Assessing the Reliability of Machine Translation in Rendering English Medical Terminology into Standardized Arabic

Abstract

Medical translation requires terminological precision, yet the reliability of neural machine translation for English–Arabic medical terminology remains inadequately examined. This inquiry evaluates DeepL’s accuracy in rendering English medical terminology into standardized Arabic. DeepL translated 1,102 medical terms from the Unified Medical Dictionary. Thirty undergraduate translation students from Mustansiriyah University, Baghdad, Iraq, independently assessed each output against the standardized Arabic equivalent and coded translations as accurate, partially accurate, or inaccurate. DeepL achieved an overall accuracy rate of 47.55%. Single-word terms with established Arabic equivalents were translated most accurately, whereas compound terms, abbreviations, and acronyms posed the greatest challenges. Qualitative error analysis identified five error categories, with semantic distortion and transliteration misuse as the most frequent. Performance also varied across medical subfields, with anatomy achieving the highest accuracy and surgical terminology the lowest. Overall, the results indicate that DeepL cannot be relied upon as a standalone tool for standardized Arabic medical translation and that systematic human post-editing remains essential.