Evaluating the Performance of Artificial Intelligence Models in Identifying Arabic Verb Patterns: A Comparative Descriptive Study of Chatgpt and Gemini 10.35781/1637-000-173-007

Abstract

Abstract

This study aimed to evaluate the effectiveness of artificial intelligence models (ChatGPT and Gemini) in processing the Arabic morphological scale (al-mīzān al-ṣarfī) in verbs, as a key component of linguistic structure analysis. The study adopted a descriptive-analytical comparative approach supported by quantitative analysis to measure performance accuracy. The sample consisted of 16 verbs representing triliteral, quadriliteral, quinquiliteral, and sextuple patterns in both regular and irregular forms, in addition to morphological phenomena such as vowel change (iʿlāl) and assimilation (idghām).
The results indicated that the error rate was )43.75%( for ChatGPT compared to )6.25% (for Gemini, while the accuracy rates were )56.25% (and )93.75%( respectively, demonstrating a clear superiority of Gemini in morphological pattern identification. Both models showed better performance in standard patterns, whereas weaknesses appeared in irregular verbs and imperative forms. The study concluded that Gemini can be considered a more effective supportive tool in teaching Arabic morphology, while emphasizing the need for linguistic review of AI outputs and the development of Arabic-specific trained models to reduce errors and hallucinations and improve computational linguistic applications.
Keywords: Artificial Intelligence, Arabic Morphology, Verb Patterns, ChatGPT, Gemini, Arabic Language Teaching
.