Google Translate for Medical Texts: A Quantitative-Qualitative Analysis of English into Arabic Package Inserts Translation

Although machine translation systems like Google Translate have made great strides, there are still concerns about their use for medical translation. Medical experts, researchers, and end-users doubt that Google Translate could pose serious risks, as it may distort the original meaning or omit vital information. This study argues that Google Translate should not be perceived as risky, mainly when translating package inserts from English into Arabic, as one example of medical texts. This argument stems from a quantitative-qualitative analysis of Google Translate’s translation performance, utilizing a corpus of 50 package inserts obtained from the Saudi Food and Drug Authority with their official Arabic translations. The quantitative analysis employed statistical measures to compare Google Translate’s output to the official translations, assess post-editing effort, validate whether end-users can distinguish between Google Translate’s output and official translations, and describe the accuracy and fluency error distribution. Simultaneously, the qualitative analysis involved a manual inspection of a random sample of 760 sentence pairs, employing Tezcan et al.’s (2018) taxonomy of translation errors to identify and categorize errors as accuracy-related or fluency-related. The results revealed significant differences between Google Translate’s output and the official translations, although these disparities were predominantly attributed to stylistic variations rather than errors. The results also showed that end-users were mostly unable to discern between Google Translate's output and the official translations. Moreover, only 165 out of the 760 sentences contained errors, with the majority being fluency-related rather than accuracy-related. Google Translate’s output, evaluated in this study, was generated in November 2023.