Prompt Writing as Metapragmatic Feedback-Seeking Practice in Generative AI-Mediated L2 Revision

Claudia Sangmi Yun^*, Hyekyung Yu^**

^*Visiting Professor, Department of English, College of Humanities and Social Science, Kangwon National University; 346 Jungang-ro, Samcheock-si, Kangwon-do, 25913, South Korea; cyun@kangwon.ac.kr
^**Instructor, Department of English Language and Literature, College of Humanities, Jeju National University; 102 Jejudaehak-ro, Jeju-si, Jeju-do, 63243, South Korea; hailey1122@jejunu.ac.kr

Abstract

As generative AI becomes increasingly integrated into L2 writing instruction, a central pedagogical challenge concerns not only students’ access to AI-generated feedback but also how they formulate requests for such feedback. This study examines the metapragmatic feedback-seeking practices evident in prompts written by 131 first-year Korean university EFL students during a four-hour classroom revision task using ChatGPT (GPT-3.5). The primary dataset comprised 655 student-generated prompts, with five prompts from each participant addressing content, coherence, organization, word choice, and sentence structure. A data-driven thematic analysis first identified three mutually exclusive pragmatic gap categories: ambiguous prompts (n = 436, 66.6%), overly broad prompts (n = 163, 24.9%), and criteria-lacking prompts (n = 56, 8.5%). Cohen’s kappa (κ = .85) was calculated for this three-category classification. A second interpretive analysis then examined how the omissions characteristic of these categories left different aspects of evaluative specification and contextual inference unresolved for the generative AI system. This distribution is interpreted analytically, not as evidence of students’ intentions. Based on the recurring omissions, this study derives the Goal–Target–Criteria (GTC) heuristic post hoc. It also offers the Reflective Question Loop (RQL) as an untested pedagogical proposal for future classroom implementation, rather than as an intervention evaluated in this study. The study contributes to sociolinguistic accounts of human-AI interaction by showing how prompt wording configures evaluative authority, inferential burden, and participant roles during L2 revision.

초록

Vol. 40 No. 4 (2022.12)

Journal Abbreviation j. Korea Saf. Manag. Sci.
Frequency Published four times annually in March, June, September, and December
Doi Prefix 10.14353/sjk.
Year of Launching 1993
Publisher The Sociolinguistic Society of Korea

Indexed/Tracked/Covered By

Online Submission

socioling.jams.or.kr

The Sociolinguistic
Society of Korea

socioling.com

Editorial Office Contact Information

- Tel: +81-10-3894-3164
- E-mail: schang@hoseo.edu

SITE MAP