Assessing Perceived Politeness In A Virtual Agent’s Request To Join A Conversational Group

Abstract: Research around the implementations of human-like interactions during conversations with embodied conversational agents is currently at the forefront of Human-Computer Interaction as a field of science. In particular, conveying politeness through verbal and nonverbal behaviors is a significant challenge when modeling life-like virtual agents, as it underlies the conventional set of unspoken rules that govern human communication. In order to assess measures of perceived persuasiveness and politeness in human-agent interaction, in this study we compare six different politeness strategies - namely Baseline, Indirect, Asking, Proposing, Commanding and Pointing - to identify which combinations of verbal and nonverbal behaviors best persuade humans in joining a virtual conversational group while immersed in Virtual Reality (VR). 45 people were asked to walk towards the group with a specific scenario in mind, and answered questions on the perceived politeness of the agents inviting them. Results show that direct strategies are more persuasive than indirect ones, and the more imposing ones among them are perceived as more offensive, less friendly and inhibiting the interactant’s freedom of action. Additionally, strategies are less effective when participants feel more immersed in VR. We conclude that those giving the counterpart a higher freedom of action are most effective in conveying politeness, while those involving no behavior at all are least effective.