A politeness control method for conversational agents considering social relationships with users

In this paper, we propose an autonomous linguistic consideration (politeness) control method for conversational agents. As an approach to realize the proposed method, we examined input information based on politeness theory in sociolinguistics that is considered to be effective for agents to control politeness. Specifically, we adapted a questionnaire study to evaluate the politeness and comfortableness of utterances in different situations of requests. The participants evaluated stimulus utterances in terms of three social factors on the politeness theory: social distance/social status difference/conversation scene. Based on the dataset collected in this questionnaire study, a regression model was constructed to predict the politeness and comfortableness rated by the participants of the experiment, adapted three social factors and politeness of utterances as explanatory variables. As a result of this questionnaire study and multiple regression analysis, a valid model of politeness was obtained. The results contribute as a model to select appropriate politeness of utterance according to three social factors on the politeness theory with the user in the proposed method.