle": [],
                        "last": "Sutskever",
                        "suffix": ""
                    },
                    {
                        "first": "Oriol",
                        "middle": [],
                        "last": "Vinyals",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Quoc",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Le",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ilya Sutskever, Oriol Vinyals, and Quoc V. Le. 2014. Sequence to sequence learning with neural net- works. arXiv 1409.3215.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Pointer networks",
                "authors": [
                    {
                        "first": "Oriol",
                        "middle": [],
                        "last": "Vinyals",
                        "suffix": ""
                    },
                    {
                        "first": "Meire",
                        "middle": [],
                        "last": "Fortunato",
                        "suffix": ""
                    },
                    {
                        "first": "Navdeep",
                        "middle": [],
                        "last": "Jaitly",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proc. of NIPS",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. 2015. Pointer networks. In Proc. of NIPS.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "text": ".H.S = R.H.S which is the correct answer. answer:B Correct Option: B Examples of solved math problems.",
                "type_str": "figure",
                "uris": null
            },
            "FIGREF1": {
                "num": null,
                "text": "For instance, in Problem 1, the equation (27x + Distribution of examples per length.",
                "type_str": "figure",
                "uris": null
            },
            "FIGREF2": {
                "num": null,
                "text": "using a softmax over the set of available operations O.",
                "type_str": "figure",
                "uris": null
            },
            "TABREF1": {
                "num": null,
                "text": "Descriptive statistics of our dataset.",
                "content": "<table/>",
                "type_str": "table",
                "html": null
            },
            "TABREF2": {
                "num": null,
                "text": "Example of a program z that would generate the output y. In v, italics indicates string types; bold indicates float types. Refer to \u00a73.3 for description of variable names.",
                "content": "<table/>",
                "type_str": "table",
                "html": null
            },
            "TABREF4": {
                "num": null,
                "text": "Results over the test set measured in Perplexity, BLEU and Accuracy.",
                "content": "<table/>",
                "type_str": "table",
                "html": null
            }
        }
    }
}