Structured Vectorial Semantics for Broad-coverage Parsing

This paper proposes a novel structured vectorial semantic framework, incorporating cognitively-motivated ‘gist’ semantics as selectional restrictions in interactive vectorial-semantic parsing. Applying vectorial semantic techniques (in particular, relational clustering over dependent headwords) in this extended framework has a predictable positive impact on parsing accuracy. The vectorial representation is conducive to a fast implementation of necessary parser operations, despite the fact that vectorial semantic techniques have dense relations to propagate through the parse chart.