Embodied Conversational Agents: Social or Nonsocial?

Abstract

Since the advent of radio, we are accustomed to hearing disembodied human voices from devices. If system designers use an embodied conversational agent (ECA) as the interface to their system, they select either a human or a synthetic voice. One prior research study shows that pairing a synthetic face with a natural voice is problematic for users. Another study showed users perceive high-quality synthetic voices as nearly equal to professional voice talent. The incongruence between the verbal and nonverbal channels of an ECA has a negative effect on ECA users. This study will demonstrate that task type (low affect-infusion vs. High affect-infusion) has an unexplored impact on the social requirements of an ECA, and that cognitive effort and trust may provide greater perceived value for high affect-infusion tasks.