This paper describes the linguistic procedure of our speech dialogue system. The procedure is composed of two processes, syntactic analysis using a finite state network, and discourse analysis using a plan recognition model. The finite state network is compiled from regular grammar. The regular grammar is described in order to accept sentences with various styles, for example ellipsis and inversion. The regular grammar is automatically generated from the skeleton of the grammar. The discourse analysis module understands the utterance, generates the next question for users and also predicts words which will be in the next utterance. For an extension number guidance task, we obtained correct recognition results for 93% of input sentences without word prediction and for 98% if prediction results include proper words.

We carried out a one year field trial of a voice-activated automatic telephone exchange service at KDD Laboratories which has about 200 branch phones. This system has DSP-based continuous speech recognition hardware which can process incoming calls in real time using a vocabulary of 300 words. The recognition accuracy was found to be 92.5% for speech read from a written text under laboratory conditions independent of the speaker. In this paper, we describe the performance of the system obtained as a result of the field trial. Apart from recognition accuracy, there was about 20% error due to out-of-vocabulary input and incorrect detection of speech endpoints which had not been allowed for in the laboratory experiments. Also, we found that the recognition accuracy for actual speech was about 18% lower than for speech read from text even if there were no out-of-vocabulary words. In this paper, we examine error variations for individual data in order to try and pinpoint the cause of incorrect recognition. It was found from experiments on the collected data that the pause model used, filled pause grammar and differences of channel frequency response seriously affected recognition accuracy. With the help of simple techniques to overcome these problems, we finally obtained a recognition accuracy of 88.7% for real data.