A comparative evaluation of machine translation vs. human translation for legal texts: A case study of translation between English and Arabic

Abstract

This study examines the comparative accuracy and fluency of Neural Machine Translations (NMTs) and Language Model-based translations (LMBTs), represented by ChatGPT and Google Translate (GT), in legal texts translations. Texts from Farahaty's "Arabic-English-Arabic Legal Translation", sourced from primary texts cited in the book and translated by scholars such as Hatim, Shunnaq, Buckley, and Farahaty were used as benchmarks for human translation (HT). Sixteen diverse texts encompassing various legal discourse subgenres were selected for analysis, with all Arabic in-text examples transliterated using the Library of Congress (LOC) system. Qualitative analysis was conducted to assess the extent to which NMTs and LLMs match HT in accuracy and fluency. The study also investigated the similarities and differences between ChatGPT and GT in their translation outputs. Findings highlight HT's superiority in producing precise, stylistically appropriate translations, compared to the challenges faced by NMTs and LLMs in capturing legal terminology and subtle linguistic nuances. Despite variations, both ChatGPT and GT demonstrate efficiency and context sensitivity, suggesting their potential as valuable tools when coupled with human post-editing. The study concludes by advocating for a hybrid approach that leverages the strengths of automated translation systems and human expertise to enhance cross-linguistic legal communication.