Driving Dialogue Dataset Collection and Analysis for a Transformer-Based Conversational System that Talks About Scenery

Abstract

In order to develop a Transformer-based spoken dialogue system for natural dialogue about the scenery like a passenger, we collected a dataset of dialogue between role of drivers and passengers who engaged in a dialogue by sharing scenery driving movies by online. The collected dataset was analyzed to determine the extent to which utterances that talked about scenery or involved self-disclosure occur. The results showed that basic chit-chat performance was most important, even for chit-chat dialogues that talks about scenery.