Introducing Syntax Information in a Stochastically-Based Semantic Case Grammar Parser

Abstract

We study the impact of introducing syntax information into a stochastic component for natural language understanding that is based on a purely semantic case grammar formalism. The parser operates in an application for train travel information retrieval, the French ARISE (Automatic Railway Information Systems for Europe) task. This application supports the development of schedule inquiry services by telephone. The semantic case grammar has been chosen in order to enhance robustness facing spontaneous speech effects. However, this robustness is likely to turn into a drawback, if the semantic analysis ignores information that is propagated by syntactic relations. Introducing additional syntax information, whose complexity is well adapted to the size of the stochastic model, may disambiguate and therefore improve the decoding.