Development of Spoken Language Understanding System : Comparison of Syntax-driven and Keyword-driven Approach

Abstract

We are developing a spoken dialogue system that accepts speaker-independent continuous utterances and responds to them. Two approaches are adopted and compared. Syntax-driven approach first applies syntactic analysis to constrain the input and passes syntactically accepted sentence candidates to semantic analysis. Keyword-driven approach performs keyword spotting and generates a lattice of keyword candidates as the input to semantic analyzer. For syntactic analysis, we realize LR parsing based on A* search, using word-pair constraint as heuristics. For keyword spotting, heuristic language model is incorporated to constrain the input containing the keywords. Semantic analyzer is based on a semantic network and a semantic representation is constructed by traversing the paths of the network nodes that are activated as the corresponding words are recognized. Experimental results shows that syntax-driven approach is superior to keyword-driven approach that is expected to be robust.