A Comparison of the Accuracy of Generative Artificial Intelligence Models in Taṣrīf and the Explanation of Wazan Meanings: A Study on Their Application in Arabic Morphology (Ṣarf)

Abstract

This study analyzes and compares the accuracy of two generative artificial intelligence models, ChatGPT and Deepseek, in producing taṣrīf (morphological conjugation) and explaining the meanings of wazan (morphological patterns) in Arabic morphology. Employing a descriptive-qualitative method with a content analysis approach, the research uses 30 fiʿl thulāthī mazīd (triliteral verbs with augmentation) representing all categories ṣaḥīḥ, maḥmūz, muḍāʿaf, muʿtal, miṯāl, ajwaf, nāqiṣ, and lafīf. Each verb was tested on ChatGPT 4.0 and Deepseek V3.1, and the outputs were verified against classical references, including al-Bināʾ, al-Maqṣūd, and al-Amthilah al-Taṣrīfiyyah. The findings show that both models produce fairly accurate results with distinctive tendencies: Deepseek aligns more closely with classical morphological rules, while ChatGPT provides more diverse and contextual explanations. Pedagogically, both have strong potential to support Arabic morphology learning through interactive and varied outputs, though teacher supervision remains essential to maintain adherence to classical standards.