The Impact of LLMs Feedback on L2 Writing Development

This paper empirically investigates the impact of feedback from Large Language Models (LLMs) on the development of writing proficiency among university-level English learners. Employing a quasi-experimental pre-test/post-test design, the study assigned 52 non-English major freshmen to either an experimental group, which received LLM-generated feedback, or a control group, which received traditional teacher feedback, over a 15-week writing intervention. Writing samples were analyzed quantitatively using the core metrics of Complexity, Accuracy, and Fluency (CAF). The results indicate that the experimental group achieved significantly greater gains in writing accuracy compared to the control group. However, the differences in improvement regarding complexity and fluency between the two groups were not statistically significant. The study concludes that LLM feedback demonstrates a distinct advantage in enhancing the accuracy of linguistic forms, though its effectiveness in promoting linguistic complexity and fluency is less clear within the current model of application. These findings provide concrete empirical support for the use of LLMs in Automated Writing Evaluation (AWE) and offer new perspectives on how L2 writing instruction can effectively harness this technology to realize innovative human-computer collaboration.