CRL/Brandeis: Description of the Diderot System as Used for MUC-5

CRL/BRANDEIS : DESCRIPTION OF THE DIDEROT SYSTEM AS USED FOR MUC-5 Jim Cowie, Louise Guthrie, Wang Jin, Rong Wang, Takahiro Waka o Computing Research Laboratory, New Mexico State University Email: jcowie@nmsu.edu & James Pustejovsky, Scott Waterman Computer Science Department, Brandeis Universit y Email: jamesp@cs.brandeis.edu Abstract This report describes the major developments over the last six months in completing th e Diderot information extraction system for the MUC-5 evaluation . Diderot is an information extraction system built at CRL and Brandeis University over th e past two years. It was produced as part of our efforts in the Tipster project. The same overall system architecture has been used for English and Japanese and for the micro-electronics and join t venture domains. The past history of the system is discussed and the operation of its major components described . A summary of scores at the 24 month workshop is given and the performance of the system on the texts selected for the system walkthrough is discussed . INTRODUCTION The Computing Research Laboratory at New Mexico State University, in collaboration with Brandeis University, was one of four sites selected to develop systems to extract relevant information automat- ically from English and Japanese texts . The systems produced by the Tipster research groups hav e already been evaluated at 12 and 18 months into the project . The performance of the Diderot System has improved both for English and Japanese . The performance in Japanese, however, is still far ahea d of our English performance . The Tipster project is, without a doubt, the largest scale Applied Natural Language Processin g task yet undertaken anywhere in the world. The government data preparation effort involved th e selection and analysis of more than 5,000 individual texts. The results of this analysis have been used to develop and test the systems produced by each site. The software used to support this huma n extraction task, both for English and Japanese, was developed and supported by the CRL under a separate subcontract . Because of the emphasis on different languages and different subject areas the research has focused on the development of general purpose, re-usable techniques. The CRL/Brandeis group have imple- mented statistical methods for focusing on the relevant parts of texts, programs which recognize an d mark names of people, places and organizations and also dates. The actual analysis of the critical parts of the texts is carried out by a parser controlled by lexical structures for the `key ' words in the text . To extend the system's coverage of English and Japanese some of the content of these lexical structure s was derived from machine readable dictionaries. These were then enhanced with information extracted from corpora. 161 The system has already been evaluated in the 4th Message Understanding Conference (MUC-4 ) where it was required to extract information from 200 texts on South American terrorism . Considering the very short development time allowed for this additional domain the system performed adequately. The system was …

Content truncated to 500 words