A Robust Semantic Parser Designed for Spoken Dialog Sytems

This article describes a knowledge oriented semantic parser designed for spoken dialog systems. It merges some syntactic and semantic aspects into a single frame to get robust and efficient analyses. The knowledge is entirely described withing an XML formalism. The parser takes as input either a string of words (the 1-best word sequence) or a word-lattice coming from an ASR (Automatic Speech Recognition) system. The output is a semantic structure of concepts representing the meaning of the utterance. From an algorithmic point of view, the parser uses a chart principle implemented with the CLIPS system, a rule-based inference engine. The article presents some results obtained by performing an evaluation based on the MEDIA French corpus. The parser is distributed as an open-source tool.