Deriving Verb-Meaning Clusters From Syntactic Structure

This paper presents a methodology for using the argument structure of sentences, as encoded by the PropBank project, to develop clusters of verbs with similar meaning and usage. These clusters can be favorably compared to the classes developed by the VerbNet project. The most interesting cases are those where the clustering methodology suggests new members for VerbNet classes which will then be associated with the semantic predicates for that class.