Disregarding Linguistics: A Critical Study of Google Translate's Syntactic/Semantic Errors in Rendering Multiword Units in English to Persian Translations

Translations rendered by machine translation have been found to be error-prone when translating medical communication language. Issues are especially frequent when Persian is either the source or target language. The source of errors has been attributed to machine translation's utilization of word-for-word translation and disregard for the formulaic nature of language. The aim of this research was, then, to investigate how multiword units (MWUs) of formulaic language (FL) are processed by a commonly used machine translator, Google Translate (GT). To do this, a medical-specific corpus was created to identify non-transparent MWUs. Adopting the framework by Simpson-Vlach and Ellis' (2010) n-gram criteria for MWUs, 20 frequently occurring MWUs were identified in the corpus. A comparison of GT's results with manual translations suggests that GT did not take FL into consideration in 50% of the data. These findings have implications for improving the accuracy of machine translation algorithms and reducing processing time.