Translating legal texts: A four-criteria evaluation of ChatGPT, DeepL, and Google Translate

Abstract

Translating legal texts—especially between linguistically and culturally distinct languages such as Arabic and English—poses significant challenges within Translation Studies. This study evaluates the performance of three machine translation tools: ChatGPT (OpenAI GPT-3.5, March 2024), Google Translate (web version, April 2024), and DeepL (free web version, April 2024), focusing on a corpus of Algerian Supreme Court judgments. Drawing on key theoretical frameworks including functionalism, equivalence and legal semiotics, we assess translation quality across four criteria: accuracy, clarity, fluency, and cultural appropriateness. Detailed documentation of tool versions, prompting strategies, and input constraints ensures transparency and supports reproducibility. Our findings indicate that while all tools exhibit limitations inherent to the complexity of legal language, ChatGPT shows particular strength in handling cultural nuances and producing more contextually appropriate translations. Nonetheless, human expertise remains vital for ensuring reliability in professional legal translation. Though the small dataset limits the generalizability of results, this study offers important insights for Translation Studies scholarship and lays groundwork for further research into the evolving role of AI in legal translation.