Comparing fluency-oriented conversational modeling and form-oriented pedagogical scaffolding in VR-supported L2 speaking instruction
Lin Chih-Chung, Chi-Kay Tang, Tzu-Hsuan Lin, Yi-Chen Hsu, Hui‐Chin Yeh · Innovation in Language Learning and Teaching · 2026
Recent advances in virtual reality and artificial intelligence have created new possibilities for English speaking instruction. However, the pedagogical value of AI-mediated innovation depends not only on emerging technologies, but also on how learner interaction, feedback, and support are configured. This quasi-experimental study examined two AI-mediated instructional orientations in virtual reality-supported workplace English speaking: a fluency-oriented conversational modeling condition and a form-oriented pedagogical scaffolding condition. Rather than comparing separate technological models, the study focused on how different pedagogical configurations of AI-mediated interaction shaped L2 speaking development. Sixty-one university EFL learners participated in a nine-week intervention involving immersive workplace communication tasks. Speaking performance was assessed in terms of fluency, accuracy, complexity, pragmatic appropriateness, and task completion, while speaking anxiety was measured using an EFL speaking anxiety scale. Results from MANCOVA and follow-up ANCOVA showed that the fluency-oriented conversational modeling condition produced significantly greater improvements in linguistic accuracy, suggesting the value of target-like modeling and reformulation. In contrast, the form-oriented pedagogical scaffolding condition led to stronger task completion and pragmatic appropriateness, as well as greater reduction in speaking anxiety. These findings indicate that AI-mediated language learning innovation should not be understood as a uniform technological intervention. Rather, different AI interactional architectures may support different dimensions of L2 speaking development. The study contributes to innovation-oriented language teaching research by showing how conversational modeling and pedagogical scaffolding can strategically support learner-centered speaking instruction in immersive environments.