File size: 40,091 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
{
    "paper_id": "M92-1007",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T03:13:22.760839Z"
    },
    "title": "BBN PLUM : MUC-4 Test Results and Analysi s",
    "authors": [
        {
            "first": "Ralph",
            "middle": [],
            "last": "Weischedel",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": "weischedel@bbn.com"
        },
        {
            "first": "Damaris",
            "middle": [],
            "last": "Ayuso",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": ""
        },
        {
            "first": "Sean",
            "middle": [],
            "last": "Boisen",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": ""
        },
        {
            "first": "Heidi",
            "middle": [],
            "last": "Fox",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": ""
        },
        {
            "first": "Herbert",
            "middle": [],
            "last": "Gish",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": ""
        },
        {
            "first": "Robert",
            "middle": [],
            "last": "Ingria",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "BBN Systems and Technologie s",
                "location": {
                    "addrLine": "10 Moulton St",
                    "postCode": "0213 8",
                    "settlement": "Cambridge",
                    "region": "MA"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Our mid-term to long-term goals in data extraction from text for the next one to three years are to achieve muc h greater portability to new languages and new domains, greater robustness, and greater scalability. The nove l aspect to our approach is the use of learning algorithms and probabilistic models to learn the domain-specific an d language-specific knowledge necessary for a new domain and new language. Learning algorithms should contribute to scalability by making it feasible to deal with domains where it would be infeasible to invest sufficient huma n effort to bring a system up. Probabilistic models can contribute to robustness by allowing for words, constructions, and forms not anticipated ahead of time and by looking for the most likely interpretation in context. We began this research agenda approximately two years ago. During the last twelve months, we have focused much of our effort on porting our data extraction system (PLUM) to a new language (Japanese) and to two ne w domains. During the next twelve months, we anticipate porting PLUM to two or three additional domains .",
    "pdf_parse": {
        "paper_id": "M92-1007",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Our mid-term to long-term goals in data extraction from text for the next one to three years are to achieve muc h greater portability to new languages and new domains, greater robustness, and greater scalability. The nove l aspect to our approach is the use of learning algorithms and probabilistic models to learn the domain-specific an d language-specific knowledge necessary for a new domain and new language. Learning algorithms should contribute to scalability by making it feasible to deal with domains where it would be infeasible to invest sufficient huma n effort to bring a system up. Probabilistic models can contribute to robustness by allowing for words, constructions, and forms not anticipated ahead of time and by looking for the most likely interpretation in context. We began this research agenda approximately two years ago. During the last twelve months, we have focused much of our effort on porting our data extraction system (PLUM) to a new language (Japanese) and to two ne w domains. During the next twelve months, we anticipate porting PLUM to two or three additional domains .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "For any group to participate in MUC is a significant investment. To be consistent with our mid-term and longterm goals, we imposed the following constraints on ourselves in participating in MUC-4 :",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We would focus our effort on semi-automatically acquired knowledge.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We would minimize effort on handcrafted knowledge, and most generally .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We would minimize MUC-specific effort . Though the three self-imposed constraints meant our overall scores on the objective evaluation were not as high as if we had focused on handtuning and handcrafting the knowledge bases, MUC-4 became a vehicle for evaluating our progress on the long-term goals.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "PLUM had demonstrated quite high recall in MUC-3 and scored among the top systems . We chose to focus o n the following goals in MUC-4 :",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "MEASURING SUCCESS IN ACHIEVING OUR SHORT-TERM GOAL S",
                "sec_num": null
            },
            {
                "text": "Increasing precision and reducing overgeneration, without hurting recall . Demonstrating a broad range of tradeoff in recall and precision. Goal 1 : Increasing precision and reducing overgeneration, without hurting recall . As the graph in Figure 1 shows, we doubled our precision in MUC-4 (compared to MUC-3) and reduced our overgeneration by roughly one third. The overall impact was to increase PLUM's F-Measure by 50% . Naturally one would ideally base thi s measurement on the new test sets for MUC-4 (TST3, TST4) ; however, between MUC-3 and MUC-4 both th e definition of the templates to be produced and the evaluation function changed dramatically, so that there was n o easy way to run the MUC-3 version of PLUM on TST3 and TST4 to produce results comparable to that of th e MUC-4 version of the system. However, since the Government had converted our MUC-3 TST2 templates to the MUC-4 format, and since we had never examined the corpus of messages or the answer key to TST2, we coul d easily use it as a basis for comparison. Figure 2 illustrates, the user ca n select from a broad range of system performance, emphasizing either recall or precision to various degrees . No system had displayed such a span favoring recall versus favoring precision in MUC-3 . Only one other system, GE's , demonstrated a broad range; at a cost of 17 points of recall, GE's system could achieve an increase of roughly 8 points of precision. For PLUM, the tradeoff of recall for precision was far more balanced . Two independent parameters primarily contributed to this : a discrete parameter controlling how aggressively o r conservatively two descriptions are fused into a single view of the same event, and a continuously variable threshol d on a classification algorithm predicting whether a paragraph is relevant or irrelevant (with respect to reporting any terrorist incident) . Together, these two parameters offer a user the ability to turn a knob to emphasize recall or precision based on their application preference .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 240,
                        "end": 248,
                        "text": "Figure 1",
                        "ref_id": null
                    },
                    {
                        "start": 1036,
                        "end": 1044,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "MEASURING SUCCESS IN ACHIEVING OUR SHORT-TERM GOAL S",
                "sec_num": null
            },
            {
                "text": "\u00ae 5/91 -MUC-3 n 5/92 -MUC-4",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "MUC-3 System vs. MUC-4 System on TST2",
                "sec_num": null
            },
            {
                "text": "Two design features stand out in our minds: partial understanding and statistical language modeling . By partia l understanding we mean that the parser and grammar are designed to find analyses for a non-overlapping sequence o f fragments. When cases of permanent, predictable ambiguity arise, such as a prepositional phrase that can b e attached in multiple ways, or most conjoined phrases, the parser finishes the analysis of the current fragment, an d begins the analysis of a new fragment . Therefore, the entities mentioned and some relations between them are processed in every sentence, whether syntactically ill-formed, complex, novel, or straightforward . Furthermore, this parsing is done using essentially domain-independent syntactic information. The semantic interpreter and the rest o f the system in turn do not assume having complete understanding.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "KEY SYSTEM FEATURE S",
                "sec_num": null
            },
            {
                "text": "The second key feature is the use of statistical algorithms to guide processing . Determining the part of speec h of highly ambiguous words is done by well-known Markov modeling techniques . To improve the recognition of Latin American names, we employed a statistically derived five-gram (five letter) model of words of Spanish origi n and a similar five-gram model of English words . This model was integrated into the part-of-speech tagger.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "KEY SYSTEM FEATURE S",
                "sec_num": null
            },
            {
                "text": "Another usage of statistical algorithms was a statistical induction algorithm to learn case frames for verbs fro m examples . This saved substantial effort compared to building the case frames by hand. The algorithm and empirical results are described in [3] . The statistical methods mentioned above were already available and used in the MUC-3 version of PLUM . A new statistical algorithm employed in MUC-4 is a classification algorithm that automatically learns features to discriminate among classes . Given a list of relevant paragraphs and a list of irrelevant ones (made available by Ne w Mexico State University), we employed a chi square measurement to determine word stems (though other features could be used as well) whose presence or absence in text is significantly correlated with the text being relevant (o r irrelevant) . Given that ordering, the user must select how many features to use . At runtime, the classifier sums the logarithm of the odds that the paragraph is relevant given the presence of the features . If the sum exceeds a userspecified threshold, the paragraph is considered relevant . If the classifier predicts that the paragraph is relevant, then events found in the paragraph can be used to generate templates ; if not, terrorist events that would otherwise have been produced from that paragraph are blocked . The performance of the overall system, given variou s thresholds of the text classifier, is shown in Figure 3 . ",
                "cite_spans": [
                    {
                        "start": 255,
                        "end": 258,
                        "text": "[3]",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1450,
                        "end": 1458,
                        "text": "Figure 3",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "KEY SYSTEM FEATURE S",
                "sec_num": null
            },
            {
                "text": "A more detailed description of the system components, their individual outputs, and their knowledge bases is presented in Ayuso et al., [1] . We expect the particular implementations to change and improve substantially during the next three years of research and development .",
                "cite_spans": [
                    {
                        "start": 136,
                        "end": 139,
                        "text": "[1]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "38",
                "sec_num": null
            },
            {
                "text": "Appendix G lists detailed test scores. A number of systems performed better on TST4 than on TST3, and som e performed significantly worse on TST4 than TST3 . The results on the two test sets were so disparate for PLUM tha t we decided to look into the causes of the abnormally low recall of PLUM on TST3 . As table 1 shows, the followin g properties of TST3 stand in stark contrast with TST4, TST1\" and the 1300 message development corpus:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "RESULT S",
                "sec_num": null
            },
            {
                "text": "\u2022 The percentage of relevant articles was abnormally high for TST3 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "RESULT S",
                "sec_num": null
            },
            {
                "text": "\u2022 The density of templates produced, i.e., the number of templates per 100 messages was abnormally high , roughly 50% higher than the 1300 message development set, TST1, and TST4 . \u2022 The density of human targets produced, i.e., the number of human target slot fills per 100 messages, was abnormally high, roughly double the number in TST1, TST4, or the development corpus . Taken together, the first two observations above suggest the following: Systems tuned to overgenerate (i .e., produce a high percentage of templates, what Hirschman labels the \"lazy merge problem\" in this volume) should perform significantl y better on TST3 than TST4. The observations above suggest, in part, why PLUM's recall for TST3 was abnormally low:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "RESULT S",
                "sec_num": null
            },
            {
                "text": "\u2022 PLUM's merging algorithm is under user-control, and was set to avoid overgeneration, therefore yieldin g worse performance on TST3 than on TST4 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Relevant Messages",
                "sec_num": null
            },
            {
                "text": "\u2022 A large number of human targets exercise two known weaknesses of the MUC-4 version of PLUM: (1) known, temporary grammar problems and (2) a challenge for discourse processing to be able to collect target s across sentences. \u2022 A bug in the official scoring program was encountered for TST3, but not for TST4 . If this bug is corrected, we estimate it would improve PLUM's scores by at least one point in recall, at least one point in precision, and at least two points in overgeneration. (That clearly is not sufficient to fully account for the discrepancy in performance on TST3 and TST4 .) One other point confirming the normalcy of TST4, contrasted with the abnormal characteristics of TST3, can b e seen in PLUM's performance under various settings . Prior to the test, we ran PLUM with numerous parameter settings on TST1, TST2, and one set of 100 messages from the development set. This predicted the setting that would maximize the F-Measure, or come indistinguishably close to the maximum F-measure. That prediction proved correct (consistent) with TST4, but was 2 points under the maximum actually achieved for TST3 via one of our optional runs. Table 2 summarizes PLUM's performance on TST3 where precision is maximized (the required run), where recall is maximized, and where F is maximized. It also lists the required run for TST4. In addition, since TST3 and TST 4 were so disparate in character, we computed the score of PLUM if TST3 and TST4 together constituted the test.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1156,
                        "end": 1163,
                        "text": "Table 2",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Relevant Messages",
                "sec_num": null
            },
            {
                "text": "We estimate that 4 person months specific to MUC-4 went into our effort . These were spent approximately as follows : domain-dependent lexical additions, 0 .5 person months; grammar, 0.5 person months; semantic rules , 0.75 person months; discourse, 1 .0 person months ; backend, 0.75 person months ; and overhead (evaluation , fulfilling requirements, etc .), 0.5 person months .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "EFFORT SPENT",
                "sec_num": null
            },
            {
                "text": "The 1300 messages of the development corpus were used at various levels as training data . PLUM was run over all 1300 messages to detect, debug, and correct any causes of system breaks . The perpetrator organization slot for all 1300 messages was used to quickly add names to the domain-dependent lexicon . After running our part-ofspeech tagger (POST) over the development corpus, the statistical algorithm for predicting words of Spanish origi n was run over the list of previously unknown words. Those predicted as Spanish in origin were then reviewed manually to add Spanish names to the lexicon .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TRAINING DATA AND TECHNIQUES",
                "sec_num": null
            },
            {
                "text": "A subset of the development set was used more intensively as training data . Approximately 95,000 words of text (about 20% of the development corpus) was tagged by the University of Pennsylvania as to part of speech an d labelled as to syntactic structure as part of the DARPA-funded TREEBANK project. The bracketed text firs t provided us with a frequency-ranked list of head verbs, head nouns, and nominal compounds . For each of these we added a pointer to the domain model element that is the most specific super-concept containing all things denoted by the verb, noun, or nominal compound. As mentioned earlier, the TREEBANK data was then used with the lexical relation to the domain model to hypothesize case frames for verbs . The automatically hypothesized verb cas e frames were then reviewed manually and added to the lexicon . This is detailed in [3] .",
                "cite_spans": [
                    {
                        "start": 858,
                        "end": 861,
                        "text": "[3]",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TRAINING DATA AND TECHNIQUES",
                "sec_num": null
            },
            {
                "text": "The 100 messages of TST1 and TST2 were used as blind test sets to measure our progress at least once a week. Throughout, we only looked at the summary output from the scoring procedure, rather than adding to the lexicon or debugging the system based on particular messages .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TRAINING DATA AND TECHNIQUES",
                "sec_num": null
            },
            {
                "text": "The training mentioned above had already been used in preparing for MUC-3, with the obvious exception that TST2 was not available in preparation for MUC-3 . What we added in MUC-4 was training regarding the relevance/irrelevance of paragraphs. We tried training at the article level; however, the fact that an article could b e mostly irrelevant except for a single paragraph mentioning a terrorist incident made the training much less effectiv e than training based on labelling individual paragraphs as irrelevant.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "TRAINING DATA AND TECHNIQUES",
                "sec_num": null
            },
            {
                "text": "Successes. Though the structure of PLUM did not change radically between MUC-3 and MUC-4, the one ne w component, a statistically based text classification algorithm, was quite successful . It was trained fully automatically . Once the system is trained on sets of text representing the various classes, such as relevant paragraphs versus irrelevant paragraphs for the MUC domain, the user need do only two things : select a cutoff for words to be used for each class in the log probability model and set a continuous variable which serves as th e threshold for inclusion in a given class . The result was a significant, continuous tradeoff in recall versus precision .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            },
            {
                "text": "A second new success in our experience this year was the heuristic for when to merge (fuse) two descriptions into a single event representation. Many of the heuristics do not require domain-specific knowledge . The heuristi c was not knowledge-intensive, yet significantly reduced overgeneration, while increasing precision .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            },
            {
                "text": "These two heuristics together enabled us to double precision and reduce overgeneration by one third, thus effecting an overall improvement in performance (F-measure) by 50% compared to the MUC-3 version of PLUM.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            },
            {
                "text": "What Limited Success. Having an investment in the terrorist domain and FBIS corpus inhibited significan t changes that we might otherwise have made, such as (1) moving away from a purely deterministic parser (i .e., a beam search of width one) to a more general probabilistically controlled beam search, (2) moving from the present semantic interpreter to a more declarative one, and (3) replacing the MUC backend with a general purpose one driven by knowledge bases derived in part from training data. Improvements Desired . Improving syntactic coverage is a priority. Increased coverage normally leads t o greater perceived ambiguity in the system; we hope to counter that through probabilistic models . A second priority is improving coverage of the discourse component. The template generator today is based on handcrafted rules of thumb. Within the next year we hope to develop and test an acquisition algorithm that would acquire most of the rules from examples in a new domain . Lastly, though the classification algorithm was a pleasant success, we believe an even more accurate classifier is possible .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            },
            {
                "text": "Lessons Learned . There are several lessons we believe we learned this year: 1 . Automatic training and acquisition of knowledge bases can yield relatively good performance at reduce d labor . Suppose one plots F-measure on TST3 and TST4 together against total effort in MUC-3 and MUC-4 together . PLUM achieved high performance with very little labor (7 person-months in MUC-4 and MUC-3 combined). Of the top eight sites in MUC-4, compared to our effort, one group put in at least 50% more effort than we did; all other groups put in roughly 2-6 times the effort in MUC-3 and MUC-4 combined .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            },
            {
                "text": "2. Substantial tradeoff in recall and precision is achievable, in particular nearly even tradeoff of recall for precision is achievable over a broad range. 3. User control of the recall/precision tradeoff is attainable via a continuous variable, e .g, via a knob the user can turn to prefer recall or to prefer precision in varying degrees . 4. A test set of 100 messages seems too small to accurately assess system performance . Scores for three systems decreased dramatically in moving from TST3 to TST4 ; of those, two were in the top eight systems. By contrast, scores for five systems increased dramatically in moving from TST3 to TST4 . Since the discrepancy between TST3 and TST4 was great for half of the systems, and since those systems divide d almost evenly on which test set gave higher performance, it is clear that a set of 100 messages, unles s carefully chosen to balance characteristics of the test set, is too small . 5. Unfortunately, given the statistics in Table 1 , TST4 seems more representative than TST3 . With hindsight it seems clear that combining the results of TST3 and TST4 would have given a useful measure . t",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 978,
                        "end": 985,
                        "text": "Table 1",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "CONCLUSIONS",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "The",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "ACKNOWLEDGMENT S",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "BBN: Description of the PLUM System as Used for MUC-4",
                "authors": [
                    {
                        "first": "D",
                        "middle": [
                            "M"
                        ],
                        "last": "Ayuso",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Boisen",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Fox",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Ingria",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    }
                ],
                "year": 1992,
                "venue": "MUC-4 Proceedings",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ayuso, D.M ., Boisen, S ., Fox, H ., Ingria, R., and Weischedel, R. \"BBN: Description of the PLUM System as Used for MUC-4\". MUC-4 Proceedings, 1992 .",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Partial Parsing, A Report on Work in Progress",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [
                            "M"
                        ],
                        "last": "Ayuso",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Bobrow",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Boisen",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Ingria",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Palmucci",
                        "suffix": ""
                    }
                ],
                "year": 1991,
                "venue": "Proceedings of the Fourth DARPA Workshop on Speech and Natural Language",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Weischedel, R., Ayuso, D.M., Bobrow, R., Boisen, S ., Ingria, R ., and Palmucci, J ., Partial Parsing, A Report on Work in Progress, Proceedings of the Fourth DARPA Workshop on Speech and Natural Language, 1991a.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Applications of Statistical Language Modelling to Natural Language Processing",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Meteer",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Schwartz",
                        "suffix": ""
                    }
                ],
                "year": 1991,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Weischedel, R., Meteer, M ., and Schwartz, Applications of Statistical Language Modelling to Natural Language Processing, unpublished manuscript, 1991b.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "COR (correct), PAR (partial), MIS (missing), and SPU (spunous), then using the definitions of recall",
                "authors": [],
                "year": null,
                "venue": "One can reasonably combine the scores by adding the scoring entries for the columns POS (possible), AC T (actual)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "One can reasonably combine the scores by adding the scoring entries for the columns POS (possible), AC T (actual), COR (correct), PAR (partial), MIS (missing), and SPU (spunous), then using the definitions of recall , precision, overgeneration, and F on the totals .",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "type_str": "figure",
                "num": null,
                "uris": null,
                "text": "PLUM's precision was doubled; overgeneration was cut by 1/3 ; overall performance (F-Measure ) increased 50% ; all without hurting recall. By varying parameters, a wide range of recall and precision can be obtained ."
            },
            "FIGREF1": {
                "type_str": "figure",
                "num": null,
                "uris": null,
                "text": "Impact of Paragraph classifier on recall and precision in the ALL TEMPLATES row ."
            },
            "TABREF1": {
                "html": null,
                "num": null,
                "content": "<table/>",
                "type_str": "table",
                "text": ""
            },
            "TABREF2": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>SLOT</td><td>REC</td><td>PRE</td><td>OVG</td><td>SLOT</td><td>REC</td><td>PRE</td><td>OVG</td></tr><tr><td>MATCHED/MISSING</td><td>30</td><td>69</td><td>10</td><td>MATCHED/MISSING</td><td>49</td><td>72</td><td>9</td></tr><tr><td>MATCHED/SPURIOUS</td><td>51</td><td>44</td><td>43</td><td>MATCHED/SPURIOUS</td><td>55</td><td>28</td><td>64</td></tr><tr><td>MATCHED ONLY</td><td>51</td><td>69</td><td>10</td><td>MATCHED ONLY</td><td>55</td><td>72</td><td>9</td></tr><tr><td>ALL TEMPLATES</td><td>30</td><td>44</td><td>43</td><td>ALL TEMPLATES</td><td>49</td><td>28</td><td>64</td></tr><tr><td>SET FILLS ONLY</td><td>33</td><td>71</td><td>14</td><td>SET FILLS ONLY</td><td>52</td><td>75</td><td>9</td></tr><tr><td>STRING FILLS ONLY</td><td>23</td><td>64</td><td>12</td><td>STRING FILLS ONLY</td><td>43</td><td>70</td><td>15</td></tr><tr><td>TEXT FILTERING</td><td>83</td><td>87</td><td>13</td><td>TEXT FILTERING</td><td>90</td><td>67</td><td>5 1</td></tr><tr><td/><td colspan=\"3\">P&amp;R 2P&amp;R P&amp;2R</td><td/><td colspan=\"3\">P&amp;R 2P&amp;R P&amp;2R</td></tr><tr><td>F-MEASURES</td><td>35 .68</td><td>40.24</td><td>32 .04</td><td>F-MEASURES</td><td>35.64</td><td>30.62</td><td>42 .61</td></tr><tr><td colspan=\"4\">Optional Run on TST3 (settings maximize F-</td><td colspan=\"3\">TST3 and TST4 Combined (settings favo r</td><td/></tr><tr><td>measure) SLOT</td><td>REC</td><td>PRE</td><td>OVG</td><td>precision) SLOT</td><td>REC</td><td>PRE</td><td>OV G</td></tr><tr><td>MATCHED/MISSING MATCHED/SPURIOUS MATCHED ONLY ALL TEMPLATES</td><td>38 52 52 38</td><td>69 38 69 38</td><td>10 50 10 50</td><td>MATCHED/MISSING MATCHED/SPURIOUS MATCHED ONLY ALL TEMPLATES</td><td>35% 52% 52% 35%</td><td>71% 44% 71% 44%</td><td>9 % 44 % 9 % 44 %</td></tr><tr><td>SET FILLS ONLY</td><td>40</td><td>70</td><td>12</td><td>SET FILLS ONLY</td><td>37%</td><td>73%</td><td>11 %</td></tr><tr><td>STRING FILLS ONLY TEXT FILTERING F-MEASURES</td><td colspan=\"3\">30 98 P&amp;R 2 P&amp;R P&amp;2R 64 15 80 20 38 .00 38 .00 38 .00</td><td>STRING FILLS ONLY TEXT FILTERING F-MEASURES</td><td colspan=\"3\">29% 83% P&amp;R 2 P&amp;R P&amp;2R 68% 12 % 81% 19 % 38.74 41 .76 36 .1 3</td></tr><tr><td colspan=\"4\">Required Run on TST4 (settings favor precision)</td><td/><td/><td/><td/></tr><tr><td>SLOT</td><td>REC</td><td>PRE</td><td>OVG</td><td/><td/><td/><td/></tr><tr><td>MATCHED/MISSING</td><td>40</td><td>72</td><td>8</td><td/><td/><td/><td/></tr><tr><td>MATCHED/SPURIOUS</td><td>53</td><td>42</td><td>47</td><td/><td/><td/><td/></tr><tr><td>MATCHED ONLY</td><td>53</td><td>72</td><td>8</td><td/><td/><td/><td/></tr><tr><td>ALL TEMPLATES</td><td>40</td><td>42</td><td>47</td><td/><td/><td/><td/></tr><tr><td>SET FILLS ONLY</td><td>42</td><td>75</td><td>9</td><td/><td/><td/><td/></tr><tr><td>STRING FILLS ONLY</td><td>36</td><td>72</td><td>1 2</td><td/><td/><td/><td/></tr><tr><td>TEXT FILTERING</td><td>8 2</td><td>7 1</td><td>29</td><td/><td/><td/><td/></tr><tr><td/><td>P&amp;R</td><td>2 P&amp;R</td><td>P&amp;2R</td><td/><td/><td/><td/></tr><tr><td>F-MEASURES</td><td>40 .98</td><td>41 .58</td><td>4038</td><td/><td/><td/><td/></tr></table>",
                "type_str": "table",
                "text": "A summary of scores on TST3 and TST4."
            }
        }
    }
}