Classification of Utterances Using Temporal Relation in Cloud-based Dialogue System

Abstract

In recent years, research on dialogue system is prosperous. we developed spoken dialogue system as the previous research. One of the challenges of the system that we developed in previous research is that temporal relation cannot be considered. In order to solve this problem, we consider a system which automatically judges whether the temporal relation is contained in robot response sentences. We used two methods as the classification method. The first is simple classification by word pattern matching. The second is classification using MLP ( Multi-Layer Perceptron. ) As a result, the classification method using MLP was more useful than the word pattern matching.