The Efficiency of Generative Language Models (LLMs) in Detecting Grammatical Errors in Arabic as a Second Language Learners: Towards Designing a System Based on Frequent Error Analysis and Correction

Abstract

This study investigates the efficiency of large language models (LLMs) in detecting grammatical errors among learners of Arabic as a second language, with a particular focus on analyzing recurrent error patterns. It further seeks to evaluate the performance of state of the art models in identifying such errors and to shed light on the underlying causes behind the most frequent ones. Findings reveal that certain models, notably ChatGPT, Claude, and Deepseek, demonstrated superior performance in error detection when compared to Gemini and Grok. However, the overall accuracy of syntactic explanations remains limited, indicating gaps in deeper grammatical anlysis. The study also highlights the decisive influence of training data quality on model performance, underscoring the need for tailored linguistic resources to enhance reliability in processing Arabic texts.