Parsing strategies for a spoken language processing system

Abstract

This paper describes the parsing strategies developed for Iris, an NLP system integrated with a speech recogniser and a dialogue manager. We will describe the syntactic notation which accounts for phenomena such as free sentence order in Spanish and in spoken language, as well as parsing strategies for sentence fragmentation into autonomous blocks. The grammar rule notation includes a novel typified semantic nuclei specification (CTAC) which simplifies the interface between the parser and the dialogue manager. In addition, the parsing module includes several strategies for the detection and correction (whenever possible) of the deficiencies originated at the speech recognition stage (over-recognition, under-recognition, close-recognition and mis-recognition). Finally, we present some provisional results obtained with the first prototype designed with the ideas described in this paper.