Inducing criteria for lexicalization parts of speech using the Cyc KB

Tom O'Hara, Michael J. Witbrock, Bjørn Aldag, Stefano Bertolo, Nancy Salay, Jon Curtis, Kathy Panton

Abstract: We present an approach for learning part-of-speech distinctions by induction over the lexicon of the Cyc knowledge base. This produces good results (74.6%) using a decision tree that incorporates both semantic features and syntactic features. Accurate results (90.5%) are achieved for the special case of deciding whether lexical mappings should use count noun or mass noun headwords. Comparable results are also obtained using OpenCyc, the publicly available version of Cyc.