Patterns of interaction in LLM based Conversation Practice: A Discourse Analysis

Abstract

This study analyzed the interactional dynamics between EFL learners and a Large Language Models (ChatGPT-4) in task-based contexts applying a mixed-method sequential explanatory design. Database consisted of human-LLM conversation transcripts complemented by semi-structured interviews with EFL learners. Sixty intermediate learners of English (IEL) at Qassim University participated in the eight conversation tasks over a period of ten weeks. Results showed statistically significant gains for the learners in discourse coherence, pragmatic subscales (indirect speech act production, request modification, and topic management), and meaning negotiation strategies across the intervention. Qualitative analysis confirmed that learners viewed ChatGPT-4 as an authentic scaffolding tool in foreign language conversation. The study has pedagogical implications encouraging the use of LLMs as additional conversational partners in EFL speaking instruction.