Strategies of Rendering Metaphor from Arabic into English: A Comparative Study of ChatGPT and Matecat

Abstract

This study examines the translation strategies employed in rendering Arabic metaphors from Naguib Mahfouz's Zuqāq al-Midaq into English, comparing the approaches of a human translator with those of two machine translation systems: ChatGPT-4 and Matecat. Drawing upon Nida's (1964) theory of formal and dynamic equivalence, the research analyzed 106 metaphors to identify predominant translation strategies. The findings reveal that the human translator consistently adopted dynamic equivalence, effectively re-creating the metaphorical meaning and cultural nuances for the target audience. In contrast, while ChatGPT-4 demonstrated a notably higher tendency towards dynamic equivalence compared to Matecat, both machine translation systems still frequently resorted to formal equivalence, resulting in more literal or less idiomatic renditions. This indicates that, despite advancements, machine translation, including advanced large language models (LLMs), continues to face significant challenges in accurately conveying the subtle complexities of figurative language. The study highlights the indispensable role of human translators in achieving nuanced and culturally sensitive metaphorical translations while also underscoring the potential of advanced AI tools to enhance the translation process when complemented by human expertise.