Human Versus Artificial Intelligence (AI): Challenges in the Legal Translation of English to Arabic Terminology at Yemeni Universities, Taiz

Abstract

This study aimed to compare the performance of human translators (HTsr) and AI (ChatGPT) based on a reference translation (RT) of 25 English legal terms (LTs), which were placed in meaningful sentences. Data was collected through a carefully designed research task, responded to by both AI and 12 female translators aged between 22 and 30, alumni from various Yemeni universities in Taiz. Analysis was conducted by tabulating the translations to organize the collected data for comparison. Findings revealed that the AI has produced three distinct categories of translation outputs: (i) correct, (ii) partially correct, and (iii) incorrect, avoiding missing translations and suggesting that AI operates within a more constrained framework. Although it produced accurate translations, especially with direct LTs, it struggled with terms that required a deep understanding of legal context, yielding literal translations. In contrast, HTrs yielded four types of translation outputs: correct, partially correct, incorrect, or nothing at all, named blank response. The possibility of producing ‘blank response’ indicates that HTrs can recognize their limitations and may choose not to provide a translation when they are unsure, adding a level of judgment that AI currently lacks. Moreover, the inability to translate some LTs reflects their insufficient depth of knowledge as well as a lack of practice in translation. In general, the significant variability in translation outputs highlighted the need for further improvement of both AI and HTrs capabilities, especially for specialized legal terminology. Finally, the study highlighted some challenges that influence the translation quality, namely linguistic, domain-specific, and technical factors, recommending that a combination of AI efficiency and human expertise might be the ideal approach to accurately translate LTs.