What are effective classroom dialogue patterns? A study on classroom dialogue patterns of pre-service and expert teachers based on large language models

Classrooms are pivotal arenas in which teachers transmit knowledge and learners cultivate understanding. Drawing on social-constructivist and dialogic-teaching theories, a substantial body of research positions classroom dialogue as the principal mechanism for the coconstruction and transformation of knowledge, as well as the critical nexus linking instruction and learning. Capitalising on authentic dialogue data from junior-secondary mathematics classrooms, we leveraged a fine-tuned large language model (LLM) to automate the encoding of classroom discourse. Descriptive statistics and lag-sequential analyses were then employed to juxtapose the interactional profiles of pre-service and expert teachers. Based on this, we derived an integrated dialogue pattern that amplifies the strengths of both cohorts and, through follow-up teacher interviews, obtained preliminary evidence of its capacity to enhance the professional growth of teachers in the pre-service and the quality of dialogue. Our findings indicate that the fine-tuned LLM achieves robust classification accuracy. Regardless of the level of expertise, prior knowledge activation dominates classroom exchanges; nevertheless, expert teachers orchestrate dialogue more strategically, eliciting higher-order thinking, and construct discourse trajectories that are demonstrably more intricate and adaptive.