Potential Use of ChatGPT-4 for Translating the Emirati Dialect Into English

For translators and conventional machine systems, Arabic dialects pose translation challenges including scarce linguistic resources and the complexity of nonstandard Arabic. However, large language models (LLMs) are able to understand contextual meaning and showcase promising translation capabilities. This study investigates ChatGPT-4’s understanding and translations of five excerpts, including 39 lexical items, from the Emirati dialect into English. The data were analyzed qualitatively for clarity, fluency, and structure. Additionally, four bilingual raters quantitatively evaluated the translated lexical items according to the context. Qualitative analysis revealed that ChatGPT-4 could understand the Emirati dialect’s linguistic, cultural, and semantic intricacies and successfully translate them into English. Quantitative evaluation, based on accuracy of meaning, word choice, naturalness, syntactic harmony, and clarity, demonstrated a high level of agreement among the raters. However, the source of the data, which was an online forum, might have impacted the results, as ChatGPT could have been trained on similar data from such forums, potentially influencing its translation outcomes. This study contributes to the discussion of LLMs’ viability in translation practice, with implications for translators, translation trainers, and tool makers.