CORPUS-BASED COMPARISON OF CHATGPT AND STUDENT WRITING ON SPEAKING PROMPTS IN VIETNAMESE EFL CONTEXTS

Abstract

This study examines written texts produced by Vietnamese EFL students and ChatGPT in response to ten common speaking prompts. Although initially designed for oral practice, the prompts were adapted into short writing tasks, ensuring comparability with ChatGPT’s written outputs. A mixed-methods approach combined quantitative corpus profiling (type–token ratio, collocational range, discourse markers) with qualitative coding of syntax, lexis, and pragmatics. The analysis revealed that ChatGPT texts displayed greater lexical diversity (TTR 0.70 vs. 0.63), more natural collocations, and a wider range of cohesive devices. In contrast, student texts relied heavily on high-frequency verbs and basic connectors. Pragmatic contrasts also emerged: ChatGPT often employed hedges and polite expressions, while student writing tended to be abrupt or overly direct. These findings suggest that ChatGPT can enrich learners’ exposure to varied vocabulary, collocational accuracy, and pragmatic awareness. However, its use should remain pedagogically guided to avoid over-reliance. The study underscores the importance of teacher mediation and aligns its findings with second language acquisition (SLA) perspectives on input, noticing, and authenticity.