Embodied Interaction in Language Learning: A Study of Mobile, VR, and Humanoid Voicebots for Learners with Social Anxiety

Social anxiety, marked by fear of being judged, often leads to avoidance of social and public situations. In language learners, it manifests as reluctance to speak due to fear of mistakes or appearing incompetent, ultimately limiting participation and hindering language acquisition. Large Language Model agents, such as ChatGPT, Gemini, and Deepseek, are now widespread and recognized across age groups. In language learning, they show strong performance in translation, grammar explanation, and text correction, offering detailed feedback. The emergence of voice-enabled chatbots further enhances their potential by enabling spoken interactions that simulate real conversational practice. Additionally, through embodiment, voicebots can be integrated into other tools, such as humanoid robots or extended reality avatars. However user experience and acceptance of these new interaction modalities remain underexplored. This study investigates the language learning experience of 86 subjects with social anxiety disorder in a between-subjects experimental design, comparing three interaction modalities: a mobile voicebot, a VR-based avatar, and a humanoid robot. The results showed no statistically significant differences between groups in user experience, however participants who interacted with the humanoid robot reported a greater sense of ease and comfort, likely due to its human-like appearance and natural conversational style, although the need for physical interaction to initiate dialogue was perceived as a limitation to fluency. Users of the mobile app expressed higher confidence in their ability to achieve better language outcomes, while those engaging with the VR platform highlighted its potential for enhanced contextual immersion.