Construction of Dialogue Corpus with Personality Information and its Application to Dialogue System

Abstract

We constructed a dialogue corpus with personality information as a basis for analyzing the relationship between the speaker’s personality traits and utterance features and realizing a dialogue system with personality. In order to avoid the influence of different dialogue partners, all the collected dialogues are with the same chatbot. We constructed a dialogue corpus with personality traits information by asking the subjects to answer a series of questionnaires about their Big Five, communication skills, social skills, and so on after the dialogue. We manually assigned JAIST dialogue act tags to all utterances in the corpus. As a first step in applying this corpus to the implementation of a dialogue system, we constructed an Encoder-Decoder model for generating responses to chat dialogues using this corpus as training data. We report the results of experiments on the reproducibility of utterance features for specific personalities.