Translating food idioms across the English-Arabic language divide: A comparative study of human translators, ChatGPT, and Gemini

Abstract

Idiomatic expressions, especially those rooted in the universal domain of food, represent one of the most culture-embedded and translation-difficult forms of language. The current study aims to examine the translation of twenty English food idioms into Arabic and twenty Arabic food idioms into English by human professional translators, as well as two of the most prominent large language models (LLMs): ChatGPT (GPT-4o) and Google Gemini Advanced. A mixed-methods approach is used to evaluate the performance of all sixty translations of each idiom on four criteria: semantic equivalence (SE)1., pragmatic appropriateness (PA), cultural adaptation (CA), and natural fluency (NF), with a five-point scale for each. A composite translation quality score is derived for each translator/agent. The results show that human professional translators outperform the two LLMs on all criteria with a mean translation quality score of 5.00, compared to 3.83 for ChatGPT and 3.60 for Gemini. The results also show the greatest performance gap between human translators and the two LLMs is on cultural adaptation. The results of the current study have important implications for the pedagogy of translation, translation practice, and the development of culturally sensitive AI translation tools.