Dialogue Act Classification in a Spoken Dialogue System

Abstract

A contribution to the understanding module in a spoken dialogue system is presented in this work. The task consists of answering telephone queries about timetables, prices and services for long distance trains in Spanish. In this system the representation of the meaning of an utterance is accomplished by means of frames, which represent the type of information of the user turn, and cases, which provide the information given in the sentence. The input of the understanding module is the output of the speech recognizer and its output is used by the dialogue manager. We focus on the classification process of the dialogue user turn with respect to the second level, i.e., the identification of the type or types of frames given in the utterance and on the effect of the spontaneous speech recognition errors in the classification accuracy. As classifiers for the user turns we employ multilayer perceptrons, in order to use specific understanding models for each type of frame.