HMM-based semantic analysis for the ESST and MEDIA tasks

A stochastic component for semantic analysis has been applied to an appointment scheduling task in English (ESST) and a hotel room reservation task in French (MEDIA). Realized as an ergodic HMM using Viterbi decoding, the parser outputs the most likely semantic representation given a transcribed utterance as input. The semantic sequences used for training and testing the parser have been derived from the semantic representations of both spoken language dialogue corpora. The HMM parameters have been estimated given the word sequences along with their semantic representation. The performance of the parser has been determined for both tasks.