A Comparative Assessment of ChatGPT, DeepSeek and Human Translations of Cultural References in the Moroccan Novel For Bread Alone

Abstract

The translatability of cultural references (CRs) has been a major problem for human translators, let alone artificial intelligence (AI) tools. This study evaluates the effectiveness of AI vis-à-vis human translations of CRs. The dataset was mined from For Bread Alone, a unique Arabic literary autobiographical masterpiece by the Moroccan author Mohamed Choukri (2000) and translated into English (1983) by Paul Bowles. The AI English translations were generated by Chat GPT (GPT) and DeepSeek (DS), Western vs Eastern AI systems. 300 English renderings of 100 CRs were analyzed qualitatively and quantitatively. The AI translations were assessed using BLEU, METEOR and chrF metrics vis-à-vis the human translation (HT), then both AI translations and HT were evaluated using Pedersen’s (2011) taxonomy of strategies (with two added strategies specifically used for this study). BLEU produced the highest results, while METEOR and chrF scores were significantly lower, and GPT scores were higher than DS in comparison to HT. The analysis of the strategies indicates that both GPT and DS relied heavily on source language oriented strategies, while HT made more use of target language oriented ones. The findings highlight the tensions between linguistic precision and cultural adequacy in both human and AI translations.