How to Get Preferred Readings in Natural Language Analysis

摘要

This paper describes a parsing program called KGW+p which is designed for integrating various sorts of knowledge to get most preferred structural descriptions of sentences. The system accepts not only a set of rnles specifying constraints which any descriptions of sentences should satisfy, but also preferential rules which are utilized in selecting most preferred descriptions among possible ones. Daring the pusing process, the preferetial rules are utilized to select feasible parsing paths. Furthermore, KGW+p is complete in the sense that it can generate all possible structural descriptions of sentences if required. The descriptions are generated in a preferential order.