Using a Shared Representation to Generate Action and Social Language for a Virtual Dialogue Environment

This paper describes a conversational system POLLy (POliteness in Language Learning) which uses a common planning representation to generate actions to be performed by embodied agents in a virtual environment, and to generate spoken utterances for dialogues about the steps involved in completing the task. In order to generate socially appropriate dialogue, Brown and Levinson’s theory of politeness is used to constrain the dialogue generation process. We report the results of a cross-cultural user experiment to test differences between user perceptions of the role of the social variables of power and social distance on the appropriate linguistic form of an utterance. Our results suggest that, contrary to Brown and Levinson’s theory, different users have different perceptions of these variables, and weight them differently in social interaction.