Generative AI Driven College English Multimodal Teaching Based on BERT Optimization: An Empirical Study on Interactive Participation and Learning Outcomes in Digital Twin Classrooms

The traditional college English teaching model has the problems of insufficient classroom interaction and limited personalized feedback, which makes it difficult to meet the personalized learning needs of students. To this end, this paper proposes a multimodal teaching framework based on BERToptimized generative AI and digital twin classrooms. By enhancing language comprehension ability through BERT, combining intelligent dialogue systems and digital twin technology, it optimizes classroom interaction and improves personalized learning support. This study adopts an experimental control design and conducts a comparative analysis of 200 college students. The results show that the number of questions in AI-enhanced classrooms increased by $35 \%$, the number of teacher-student dialogues increased by $50 \%$, and the proportion of group discussion time increased by $25 \%$. In terms of learning outcomes, the writing scores of students in the experimental group increased by 1.8 points (out of 10 points), oral fluency increased by $20 \%$, and reading comprehension accuracy increased by $15 \%$, all of which were significantly better than those in the control group. Generative AI combined with multimodal teaching and digital twin technology can effectively improve classroom interactivity and learning outcomes, provide a new direction for the development of intelligent education, and provide theoretical support for the application of AI in language teaching.