Virtual agents aligning to their users. Lexical alignment in human–agent-interaction and its psychological effects

Abstract

This study investigated the impact of agent-sided lexical (non-)alignment on users’ psycho-physiological responses and their evaluation. In an object-naming-matching game played in two game rounds, all participants interacted with a virtual agent who lexically aligns with user/participant language choices during the first half of the interaction, but not during the second half, and vice-versa. Participants were asked to evaluate the agent after each round with regard to perceived likability, competence, and autonomy. Measures for participants’ current positive and negative affect were administered before the study start and after each round. Results indicated that the temporal occurrence of (non-)alignment had an effect on the evaluation of the agent regarding its competence. The agent was rated more positively if it changed its behavior from non-alignment to alignment and participants rated the agent more negatively when it stopped aligning during the course of the game. This demonstrates the importance of the dynamics in alignment processes. User-sided alignment to the agent’s word choices decreased over the course of the interaction. Concerning participant’s self-reported positive and negative affect and physiological arousal, analyses showed no effect of the experimental condition but rather a general decrease over the course of the interaction. This finding seems to reflect fatigue due to the long experimental procedure. We discuss methodological improvements regarding psycho-physiology for future work.