Lexical Resources for Natural Language Processing Systems and the Problem of Polysemy

In this paper I evaluate extant lexical acquisition techniques for NLP systems with respect to the problem of polysemy. I argue that available lexical resources are inadequate for creating computational lexica capable of addressing the generativity of language use. I suggest several desiderata for lexical resources, to increase their utility for lexical acquisition in the context of polysemy: fully parsed corpora, and abstract semantic tagging of both verbs and nouns. These desiderata derive from a worked example of how lexical acquisition might proceed specifically in the context of handling the phenomenon of logical metonymy. I propose that both effective solutions to the word sense disambiguation problem and the development of adequate lexical resources for tackling this problem on a large scale must be informed by linguistic analysis.