Optimization of Non-Verbal Information for English Conversation Agents Using Interactive Evolutionary Computation

Abstract

As English becomes increasingly important in our globalized world, current agent-based con-(breakpoint)versation practice environments often fail to sustain learner motivation, in part due to the agents’ inability to adapt their behavior to learners preferences. This study proposes optimizing a conversational agent’s non-verbal cues, such as nodding and voice characteristics, through interactive evolutionary computation to enhance users’ motivation for learning English conversation. Participants engaged in role-play scenarios across eight different scenarios, including libraries and cafes dialogue situations. Then, they evaluated the agent after each interaction, with the feedback used to iteratively optimize the agent’s non-verbal expres-(breakpoint)sions. As a result, approximately 90 % of participants reported increased willingness to interact with the optimized agent, suggesting personalizing non-verbal behavior hold the potential to significantly improve language learners’ motivation.