Building a Large Lexical Databank Which Provides Deep Semantics

This paper reports on the design of a lexical database for English which is currently under construction ("FrameNet-2" ), and describes the kinds of linguistic facts that the database is intended to make available, for both human and computer consumers. Building on a recently completed pilot study ("FrameNet-1" ), it is centered on the nature of the relation between lexical meanings and the conceptual structures which underlie them (semantic frames). The database will show the semantic and syntactic combinatorial possibilities (based on frame membership) of the lexical items it includes, as these are documented through grammatical and semantic annotations of sentences extracted from a large corpus of contemporary written English. The notions of profiling within a frame, frame inheritance including multiple inheritance, frame blending, and frame composition will be explained and illustrated, and the manner of storing information about them in the database will be outlined. The building of the database, with its necessary labor-intensive manual component, will be explained.