Exploring student engagement with multimodal generative AI in task-based Chinese language learning

Research on generative artificial intelligence (GenAI) in language learning has largely focused on text-based writing support, with less attention to multimodal functions and their role in task-based contexts. Thus, this study investigated how 20 Japanese university students engaged with multimodal GenAI in a 14-week Chinese as a foreign language course. Data included pre- and post-semester surveys, learner reflections, and two video tasks designed around real-world communication. A mixed-methods analysis revealed that GenAI facilitated preparation and was perceived as lowering affective barriers to participation, contributing to measurable gains in vocabulary range, structural accuracy, and task achievement. However, improvements in fluency, tonal accuracy, and delivery confidence were limited. Learners valued efficiency and accessibility but often questioned the reliability and authenticity of AI outputs, with many continuing to rely on teachers and peers for reassurance. Learner engagement across affective, cognitive, and behavioral dimensions ranged from deep, evaluative use of AI resources to surface-level dependence on unedited outputs. The findings highlight both the potential and the limitations of AI-mediated engagement: while multimodal GenAI can scaffold short-term task completion, long-term development requires carefully designed tasks that sustain challenge, foster critical AI literacy, and build relational trust.