NOTICING INTERLINGUAL LEXICAL ERRORS IN TURKISH EFL WRITING: A COMPARISON OF LEARNER, TEACHER, AND CHATGPT ERROR FLAGGING AND FEEDBACK TYPES

Abstract

Written corrective feedback (WCF) from teachers and, more recently, from artificial intelligence (AI) tools has been widely studied; however, little attention has been paid to comparing how these sources address interlingual lexical errors. This study primarily aims to examine how the learners, the teacher, and AI feedback generated by ChatGPT differ in noticing and addressing interlingual lexical errors in EFL university writing. The present study was carried out at a state university where the English preparatory program introduces systematic writing instruction, progressing from paragraph writing (B1) to essay writing (early B2). Adopting a descriptive case study design, the study analysed 18 opinion texts produced by 12 B1-level students enrolled in an English preparatory programme at a Turkish state university. Six students each produced one opinion paragraph, while six different students each produced two opinion essays. A total of 81 error instances, identified through retrospective learner self-flagging, teacher feedback, and AI feedback, were classified according to James’s (1998) taxonomy, and the feedback types were coded following Ellis’s (2009) typology. The results showed that semantic errors constituted the majority of the instances (80.2%). The AI tool noticed the largest share of the errors (85.2%), followed by the learners (45.7%) and the teacher (28.4%), while all three sources converged on only 11 instances (13.6%). The teacher mostly provided indirect feedback with error codes, whereas the AI tool almost always gave direct feedback with an explanation. In this sense, it can be said that error identification works best as a complementary process, in which learner awareness, teacher judgement, and AI assistance each cover a different part of the errors.