| { | |
| "paper_id": "M92-1020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T03:13:19.289878Z" | |
| }, | |
| "title": "UNIVERSITY OF MARYLAND/CONQUEST : MUC-4 TEST RESULTS AND ANALYSI S", | |
| "authors": [ | |
| { | |
| "first": "James", | |
| "middle": [], | |
| "last": "Mayfield", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Maryland Baltimore County Baltimore", | |
| "location": { | |
| "postCode": "21228-5398", | |
| "region": "MD", | |
| "country": "USA" | |
| } | |
| }, | |
| "email": "mayfield@cs.umbc.edu" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "Only a subset of the MUC-4 slots were generated by the UM/ConQuest ICTOAN system. Our scores fo r the slots we attempted to fill are shown in Figure 1. 'This slot was filled once inadvertentally due to a bug in the semantic net .", | |
| "pdf_parse": { | |
| "paper_id": "M92-1020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "Only a subset of the MUC-4 slots were generated by the UM/ConQuest ICTOAN system. Our scores fo r the slots we attempted to fill are shown in Figure 1. 'This slot was filled once inadvertentally due to a bug in the semantic net .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "The most significant limiting factor we experienced was development time . The ICTOAN system was writte n from scratch during the first five months of 1992 ; only the template generation software was reused fro m our MUG3 system . The effect of the short development time was compounded by our selection of C as the programming language for the project . We chose C for its speed, and to gain leverage from existing ConQuest software . However, because C is such a low-level language, development of a piece of code in C takes longer than development of equivalent code in a higher-level language such as LISP.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "LIMITING FACTORS", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Recall scores for the evaluation system were uniformly poor . The main factors contributing to the low recall scores were :", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "LIMITING FACTORS", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u2022 The evaluation system did not attempt to fill all template slots . In particular, we did not attempt t o guess any slot values that were not known, even when guessing would lead to a significant partial scor e (e .g. in the data slot) . The following slots were not filled : \u2022 No grammar rules were in place for performing semantic analysis of nominalized attack actions, or fo r interpretation of passive verbs . Therefore, the only attacks that were interpreted were ones that wer e described by active verbs . A low percentage of the reported attacks were described in this way in th e test corpora.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "LIMITING FACTORS", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u2022 No merging of co-referential phrases or sentences was performed . Thus, two interpretable descriptions of the same attack always led to the generation of two different templates . In addition to increasing overgeneration, this had the effect of reducing recall for slots that were filled in templates that wer e eventually discarded by the scoring program .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "LIMITING FACTORS", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u2022 Although significant hooks are in place in the system for discourse segmentation, the lack of merging of co-referential items made discourse segmentation algorithms irrelevant for the purposes of th e evaluation .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "LIMITING FACTORS", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Most of the development time for the ICTOAN system was spent on writing the basic system code and on developing the semantic net of world knowledge . Relatively little time was spent in grammar-writing. Because we had the Proximity Linguistic System available for our dictionary, little time was spent o n dictionary development . This turned out to be problematic, because the dictionary contained many obscur e or questionable word senses . For example, the word `the' was listed as an adverb (as in `the higher the fewer') . Dictionary cleanup is the task we would most like to redo .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "ALLOCATION OF RESOURCE S", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The two main successes of the ICTOAN system were its speed and its flexibility .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "SUCCESSES", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The system is quite fast . For example, the system processed the 100 stories in the TST3 corpus in unde r twenty minutes on a lightly-loaded Sun Sparc II with 24MB of memory . The evaluation system had a number of known memory leaks . Because of these leaks, we chose to break up the input texts into individual file s and run ICTOAN separately on each . Given time to eliminate these leaks, we believe that the system is capable of processing 100 texts on the same platform in under twelve minutes .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Speed", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "One result made possible by the system's speed is that the ICTOAN system does no skimming. That is, the system does not look for `hot' areas of the text and concentrate its processing power on those areas . Rather, the system processes every sentence of every story to the best of its abilities.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Speed", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The ICTOAN system architecture is best viewed as a set of parallel streams of data flowing through a pipeline of processes . This stream-based architecture makes it possible to easily integrate top-down an d bottom-up processing . Bottom-up processes can be written to read ,data from low-level streams and buil d items to be placed on high-level streams . Top-down processes can be written to search for desired elements on low-level streams, based on the items that appear in high-level streams.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Flexibility", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Since individual processes may be placed in any order along the pipeline, it is easy to explore th e interactions between for example top-down and bottom-up processes . The user needs only to modify a configuration file that indicates which processes should be used and in what order . Furthermore, the overal l effectiveness of a particular process can be evaluated by comparing the performance of the system runnin g with the component of interest against the performance of the system running without that component .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Flexibility", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "We believe that the flexibility of the system will make it easily adaptable to other uses . For example, we intend to use all of the system except the template generator as an automatic text-to-hypertext conversio n system [1] .", | |
| "cite_spans": [ | |
| { | |
| "start": 223, | |
| "end": 226, | |
| "text": "[1]", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "acknowledgement", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Using semantic nets to enrich hypertext links", | |
| "authors": [ | |
| { | |
| "first": "James", | |
| "middle": [], | |
| "last": "Mayfield", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Charles", | |
| "middle": [], | |
| "last": "Nicholas", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1992, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "James Mayfield and Charles Nicholas. Using semantic nets to enrich hypertext links . Technical Repor t CS-92-02, University of Maryland Baltimore County, Baltimore, MD, 1992 .", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": {} | |
| } | |
| } |