Author: Noriko Tomuro
Journal-ref: Accepted for publication at the workshop on Lexical Semantics in Context: Corpus, Inference and Discourse at the10th European Summer School in Logic, Language and Information (ESSLLI-98).
This paper describes a semi-automatic method of inducing underspecified semantic classes from WordNet verbs and nouns. An underspecified semantic class is an abstract semantic class which represents systematic polysemy: a set of word senses that are related in systematic and predictable ways. The method first applies a simple statistical technique to extract sense cooccurrence from the ambiguous words, and creates a type dependency graph after a manual filtering step. Then, the underspecified classes are automatically induced by partitioning the ambiguous senses according to the nodes covered in the graph. We discuss the advantages and difficulties of our method by comparing our results and those in CORELEX (Buitelaar, 1997, 1998).
Paper: Full paper (postscript 200k)