Although recent advancements in machine translation have improved lexical and grammatical accuracy, assessing its effectiveness in rendering contextually and semantically accurate translations remains inadequate. This gap overlooks how systems like Google Translate handle specialized terms and subtle shifts in language style, particularly when working with field-specific texts. The objective of this study was to evaluate the performance of Google Translate in rendering natural science texts from English into Indonesian, with particular focus on its semantic accuracy when compared to human translation. The data consisted of 76 sentences drawn from six scientific texts, which were segmented into 41 single words as well as 35 multi-word phrases obtained from professional ProZ translator portfolios, and subsequently translated using Google Translate. The translations were analyzed for inaccuracies, classified as under-translation, over-translation, and mistranslation, drawing on Baker’s (1992) theory of propositional meaning and also Nida’s (1964) receptor-oriented framework. The analysis revealed that while Google Translate frequently produced grammatically correct structures, it often failed to generate contextually appropriate or domain-specific terms, resulting in semantic inequi-valence with human translations. The findings indicate that these limitations are not inherent to the lexical output itself, but rather emerge from the system’s inability to capture nuanced meanings, specialized registers, as well as situational contexts. Therefore, human translators remain indispensable in ensuring accuracy and reliability in field-specific translations, whereas machine translation is best positioned as a supportive tool for general comprehension.
Arum et al. (Thu,) studied this question.