FEEDBACK QUALITY OF CHATGPT IN REVISING FRENCH TEXTS

DOI: 10.18173/2354-1075.2025-0034

Abstract

This study investigates the effectiveness of ChatGPT's feedback in revising French texts written by A2+ level learners. ChatGPT's feedback was divided into three categories: error detection, error analysis, and correction suggestion. A mixedmethods analysis of 20 learner texts showed that the AI's rate of inaccurate error detection was 13%, bad error analysis was 18%, and bad correction suggestion was 16%. Over half of the incorrect feedback was overcorrection, while 29% of the flawed error analysis was due to ChatGPT's lack of meta-linguistic knowledge to correctly explain the nature of the error. While most studies on the application of ChatGPT in foreign language teaching use questionnaires as the corpus, this research uses authentic texts, providing a more nuanced insight into the limitations of this tool. The study recommends that teachers, when applying ChatGPT in writing instruction, create prompts that clearly specify the language level of the text and the target audience. In cases of ineffective feedback, students should provide additional instructions to ChatGPT by clarifying their intended meaning or specifying the desired language norm.