File size: 62,119 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
{
    "paper_id": "I05-1046",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T07:24:41.732296Z"
    },
    "title": "Web-Based Unsupervised Learning for Query Formulation in Question Answering",
    "authors": [
        {
            "first": "Yi-Chia",
            "middle": [],
            "last": "Wang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Ta Hsueh Rd",
                "location": {
                    "postBox": "300",
                    "postCode": "1001",
                    "settlement": "Hsinchu",
                    "region": "R.O.C",
                    "country": "Taiwan"
                }
            },
            "email": ""
        },
        {
            "first": "Jian-Cheng",
            "middle": [],
            "last": "Wu",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Tsing Hua University",
                "location": {
                    "addrLine": "Section 2 Kuang Fu Road",
                    "postCode": "101, 300",
                    "settlement": "Hsinchu",
                    "region": "R.O.C",
                    "country": "Taiwan"
                }
            },
            "email": ""
        },
        {
            "first": "Tyne",
            "middle": [],
            "last": "Liang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Ta Hsueh Rd",
                "location": {
                    "postBox": "300",
                    "postCode": "1001",
                    "settlement": "Hsinchu",
                    "region": "R.O.C",
                    "country": "Taiwan"
                }
            },
            "email": "tliang@cis.nctu.edu.tw"
        },
        {
            "first": "Jason",
            "middle": [
                "S"
            ],
            "last": "Chang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Tsing Hua University",
                "location": {
                    "addrLine": "Section 2 Kuang Fu Road",
                    "postCode": "101, 300",
                    "settlement": "Hsinchu",
                    "region": "R.O.C",
                    "country": "Taiwan"
                }
            },
            "email": "jschang@cs.nthu.edu.tw"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Converting questions to effective queries is crucial to open-domain question answering systems. In this paper, we present a web-based unsupervised learning approach for transforming a given natural-language question to an effective query. The method involves querying a search engine for Web passages that contain the answer to the question, extracting patterns that characterize fine-grained classification for answers, and linking these patterns with n-grams in answer passages. Independent evaluation on a set of questions shows that the proposed approach outperforms a naive keywordbased approach in terms of mean reciprocal rank and human effort.",
    "pdf_parse": {
        "paper_id": "I05-1046",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Converting questions to effective queries is crucial to open-domain question answering systems. In this paper, we present a web-based unsupervised learning approach for transforming a given natural-language question to an effective query. The method involves querying a search engine for Web passages that contain the answer to the question, extracting patterns that characterize fine-grained classification for answers, and linking these patterns with n-grams in answer passages. Independent evaluation on a set of questions shows that the proposed approach outperforms a naive keywordbased approach in terms of mean reciprocal rank and human effort.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "An automated question answering (QA) system receives a user's natural-language question and returns exact answers by analyzing the question and consulting a large text collection [1, 2] . As Moldovan et al. [3] pointed out, over 60% of the QA errors can be attributed to ineffective question processing, including query formulation and query expansion.",
                "cite_spans": [
                    {
                        "start": 179,
                        "end": 182,
                        "text": "[1,",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 183,
                        "end": 185,
                        "text": "2]",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 207,
                        "end": 210,
                        "text": "[3]",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "A naive solution to query formulation is using the keywords in an input question as the query to a search engine. However, it is possible that the keywords may not appear in those answer passages which contain answers to the given question. For example, submitting the keywords in \"Who invented washing machine?\" to a search engine like Google may not lead to retrieval of answer passages like \"The inventor of the automatic washer was John Chamberlain.\" In fact, by expanding the keyword set (\"invented\", \"washing\", \"machine\") with \"inventor of,\" the query to a search engine is effective in retrieving such answer passages as the top-ranking pages. Hence, if we can learn how to associate a set of questions (e.g. (\"who invented \u2026?\") with effective keywords or phrases (e.g. \"inventor of\") which are likely to appear in answer passages, the search engine will have a better chance of retrieving pages containing the answer.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we present a novel Web-based unsupervised learning approach to handling question analysis for QA systems. In our approach, training-data questions are first analyzed and classified into a set of fine-grained categories of question patterns. Then, the relationships between the question patterns and n-grams in answer passages are discovered by employing a word alignment technique. Finally, the best query transforms are derived by ranking the n-grams which are associated with a specific question pattern. At runtime, the keywords in a given question are extracted and the question is categorized. Then the keywords are expanded according the category of the question. The expanded query is the submitted to a search engine in order to bias the search engine to return passages that are more likely to contain answers to the question. Experimental results indicate the expanded query indeed outperforms the approach of directly using the keywords in the question.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Recent work in Question Answering has attempted to convert the original input question into a query that is more likely to retrieve the answers. Hovy et al. [2] utilized WordNet hypernyms and synonyms to expand queries to increase recall. Hildebrandt et al. [4] looked up in a pre-compiled knowledge base and a dictionary to expand a definition question. However, blindly expanding a word using its synonyms or dictionary gloss may cause undesirable effects. Furthermore, it is difficult to determine which of many related word senses should be considered when expanding the query.",
                "cite_spans": [
                    {
                        "start": 157,
                        "end": 160,
                        "text": "[2]",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 258,
                        "end": 261,
                        "text": "[4]",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Radev et al. [5] proposed a probabilistic algorithm called QASM that learns the best query expansion from a natural language question. The query expansion takes the form of a series of operators, including INSERT, DELETE, REPLACE, etc., to paraphrase a factual question into the best search engine query by applying Expectation Maximization algorithm. On the other hand, Hermjakob et al. [6] described an experiment to observe and learn from human subjects who were given a question and asked to write queries which are most effective in retrieving the answer to the question. First, several randomly selected questions are given to users to \"manually\" generate effective queries that can bias Web search engines to return answers. The questions, queries, and search results are then examined to derive seven query reformulation techniques that can be used to produce queries similar to the ones issued by human subjects.",
                "cite_spans": [
                    {
                        "start": 13,
                        "end": 16,
                        "text": "[5]",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 388,
                        "end": 391,
                        "text": "[6]",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "In a study closely related to our work, Agichtein et al. [7] presented Tritus system that automatically learns transforms of wh-phrases (e.g. expanding \"what is\" to \"refers to\") by using FAQ data. The wh-phrases are restricted to sequences of function word beginning with an interrogative, (i.e. who, what, when, where, why, and how). These wh-phrases tend to coarsely classify questions into a few types.",
                "cite_spans": [
                    {
                        "start": 57,
                        "end": 60,
                        "text": "[7]",
                        "ref_id": "BIBREF6"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Tritus uses heuristic rules and thresholds of term frequencies to learn transforms.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "In contrast to previous work, we rely on a mathematical model trained on a set of questions and answers to learn how to transform the question into an effective query. Transformations are learned based on a more fine-grained question classification involving the interrogative and one or more content words.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "The method is aimed at automatically learning of the best transforms that turn a given natural language question into an effective query by using the Web as corpus. To that end, we first automatically obtain a collection of answer passages (APs) as the training corpus from the Web by using a set of (Q, A) pairs. Then we identify the question pattern for each Q by using statistical and linguistic information. Here, a question pattern Q p is defined as a question word plus one or two keywords that are related to the question word. Q p represents the question intention and it can be treated as a preference indicative for fine-grained type of named entities. Finally, we decide the transforms Ts for each Q p by choosing those phrases in the APs that are statistically associated with Q p and adjacent to the answer A. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transforming Question to Query",
                "sec_num": "3"
            },
            {
                "text": "For training purpose, a large amount of question/answer passage pairs are mined from the Web by using a set of question/answer pairs as seeds. More formally, we attempt to retrieve a set of (Q, AP) pairs on the Web for training purpose, where Q stands for a natural language question, and AP is a passage containing at least one keyword in Q and A (the answer to Q). The seed data (Q, A) pairs can be acquired from many sources, including trivia game Websites, TREC QA Track benchmarks, and files of Frequently Asked Questions (FAQ). The output of this training-data gathering process is a large collection of (Q, AP) pairs. We describe the procedure in details as follows: 1. For each (Q, A) pair, the keywords k 1 , k 2 ,\u2026, k n are extracted from Q by removing stopwords. 2. Submit (k 1 , k 2 ,\u2026, k n , A) as a query to a search engine SE. 3. Download the top n summaries returned by SE. 4. Separate sentences in the summaries, and remove HTML tags, URL, special character references (e.g., \"<\"). 5. Retain only those sentences which contain A and some k i .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Search the Web for Relevant Answer Passages",
                "sec_num": "3.1"
            },
            {
                "text": "Consider the example of gathering answer passages from the Web for the (Q, A) pair where Q = \"What is the capital of Pakistan?\" and A = \"Islamabad.\" See Table 1 for the query submitted to a search engine and potential answer passages returned.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 153,
                        "end": 160,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Search the Web for Relevant Answer Passages",
                "sec_num": "3.1"
            },
            {
                "text": "This subsection describes the presented identification of the so-called \"question pattern\" which is critical in categorizing a given question and transforming the question into a query.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Analysis",
                "sec_num": "3.2"
            },
            {
                "text": "Formally, a \"question pattern\" for any question is defined as following form:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Analysis",
                "sec_num": "3.2"
            },
            {
                "text": "question-word head-word+",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Analysis",
                "sec_num": "3.2"
            },
            {
                "text": "where \"question-word\" is one of the interrogatives (Who/What/Where/When/How) and the \"head-word\" represents the headwords in the subsequent chunks that tend to reflect the intended answer more precisely. If the first headword is a light verb, an additional headword is needed. For instance, \"who had hit\" is a reasonable question pattern for \"Who had a number one hit in 1984 with 'Hello'?\", while \"who had\" seems to be too coarse.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Analysis",
                "sec_num": "3.2"
            },
            {
                "text": "In order to determine the appropriate question pattern for each question, we examined and analyzed a set of questions which are part-of-speech (POS) tagged and phrase-chunked. With the help of a set of simple heuristic rules based on POS and chunk information, fine-grained classification of questions can be carried out effectively.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Analysis",
                "sec_num": "3.2"
            },
            {
                "text": "After analyzing recurring patterns and regularity in quizzes on the Web, we designed a simple procedure to recognize question patterns. The procedure is based on a small set of prioritized rules.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Question Pattern Extraction",
                "sec_num": null
            },
            {
                "text": "The question word which is one of the wh-words (\"who,\" \"what,\" \"when,\" \"where,\" \"how,\" or \"why\") tagged as determiner or adverbial question word. According to the result of POS tagging and phrase chunking, we further decide the main verb and the voice of the question. Then, we apply the following expanded rules to extract words to form question patterns: Rule 1: Question word in a chunk of length more than one (see Example (1) in Table 2 ). Qp = question word + headword in the same chunk By exploiting linguistic information of POS and chunks, we can easily form the question pattern. These heuristic rules are intuitive and easy to understand. Moreover, the fact that these patterns which tend to recur imply that they are general and it is easy to gather training data accordingly. These question patterns also indicate a preference for the answer to be classified with a fine-grained type of proper nouns. In the next section, we describe how we exploit these patterns to learn the best questionto-query transforms. ",
                "cite_spans": [
                    {
                        "start": 427,
                        "end": 430,
                        "text": "(1)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 434,
                        "end": 441,
                        "text": "Table 2",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Question Pattern Extraction",
                "sec_num": null
            },
            {
                "text": "This section describes the procedure for learning transforms Ts which convert the question pattern Q p into bigrams in relevant APs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Learning Best Transforms",
                "sec_num": "3.3"
            },
            {
                "text": "We use word alignment techniques developed for statistical machine translation to find out the association between question patterns in Q and bigrams in AP. The reason why we use bigrams in APs instead of unigrams is that bigrams tend to have more unique meaning than single words and are more effective in retrieving relevant passages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "We use Competitive Linking Algorithm [8] to align a set of (Q, AP) pairs. The method involves preprocessing steps for each (Q, AP) pair so as to filter useless information:",
                "cite_spans": [
                    {
                        "start": 37,
                        "end": 40,
                        "text": "[8]",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "1. Perform part-of-speech tagging on Q and AP. 2. Replace all instances of A with the tag <ANS> in APs to indicate the location of the answers. 3. Identify the question pattern, Q p and keywords which are not a named entity. We denote the question pattern and keywords as q 1 , q 2 , ..., q n . 4. Convert AP into bigrams and eliminate bigrams with low term frequency (tf) or high document frequency (df). Bigrams composed of two function words are also removed, resulting in bigrams a 1 , a 2 , ..., a m .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "We then align q's and a's via Competitive Linking Algorithm (CLA) procedure as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "Input: A collection C of (Q; A) pairs, where (Q; A) = (q 1 = Q p , q 2 , q 3 , ..., q n ; a 1 , a 2 , ..., a m ) Output: Best alignment counterpart a's for all q's in C 1. For each pair of (Q; A) in C and for all q i and a j in each pair of C, calculate LLR(q i , a j ), logarithmic likelihood ratio (LLR) between q i and a j , which reflects their statistical association. 2. Discard (q, a) pairs with a LLR value lower than a threshold.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "3. For each pair of (Q; A) in C and for all q i and a j therein, carry out Steps 4-7: 4. Sort list of (q i , a j ) in each pair of (Q ; A) by decreasing LLR value. 5. Go down the list and select a pair if it does not conflict with previous selection. 6. Stop when running out of pairs in the list. 7. Produce the list of aligned pairs for all Qs and APs. 8. Tally the counts of aligning (q, a). 9. Select top k bigrams, t 1 , t 2 , ..., t k , for every question pattern or keyword q.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "The LLR statistics is generally effective in distinguishing related terms from unrelated ones. However, if two terms occur frequently in questions, their alignment counterparts will also occur frequently, leading to erroneous alignment due to indirect association. CLA is designed to tackle the problem caused by indirect association. Therefore, if we only make use of the alignment counterpart of the question pattern, we can keep the question keywords in Q so as to reduce the errors caused by indirect association. For instance, the question \"How old was Bruce Lee when he died?\" Our goal is to learn the best transforms for the question pattern \"how old.\" In other words, we want to find out what terms are associated with \"how old\" in the answer passages. However, if we consider the alignment counterparts of \"how old\" without considering those keyword like \"died,\" we run the risk of getting \"died in\" or \"is dead\" rather than \"years old\" and \"age of.\" If we have sufficient data for a specific question pattern like \"how long,\" we will have more chances to obtain alignment counterparts that are effective terms for query expansion.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Alignment Across Q and AP",
                "sec_num": null
            },
            {
                "text": "In addition to the association strength implied with alignment counts and cooccurrence, the distance of the bigrams to the answer should also be considered. We observe that terms in the answer passages close to the answers intuitively tend to be useful in retrieving answers. Thus, we calculate the bigrams appearing in a window of three words appearing on both sides of the answers to provide additional constraints for query expansion.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Distance Constraint and Proximity Ranks",
                "sec_num": null
            },
            {
                "text": "The selection of the best bigrams as the transforms for a specific question pattern is based on a combined rank of alignment count and proximity count. It takes the average of these two counts to re-rank bigrams. The average rank of a bigram b is The n top-ranking bigrams for a specific type of question will be chosen to transform the question pattern into query terms. For the question pattern \"how old,\" the candidate bigrams with alignment ranks, co-occurring ranks, and average ranks are shown in Table 3 . ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 503,
                        "end": 510,
                        "text": "Table 3",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Combing Alignment and Proximity Ranks",
                "sec_num": null
            },
            {
                "text": "At runtime, a given question Q submitted by a user is converted into one or more keywords and a question pattern, which is subsequently expanded in to a sequence of query terms based on the transforms obtained at training. We follow the common practice of keyword selection in formulating Q into a query:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "\u2022 Function words are identified and discarded.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "\u2022 Proper nouns that are capitalized or quoted are treated as a single search term with quotes.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "Additionally, we expand the question patterns based on alignment and proximity considerations:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "\u2022 The question pattern Q p is identified according to the rules (in Section 3.2) and is expanded to be a disjunction (sequence of ORs) of Q p 's headword and n topranking bigrams (in section 3.3)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "\u2022 The query will be a conjunction (sequence of ANDs) of expanded Q p , proper names, and remaining keywords. Except for the expanded Q p , all other proper names and keywords will be in the original order in the given question for the best results. Expanded query Boolean query: ( \"old\" OR \"age of\" OR \"years old\" ) AND \"Bruce Lee\" AND \"died\" Equivalent Google query: (old || \"age of\" || \"years old\") \"Bruce Lee\" died For example, formulating a query for the question \"How old was Bruce Lee when he died?\" will result in a question pattern \"how old.\" Because there is a proper noun \"Bruce Lee\" in the question and a remaining keyword \"died,\" the query becomes \"( 'old' OR 'age of' OR 'years old' ) AND 'Bruce Lee' AND 'died.'\" Table 4 lists the query formulating for the example question.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 727,
                        "end": 734,
                        "text": "Table 4",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Runtime Transformation of Questions",
                "sec_num": "3.4"
            },
            {
                "text": "The proposed method is implemented by using the Web search engine, Google, as the underlying information retrieval system. The experimental results are also justified with assessing the effectiveness of question classification and query expansion.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments and Evaluation",
                "sec_num": "4"
            },
            {
                "text": "We used a POS tagger and chunker to perform shallow parsing of the questions and answer passages. The tagger was developed using the Brown corpus and WordNet. The chunker is built from the shared CoNLL-2000 data provided by CoNLL-2000. The shared task CoNLL-2000 provides a set of training and test data for chunks. The chunker we used produces chunks with an average precision rate of about 94%.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments and Evaluation",
                "sec_num": "4"
            },
            {
                "text": "The 200 questions from TREC-8 QA Track provide an independent evaluation of how well the proposed method works for question pattern extraction works. We will also give an error analysis. Two human judges both majoring in Foreign Languages were asked to assess the results of question pattern extraction and give a label to each extracted question pattern. A pattern will be judged as \"good\" if it clearly expresses the answer preference of the question; otherwise, it is tagged as \"bad.\" The precision rate of extraction for these 200 questions is shown in Table 5 . The second column indicates the precision rate when both of two judges agree that an extracted question pattern is \"good.\" In addition, the third column indicates the rate of those question patterns that are found to be \"good\" by either judge. The results imply that the proposed pattern extraction rules are general, since they are effective even for questions independent of the training and development data. Table 6 shows evaluation results for \"two 'good' labels\" of the first five questions.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 557,
                        "end": 564,
                        "text": "Table 5",
                        "ref_id": null
                    },
                    {
                        "start": 979,
                        "end": 986,
                        "text": "Table 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Question Patterns",
                "sec_num": "4.1"
            },
            {
                "text": "We summarize the reasons behind these bad patterns: \u2022 Incorrect part-of-speech tagging and chunking \u2022 Imperative questions such as \"Name the first private citizen to fly in space.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Question Patterns",
                "sec_num": "4.1"
            },
            {
                "text": "\u2022 Question patterns that are not specific enough For instance, the system produces \"what name\" for \"What is the name of the chronic neurological autoimmune disease which \u2026 ?\", while the judges suggested that \"what disease.\". Indeed, some of the patterns extracted can be modified to meet the goal of being more fine-grained and indicative of a preference to a specific type of proper nouns or terminology.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Question Patterns",
                "sec_num": "4.1"
            },
            {
                "text": "We implemented a prototype of the proposed method called Atlas (Automatic Transform Learning by Aligning Sentences of question and answer). To develop the system of Atlas, we gathered seed training data of questions and answers from a trivia game website, called QuizZone 1 . We collected the questions posted in June, 2004 on QuizZone and obtained 3,851 distinct question-answer pairs. We set aside the first 45 questions for testing and used the rest for training. For each question, we form a query with question keywords and the answer and submitted the query to Google to retrieve top 100 summaries as the answer passages. In all, we collected 95,926 answer passages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "At training time, we extracted a total of 338 distinct question patterns from 3,806 questions. We aligned these patterns and keywords with bigrams in the 95,926 answer passages, identified the locations of the answers, and obtained the bigrams appearing within a distance of 3 of the answers. At runtime, we use the top-ranking bigram to expand each question pattern. If no such bigrams are found, we use only the keyword in the question patterns. The expanded terms for question pattern are placed at the beginning of the query.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "We submitted forty-five keyword queries and the same number of expanded queries generated by Atlas for the test questions to Google and obtained ten returned summaries for evaluation. For the evaluation, we use three indicators to measure the performance. The first indicator is the mean reciprocal rank (MRR) of the first relevant document (or summary) returned. If the r-th document (summary) returned is the one with the answer, then the reciprocal rank of the document (summary) is 1/r.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "The mean reciprocal rank is the average reciprocal rank of all test questions. The second indicator of effective query is the recall at R document retrieved (Recall at R). The last indicator measures the human effort (HE) in finding the answer. HE is defined as the least number of passages needed to be viewed for covering all the answers to be returned from the system.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "The average length of these test questions is short. We believe the proposed question expansion scheme helps those short sentences, which tend to be less effective in retrieving answers. We evaluated the expanded queries against the same measures for summaries returned by simple keyword queries. Both batches of returned summaries for the forty-five questions were verified by two human judges.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "As shown in Table 7 , the MRR produced by keyword-based scheme is slightly lower than the one yielded by the presented query expansion scheme. Nevertheless, such improvement is encouraging by indicating the effectiveness of the proposed method. Table 8 lists the comparisons in more details. It is found that our method is effective in bringing the answers to the top 1 and top 2 summaries as indicated by the high Recall of 0.8 at R = 2. In addition, Table 8 also shows that less user's efforts are needed by using our approach. That is, for each question, the average of summaries required to be viewed by human beings goes down from 2.7 to 2.3.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 12,
                        "end": 19,
                        "text": "Table 7",
                        "ref_id": null
                    },
                    {
                        "start": 245,
                        "end": 252,
                        "text": "Table 8",
                        "ref_id": null
                    },
                    {
                        "start": 452,
                        "end": 459,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "In the end, we found that those bigrams containing a content word and a function word turn out to be very effective. For instance, our method tends to favor transforms Table 7 . Evaluation results of MRR Performances MRR GO (Direct keyword query for Google) 0.64 AT+GO (Atlas expanded query for Google) 0.69 Table 8 . Evaluation Result of Recall at R and Human Effort such as \"who invented\" to bigrams such as \"invented by,\" \"invent the,\" and \"inventor of.\" This contrasts to conventional wisdom of using a stoplist of mostly function words and excluding them from consideration in a query. Our experiment also shows a function word as part of a phrasal term seems to be very effective, for it indicate an implied relation with the answer.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 168,
                        "end": 175,
                        "text": "Table 7",
                        "ref_id": null
                    },
                    {
                        "start": 308,
                        "end": 315,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Query Expansion",
                "sec_num": "4.2"
            },
            {
                "text": "In this paper, we introduce a method for learning query transformations that improves the ability to retrieve passages with answers using the Web as corpus. The method involves question classification and query transformations using a learning-based approach. We also describe the experiment with over 3,000 questions indicates that satisfactory results were achieved. The experimental results show that the proposed method provides effective query expansion that potentially can lead to performance improvement for a question answering system. A number of future directions present themselves. First, the patterns learned from answer passages acquired on the Web can be refined and clustered to derive a hierarchical classification of questions for more effective question classification. Second, different question patterns, like \"who wrote\" and \"which author\", should be treated as the same in order to cope with data sparseness and improve system performance. On the other hand, an interesting direction is the generating pattern transformations that contain the answer extraction patterns for different types of questions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "5"
            },
            {
                "text": "QuizZone (http://www.quiz-zone.co.uk)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "Recall",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Rank count",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "IBM's statistical question answering system",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Ittycheriah",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Franz",
                        "suffix": ""
                    },
                    {
                        "first": "W.-J",
                        "middle": [],
                        "last": "Zhu",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Rathaparkhi",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proceedings of the TREC-9 Question Answering Track",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ittycheriah, A., Franz, M., Zhu, W.-J., and Rathaparkhi, A. 2000. IBM's statistical question answering system. In Proceedings of the TREC-9 Question Answering Track, Gaithersburg, Maryland.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Question answering in Webclopedia",
                "authors": [
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Hovy",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Gerber",
                        "suffix": ""
                    },
                    {
                        "first": "U",
                        "middle": [],
                        "last": "Hermjakob",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Junk",
                        "suffix": ""
                    },
                    {
                        "first": "C.-Y",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proceedings of the TREC-9 Question Answering Track",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hovy, E., Gerber, L., Hermjakob, U., Junk, M., and Lin, C.-Y. 2000. Question answering in Webclopedia. In Proceedings of the TREC-9 Question Answering Track, Gaithersburg, Maryland.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Performance Issues and error Analysis in an Open-Domain Question Answering System",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Moldovan",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Pasca",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Harabagiu",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Surdeanu",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 40th Annual Meeting of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Moldovan D., Pasca M., Harabagiu S., & Surdeanu M. 2002. Performance Issues and error Analysis in an Open-Domain Question Answering System. In Proceedings of the 40th Annual Meeting of ACL, Philadelphia, Pennsylvania.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Answering definition questions with multiple knowledge sources",
                "authors": [
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Hildebrandt",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Katz",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the 2004 Human Language Technology Conference and the North American Chapter of the Association for Computational",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hildebrandt, W., Katz, B., & Lin, J. 2004. Answering definition questions with multiple knowledge sources. In Proceedings of the 2004 Human Language Technology Conference and the North American Chapter of the Association for Computational.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Mining the web for answers to natural language questions",
                "authors": [
                    {
                        "first": "D",
                        "middle": [
                            "R"
                        ],
                        "last": "Radev",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Qi",
                        "suffix": ""
                    },
                    {
                        "first": "Z",
                        "middle": [],
                        "last": "Zheng",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Blair-Goldensohn",
                        "suffix": ""
                    },
                    {
                        "first": "Z",
                        "middle": [
                            "Z W"
                        ],
                        "last": "Fan",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "M"
                        ],
                        "last": "Prager",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the International Conference on Knowledge Management (CIKM-2001)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Radev, D. R., Qi, H., Zheng, Z., Blair-Goldensohn, S., Fan, Z. Z. W., and Prager, J. M. 2001. Mining the web for answers to natural language questions. In Proceedings of the International Conference on Knowledge Management (CIKM-2001), Atlanta, Georgia.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Natural Language Based Reformulation Resource and Web Exploitation for Question Answering",
                "authors": [
                    {
                        "first": "U",
                        "middle": [],
                        "last": "Hermjakob",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Echihabi",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Marcu",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceeding of TREC-2002",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hermjakob, U., Echihabi, A., and Marcu, D. 2002. Natural Language Based Reformulation Resource and Web Exploitation for Question Answering. In Proceeding of TREC-2002, Gaithersburg, Maryland.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Learning to find answers to questions on the Web",
                "authors": [
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Agichtein",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Lawrence",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Gravano",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "ACM Transactions on Internet Technology (TOIT)",
                "volume": "4",
                "issue": "",
                "pages": "129--162",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Agichtein, E., Lawrence, S., and Gravano, L. Learning to find answers to questions on the Web. 2003. In ACM Transactions on Internet Technology (TOIT), 4(2):129-162.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "A Word-to-Word Model of Translational Equivalence",
                "authors": [
                    {
                        "first": "I",
                        "middle": [
                            "D"
                        ],
                        "last": "Melamed",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proceedings of the 35st Annual Meeting of ACL",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Melamed, I. D. 1997. A Word-to-Word Model of Translational Equivalence. In Proceedings of the 35st Annual Meeting of ACL, Madrid, Spain.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Using the Web as Corpus for Un-supervised Learning in Question Answering",
                "authors": [
                    {
                        "first": "Yi-Chia",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Jian-Cheng",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Tyne",
                        "middle": [],
                        "last": "Liang",
                        "suffix": ""
                    },
                    {
                        "first": "Jason",
                        "middle": [
                            "S"
                        ],
                        "last": "Chang",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of Rocling",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yi-Chia Wang, Jian-Cheng Wu, Tyne Liang, and Jason S. Chang. 2004. Using the Web as Corpus for Un-supervised Learning in Question Answering, Proceedings of Rocling 2004, Taiwan.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "type_str": "figure",
                "uris": null,
                "num": null,
                "text": "Rank avg (b) = (Rank align (b)+ Rank prox (b))/2, where Rank align (b) is the rank of b's alignment count and Rank prox (b) is the rank of b's proximity count."
            },
            "TABREF0": {
                "text": "An",
                "html": null,
                "num": null,
                "content": "<table><tr><td colspan=\"2\">example of converting a question (Q) with its answer (A) to a SE query and</td></tr><tr><td>retrieving answer passages (AP)</td><td/></tr><tr><td>(Q, A)</td><td>AP</td></tr><tr><td>What is the capital of Pakistan? Answer:( Islamabad)</td><td>Bungalow For Rent in Islamabad, Capital Pakistan. Beautiful Big House For \u2026 Islamabad is the capital of Pakistan. Current</td></tr><tr><td>(k 1 , k 2 ,\u2026, k n , A)</td><td>time, \u2026</td></tr><tr><td>capital, Pakistan, Islamabad</td><td>\u2026the airport which serves Pakistan's capital Islamabad, \u2026</td></tr></table>",
                "type_str": "table"
            },
            "TABREF1": {
                "text": "If none of the above rules are applicable, the question pattern is the question word.",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Rule 2: Question word followed by a light verb and Noun Phrase(NP) or</td></tr><tr><td>Prepositional Phrase(PP) chunk (Example (2)).</td></tr><tr><td>Qp = question word + light verb +headword in the following NP or PP chunk</td></tr><tr><td>Rule 3: Question word followed immediately by a verb (Example (3)).</td></tr><tr><td>Qp = question word + headword in the following Verb Phrase(VP) or NP chunk</td></tr><tr><td>Rule 4: Question word followed by a passive VP (Example (4)).</td></tr><tr><td>Qp = Question word + \"to be\" + headword in the passive VP chunk</td></tr><tr><td>Rule 5: Question word followed by the copulate \"to be\" and an NP (Example (5)).</td></tr><tr><td>Qp = Question word + \"to be\" + headword in the next NP chunk</td></tr><tr><td>Rule 6:</td></tr></table>",
                "type_str": "table"
            },
            "TABREF2": {
                "text": "Example questions and question patterns (of words shown in bold) (1) Which female singer performed the first song on Top of the Pops? (2) Who in 1961 made the first space flight? What is the second longest river in the world?",
                "html": null,
                "num": null,
                "content": "<table><tr><td>(3) Who painted \"The Laughing Cavalier\"?</td></tr><tr><td>(4) What is a group of geese called?</td></tr><tr><td>(5)</td></tr></table>",
                "type_str": "table"
            },
            "TABREF3": {
                "text": "Average rank calculated from for the bigram counterparts of \"how old\"",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Bigrams</td><td colspan=\"4\">Alignment Rank Proximity Rank Avg. Rank Final Rank</td></tr><tr><td>age of</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>years old</td><td>2</td><td>2</td><td>2</td><td>2</td></tr><tr><td>ascend the</td><td>3</td><td>-</td><td>-</td><td>-</td></tr><tr><td>throne in</td><td>4</td><td>3</td><td>3.5</td><td>3</td></tr><tr><td>the youngest</td><td>3</td><td>-</td><td>-</td><td>-</td></tr><tr><td>\u2026</td><td>\u2026</td><td>\u2026</td><td>\u2026</td><td>\u2026</td></tr></table>",
                "type_str": "table"
            },
            "TABREF4": {
                "text": "An example of transformation from question into query",
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td>Question</td><td/></tr><tr><td colspan=\"2\">How old was Bruce Lee when he died?</td><td/></tr><tr><td>Question pattern</td><td>Proper noun</td><td>Keyword</td></tr><tr><td>how old</td><td/><td/></tr><tr><td>Transformation</td><td>\"Bruce Lee\"</td><td>died</td></tr><tr><td>age of, years old</td><td/><td/></tr></table>",
                "type_str": "table"
            },
            "TABREF5": {
                "text": "Evaluation results of question pattern extraction The first five questions with question patterns and judgment",
                "html": null,
                "num": null,
                "content": "<table><tr><td colspan=\"2\">Two \"good\" labels</td><td colspan=\"2\">At least one \"good\" label</td></tr><tr><td>Precision (%)</td><td>86</td><td>96</td></tr><tr><td>Question</td><td/><td>Question pattern</td><td>Judgment</td></tr><tr><td colspan=\"2\">Who is the author of the book, \"The Iron Lady: A Biography of Margaret Thatcher\"?</td><td>Who-author</td><td>good</td></tr><tr><td colspan=\"2\">What was the monetary value of the Nobel Peace Prize in 1989?</td><td>What value</td><td>good</td></tr><tr><td colspan=\"2\">What does the Peugeot company manufacture?</td><td>What do manufacture</td><td>good</td></tr><tr><td colspan=\"2\">How much did Mercury spend on advertising in 1993?</td><td>How much</td><td>good</td></tr><tr><td colspan=\"2\">What is the name of the managing director of Apricot Computer?</td><td>What name</td><td>bad</td></tr></table>",
                "type_str": "table"
            }
        }
    }
}