Investigating Machine Translation Quality Across Genres
Sandra Navarro · 2024
This chapter focuses on machine translation (MT) quality assessment from the translator&s;s perspective. The research is motivated by the assumption that MT raw output is influenced by the linguistic patterns of translated texts contained in the training data, thus resulting in unusual features typical of has been referred to as ‘machine translationese’. Further, domain and text genre play a significant role in general-purpose MT quality performance. In this chapter, we provide a corpus-based comparative analysis of phraseologies in a corpus comprised of machine-translated texts and naturally occurring texts across three different text genres (abstracts, instruction manuals, travel brochures) in two language directions (English Portuguese). Results revealed that while machine-translated instruction manuals presented the highest level of adherence to the phraseology of original texts, abstracts demonstrated an intermediate level and travel brochures showed the lowest level of performance. The results and discussions herein shed light on the potential of adopting corpus linguistics (CL) principles and methods to the ever-more urgent task of assessing MT quality.