Construction and Optimization of a Question and Answer Database for a Real-environment Speech-oriented Guidance System

In this paper, we describe a question and answer database (QADB), which is useful for constructing a spoken dialog system. For building a real-environment speech-oriented dialog system, real user由ta釘e required. In order to col1ect these data, we have been employing a spoken guidance sytem since five years. A database of trainscriptions and appropriate responses, etc. has been constructed for 2 ye訂s of the collected utterances. Furthermore we introduce a QADB optimization method to exclude inappropriate data. Using this method, response accuracy improves by 1 .6% to 75.7% absolute and by 2.9% to 59.8% about adults and children, respectively.