Abstract While large language models (LLMs) such as ChatGPT are increasingly integrated into language education, little is known about their use for real-time spoken interaction in Chinese as a Foreign Language (CFL), particularly in formal K–12 instructional settings. This study investigates how 28 high school CFL learners in Austria engaged in tightly scripted question–answer interactions with ChatGPT’s Voice Mode during a classroom-based oral practice task. Data includes 28 audio recordings, six video recordings, and 26 written reflections collected immediately after the task. The analysis draws on a conversation-analytic perspective, focusing on task initiation, conversation openings, repair, and the role of teacher scaffolding, and is complemented by a thematic analysis of learner reflections. Findings show that ChatGPT (GPT-4) can be successfully constrained to deliver predictable, beginner-level interactions in Mandarin, offering learners a safe space to practice formulaic expressions and basic turn-taking routines, however, teacher support during task initiation and breakdowns was crucial to sustaining the interaction. Learners’ reflections revealed ambivalence: while many described the experience as motivating, novel, and low-anxiety, others expressed concerns regarding robotic prosody, occasional glitches, and the absence of interactional features typical of human interlocutors. We conclude that ChatGPT’s Voice Mode holds promise as a complementary tool for oral practice in beginner CFL classrooms, provided that its use is carefully mediated by teachers and supported by further AI development to enable seamless code-switching and multilingual feedback.
Renner et al. (Fri,) studied this question.