A Study on Conversational Content Recognition Method Using JapaneseWordNet for Robot-Assisted Therapies

Abstract

This paper proposes a conversational content recognition method for robot-assisted therapies. In the proposed method, nouns in the conversation are first extracted by morphological analysis techniques. As the next step, the dominant conception of them, which is called “synset” in this paper, are obtained by using a concept dictionary. And finally the conversational topic is determined considering the number and kinds of obtained synsets. In this paper, Japanese Word-Net was employed as concept dictionary, and evaluation experiments using daily conversation voice data recorded in the welfare facility were conducted. The obtained results by the proposed method were quite similar to those by human and indicated that the proposed method had enough possibility to recognize conversation among some persons. This paper describes the detail of the proposed method, experimental results and also does some problems about the proposed method.