Publikation
Specification and Underspecification in Lexical Semantic Processing for Information Extraction
Paul Buitelaar
In: Bad Teinach Workshop on Underspecification, May. Bad Teinach Workshop on Underspecification, 1998.
Zusammenfassung
The work reported here is part of a research effort to extend an existing information extraction system (SMES) with more sophisticated lexical semantic capabilities, including the use of underspecified representations, in order to make the important step of domain modelling, in porting the system to different application areas, more systematic and generally to increase the accuracy of the system. In general, information extraction (IE) is a rather scaled down version of natural language understanding, in which there is little room for the processing of deep (i.e. lexical semantic) knowledge. However, in obvious ways, underspecified lexical semantic representations can be seen as shallow lexical meanings that have a similar status as chunks in syntactic processing. Therefore, the use of such representations in IE seems an attractive option to extend the semantic capabilities of these systems, although it is not entirely clear where underspecification will be useful and where not. It is the topic of this paper to investigate this into some respect.