A Study of Emotion Regulation and Intervention in Second Language Learning at Vocational Undergraduate Level in the Context of Multimodel Large Language Model

The emergence of multimodal large language models (MMLLMs)—such as GPT-4, Gemini, and Sora—has introduced a paradigm shift in second language (L2) education, enabling not only cognitive enhancement but also affective support through real-time, multimodal interaction. However, their potential for scaffolding learner emotion regulation remains underexamined, particularly within vocational undergraduate contexts where students often face heightened anxiety, low academic self-concept, and limited emotional self-regulation capacity. This study investigates the emotional characteristics of vocational L2 learners and examines how MMLLMs can be pedagogically leveraged to foster emotionally responsive, context-aware, and professionally relevant instruction. Drawing on a mixed-methods approach involving classroom observation and semi-structured interviews, the findings suggest that MMLLMs, when integrated with targeted strategies—such as emotion-aware adaptation, multimodal content delivery, personalized feedback, and vocational task simulation—significantly enhance both emotional resilience and communicative competence. The study provides theoretical insights and practical implications for developing affect-sensitive, AI-assisted L2 pedagogy tailored to the needs of vocational learners.