User attitude towards an embodied conversational agent: Effects of the interaction mode

We describe how the interaction mode with an embodied conversational agent (ECA) affects the users’ perception of the agent and their behavior during interaction, and propose a method to recognize the social attitude of users towards the agent from their verbal behavior. A corpus of human–ECA dialogues was collected with a Wizard-of-Oz study in which the input mode of the user moves was varied (written vs. speech-based). After labeling the corpus, we evaluated the relationship between input mode and social attitude of users towards the agent. The results show that, by increasing naturalness of interaction, spoken input produces a warmer attitude of users and a richer language: this effect is more evident for users with a background in humanities. Recognition of signs of social attitude is needed for adapting the ECA's verbal and nonverbal behavior.