Transparency and Building Lexical Dependency Graphs

Abstract

The authors are participants in a project, called "FrameNet" which is aimed at building a large computer lexicon of contemporary written English and making it accessible through the World Wide Web for both computational and lexicographic interests. In the process of designing this resource, we have to keep in mind how it can serve its intended applications. A prerequisite to most imaginable NLP applications is word sense disambiguation (WSD), the automatic process by which a word in a linguistic context can be (probabilistically) assigned its locally intended meaning. This paper will introduce FrameNet and will characterize a facility, based on its tools and data, that could in principle be directed to WSD efforts, and will suggest how both technical (engineering) and linguistic considerations must be called on to build it.