A comparative study of pragmatic performance of internal modification in continuation tasks of online reviews: Generative AI versus EFL learners

Recent advances in Artificial Intelligence (AI) and Large Language Models (LLMs) have underscored their potential in second language (L2) pragmatics instruction. This study conducts a comparative analysis of 40 ChatGPT-generated and 40 Chinese EFL learner-produced online continuation reviews, systematically examining pragmatic internal modification across emotional valences. The findings reveal that: 1) In positive reviews, ChatGPT demonstrated superior diversity of lexical intensifiers and appropriate punctuation usage, whereas learners exhibited greater contextual sensitivity in modulating punctuation emphasis across different levels of emotion; 2) In negative reviews, ChatGPT displayed stronger command of genre-specific modality hedges and understaters, but lagged in flexibility in syntactic negating device; 3) Valence effects were stronger on ChatGPT's modification choices compared to EFL learners, suggesting GenAI's priority to predictability and safety and learners' emphasis on authenticity and adaptability. These findings not only enhance our understanding of LLM's pragmatic production capability with regard to internal modification but also propose a tripartite integration framework for ChatGPT in L2 pragmatics instruction: as a pragmatic input facilitator, automated feedback provider, and conversational tutor.