| Abstract |
TACITUS is a system for interpreting natural language texts that has been under development since 1985. It has a preprocessor and postprocessor currently tailored to the MUC-3 application. It performs a syntactic analysis of the sentences in the text, using a fairly complete grammar of English, producing a logical form in first-order predicate calculus. Pragmatics problems are solved by abductive inference in a pragmatics, or interpretation, component. The original purpose of TACITUS was to aid us in investigating the problems of inferencing in natural language. For that reason, the system employed a straight-line modularization, with syntactic analysis being done by the already-developed DIALOGIC parser and grammar; only the correct parse was chosen and passed on the inferencing component. With the discovery of the abduction framework in 1987, we realized that the proper way to deal with syntax-pragmatics interactions was in a unified abductive framework. However, the overhead in implementing such an approach at the level of coverage that the DIALOGIC system already provided would have been enormous, so that effort was not pursued, and we continued to focus on pragmatics problem. When we began to participate in the MUC-2 and MUC-3 evaluations, we could no longer chose manually which syntactic analysis to process, so we began to invest more effort in the implementation of heuristics for choosing the right parse. We do not view this as the ideal way of handling syntax-pragmatics interactions, but, on the other hand, it has forced us into the development of these heuristics to a point of remarkable success, as an analysis of our results in the latest evaluation demonstrate. We developed a preprocessor for MUC-2 and modified it for MUC-3. Our relevance filter was developed for MUC-3, as was our current template-generation component. |
