Evaluating the Translation of Egyptian Colloquial Arabic Body Idioms into English: A Comparative Analysis of Four AI Tools

Artificial intelligence (AI) is increasingly being used in translation due to its speed and accessibility. However, idioms, particularly culturally specific ones like body idioms in Egyptian Colloquial Arabic (ECA), pose significant challenges due to their non-compositional and context-dependent nature. This study assessed the effectiveness of four AI tools—ChatGPT, Claude, Gemini, and DeepSeek—in translating ECA body idioms into English. 14 idioms were selected, and 12 expert translators were asked to assess AI-generated translations based on accuracy of meaning, naturalness, and cultural appropriateness using a Likert scale ranging from 0 (worst) to 5 (best). The results reveal notable variability among the tools. Gemini and ChatGPT showed nearly identical performance, scoring 3.53 and 3.52, respectively. DeepSeek performed moderately (3.15), while Claude scored lowest (2.60). Common issues included overly literal translations, reversal of meaning, tone misinterpretation, and outdated phrasing. Despite some successes, all tools struggled to consistently capture the figurative essence and cultural depth of the idioms. The study highlights the persistent difficulty of translating idioms with AI, particularly between culturally distant language varieties like ECA and English. While AI shows potential, it is not yet reliable for independent idiom translation. Human expertise remains essential for ensuring accuracy, naturalness, and cultural appropriateness. The results advocate for a hybrid approach combining AI capabilities with human oversight and emphasize the need for further research into AI’s handling of figurative language and cultural nuances.