Beyond Lexical Equivalence: A Context-Sensitive Translation Model for Evaluating ChatGPT English Translation of Culturally Loaded Arabic Expressions

Abstract

Recent advances in artificial intelligence (AI) have substantially improved machine translation.  However, the translation of culturally loaded expressions remains challenging in that such expressions require semantic, pragmatic and cultural interpretation beyond lexical equivalence. Existing research has primarily evaluated AI translation quality by comparing AI outputs with human translations across specific genres. Nevertheless, limited attention has been given to developing practical frameworks for assessing and improving AI translation of culturally embedded Arabic expressions. This study investigates the ability of ChatGPT to translate culturally loaded Arabic expressions into English and proposes a Context-Sensitive Translation Model (CSTM) for evaluating and improving AI-assisted translation. Grounded in translation equivalence theory and principles of culture-specific translation, the study adopts a qualitative descriptive design. A purposive sample of 16 Arabic cultural terms which represent eight cultural categories was translated using ChatGPT and then analyzed through three evaluative dimensions, namely, semantic, pragmatic and cultural equivalence. The findings obtained reveal that ChatGPT consistently produces fluent and semantically acceptable translations. However, it frequently fails to preserve the deeper cultural, legal, religious and social meanings embedded in the source expressions. Although ChatGPT successfully conveyed the general lexical meaning of the selected terms, it could only partially maintain their pragmatic and cultural significance. The study demonstrated that lexical accuracy alone is insufficient for culturally appropriate translation. The study proposes CSTM as a practical framework for evaluating AI-generated translations and recommends integrating semantic, pragmatic and cultural considerations into AI translation systems while maintaining human involvement in translating culturally sensitive texts.