| Sumario: | This study aims to evaluate the ability of three emerging Large Language Models (LLMs) — ChatGPT, Gemini, and Jais — that support the Arabic language and are widely accessible. The focus was on their performance in drafting formal Arabic email correspondence for professional and administrative purposes. The study used a descriptive-analytical methodology. It analyzed the outputs of the models across six specific purposes: leave requests, transfer requests, resignation letters, administrative assignments, instruction delivery, and recommendation requests. A unified prompt was designed to ensure variation in the sender’s and recipient’s gender and job position. The models’ outputs were assessed based on clear structural and linguistic criteria, using both quantitative and qualitative analysis. The results showed that Gemini achieved the highest overall score at 88.3%, characterized by offering multiple response options and detailed explanations. Jais ranked second with 81.3%, demonstrating stronger cultural alignment, especially in opening and closing greetings. ChatGPT scored 80.4% but achieved the highest linguistic accuracy rate among the three models. Despite these strengths, all three models required human review and intervention to ensure full quality, clarity, and cultural appropriateness. The study concluded that while LLMs can assist in drafting Arabic emails, they cannot yet fully replace human oversight. It recommended expanding research efforts on Arabic email writing standards. More descriptive and normative studies are needed. The study also highlighted the importance of incorporating cultural dimensions into Arabic communication frameworks. Developing culturally sensitive and linguistically accurate models will improve their usability, relevance, and overall effectiveness in real-world Arabic contexts.
|
|---|