File size: 14,829 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 | {
"paper_id": "M91-1012",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T03:15:31.645812Z"
},
"title": "MCDONNELL DOUGLAS ELECTRONIC SYSTEMS COMPANY : MUC-3 Test Results and Analysi s",
"authors": [
{
"first": "David",
"middle": [],
"last": "De Hilster",
"suffix": "",
"affiliation": {
"laboratory": "Advanced Computing Technologies Lab McDonnell Douglas Electronics Systems Compan y",
"institution": "East Saint Andrew Plac e Santa Ana",
"location": {
"postCode": "1801, 92705-652 0",
"region": "California"
}
},
"email": ""
},
{
"first": "Amnon",
"middle": [],
"last": "Meyers",
"suffix": "",
"affiliation": {
"laboratory": "Advanced Computing Technologies Lab McDonnell Douglas Electronics Systems Compan y",
"institution": "East Saint Andrew Plac e Santa Ana",
"location": {
"postCode": "1801, 92705-652 0",
"region": "California"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "INLET scored 25% recall and 35% precision after interactive correction. Noninteractive scores were 22% and 32%, respectively. Relative to other sites, we ranked 7th i n recall and 9th in precision. Because we are in transition from one system (VOX) to another (INLET), our score s reflect the performance of a very new and incomplete analyzer, as well as a small vocabulary and knowledge base. To be able to participate in MUC3, we implemented only a skimming capability, rather than a full-fledged syntax-driven language analyzer. Considering the statu s of our project, we feel INLET's performance is highly commendable. We ourselves were surprised by the success of skimming alone in performance on the MUC3 task, especiall y considering the preliminary nature of our work in this area .",
"pdf_parse": {
"paper_id": "M91-1012",
"_pdf_hash": "",
"abstract": [
{
"text": "INLET scored 25% recall and 35% precision after interactive correction. Noninteractive scores were 22% and 32%, respectively. Relative to other sites, we ranked 7th i n recall and 9th in precision. Because we are in transition from one system (VOX) to another (INLET), our score s reflect the performance of a very new and incomplete analyzer, as well as a small vocabulary and knowledge base. To be able to participate in MUC3, we implemented only a skimming capability, rather than a full-fledged syntax-driven language analyzer. Considering the statu s of our project, we feel INLET's performance is highly commendable. We ourselves were surprised by the success of skimming alone in performance on the MUC3 task, especiall y considering the preliminary nature of our work in this area .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Like many other developers of NLP systems, we have had no reason, up to now, to parameterize the trade-off between recall and precision . At various points in the past fe w months, we had tested heuristics that increased recall while reducing precision .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TEST SETTING S",
"sec_num": null
},
{
"text": "Participation in MUC3 has led us to examine such a parameter more seriously . In particular, a confidence metric, whereby the system assesses its own confidence i n understanding a text, seems to be a useful component of NLP system development . We can attac h such confidence factors to every action that the system takes in processing text, in extractin g information from text, and in correlating the information to produce output .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TEST SETTING S",
"sec_num": null
},
{
"text": "As an NLP system improves and its knowledge of linguistics and the domain become mor e complete, confidence factors can be raised . Additionally, an arsenal of heuristics could be mad e available, depending on the confidence threshold assigned to the language processing task .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TEST SETTING S",
"sec_num": null
},
{
"text": "Assignment of confidence ratings can be assisted by empirical data . For example, we can assign an overall system confidence in identifying the incident type by scoring performance fo r the entire MUC3 corpus .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TEST SETTING S",
"sec_num": null
},
{
"text": "Once the skimmer was completed (April 22), we spent approximately 2 man-month s customizing INLET to the MUC3 domain and task . A substantial portion of that time was spent i n implementing code for filling the various slots in the MUC3 template and in developin g heuristics to improve recall and precision . Somewhat less time was spent in building an d applying generic and domain-specific grammar rules . System testing was relatively painless , because INLET typically processed 100 messages in 45 minutes, and several Sun Sparcstation s were typically available for running tests . Unfortunately, we allowed too little time fo r vocabulary addition, so that many important words and phrases were omitted (e .g . , \"machinegun\", \"automatic weapons\") .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "EXPENDITURE OF EFFOR T",
"sec_num": null
},
{
"text": "Our main problems resulted from working with a new and incomplete system . Too ofte n we had to devote our time to fixing bugs or making improvements in the skimmer, in th e graphic representation tools, and in the knowledge addition tools . Starting with a smal l vocabulary and little linguistic and domain knowledge was disadvantageous . Adding a lot o f knowledge to the system over a short period of time caused many problems to surface (e .g., initializing the system became a time-waster, system tables overflowed several times) . Lack o f an internal representation for information extracted from the text was yet another limitation o n development .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LIMITING FACTOR S",
"sec_num": null
},
{
"text": "We used the entire development corpus, including the key templates, for gatherin g domain information . For example, we used the key templates to get fairly complete lists o f perpetrators, targets, instruments, and so on . Similarly, we searched the corpus for keywords , temporal, locative, and other patterns . Many of our domain-specific grammar rules wer e crafted using the results of such searches .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TRAININ G",
"sec_num": null
},
{
"text": "The first 100 messages of the development set served as a primary development an d testing vehicle . TST1 messages were run occasionally in order to gauge progress on unseen message sets .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TRAININ G",
"sec_num": null
},
{
"text": "In order to shake out bugs in the system, we processed half the development set in batches of 100 messages several days before the testing deadline .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TRAININ G",
"sec_num": null
},
{
"text": "In general, skimming worked much better than expected . Based merely on our initial results for MUC3, we conclude that skimming is a powerful adjunct to deeper processing of text . We feel that with several months' work, continued development of skimming techniques , combined with knowledge base and vocabulary development, would substantially raise our MUC 3 score . Skimming provides extremely fast, simple, and robust text processing . While keyword and pattern-based methods for NLP have usually met with scorn, we feel a review of thes e methods is called for .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "STRENGTHS AND WEAKNESSE S",
"sec_num": null
},
{
"text": "On the other hand, we are aware of the limitations of any approach that doesn't analyz e text as deeply as possible . In order to segment incidents with great accuracy, linguistic contex t as well as script-level understanding of the text are required . Many reference resolutio n problems also require such knowledge .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "STRENGTHS AND WEAKNESSE S",
"sec_num": null
},
{
"text": "In the near future, we will merge our skimming capability with a bottom-up syntacti c analysis mechanism, and also incorporate script-based understanding mechanisms .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "STRENGTHS AND WEAKNESSE S",
"sec_num": null
},
{
"text": "The INLET customization tools have proved their worth by supporting hierarchy , grammar rule, and vocabulary addition . Even our qualified success would have been impossibl e without the effectiveness of the knowledge addition framework .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "STRENGTHS AND WEAKNESSE S",
"sec_num": null
},
{
"text": "Our system is fairly good at determining the incident type, using a hierarchy of ke y words and patterns . With just a few specialized rules, the system is able to process appositive s to find perpetrators, perpetrator organizations, physical targets, and human targets . An extensive temporal grammar was developed, though not much correlation of multiple tempora l references has been implemented . A similar situation holds for locative phrases .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "HITS AND MISSES",
"sec_num": null
},
{
"text": "Simple gaps in knowledge and vocabulary caused many misses on the TST2 messages . Missing vocabulary (e .g., \"killings\"), missing domain rules (e .g., \"explosion caused damage\") , missing generic rules (e.g., \"actor participated in action on object\"), and missing mechanism s of various kinds led to substantially lower performance than we would expect of a more matur e INLET system . Missing mechanisms include lack of threat handling, lack of any inferencin g capability, lack of spelling correction, and lack of rejection of incidents for even simple reason s (e.g., an abstract object such as the \"economy\" is attacked) .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "HITS AND MISSES",
"sec_num": null
},
{
"text": "Because INLET is a customization shell, portability of the specific knowledge added to th e system is not a major concern . In 2 man-months, we were able to achieve a 25% recall and 35% precision score with a relatively immature INLET system . When the system is completed , we expect similar customization time to result in a better system for the particular domain an d task .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PORTABILIT Y",
"sec_num": null
},
{
"text": "The skimmer framework and knowledge addition framework are generic, as is the core knowledge base and vocabulary . On top of this layer is a substantial body of domain-specifi c code and knowledge, which would necessarily have to be replaced for a new domain and task .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "PORTABILIT Y",
"sec_num": null
},
{
"text": "We have demonstrated that the INLET knowledge addition framework and skimmer ca n quickly support customization to a new domain and task . We have found that the graphic interface for knowledge addition has speeded up customization over a system like VOX . Finally , we have found that skimming is a critical adjunct to deeper NLP .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LESSONS LEARNED",
"sec_num": null
},
{
"text": "Our participation in MUC3 has shown us the high value of formal testing and compariso n with other NLP efforts . We intend to continue using the MUC3 corpus and testing system for ou r system development, test, and evaluation .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LESSONS LEARNED",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {},
"ref_entries": {}
}
} |