Post-Editing Neural MT in Medical LSP: Lexico-Grammatical Patterns and Distortion in the Communication of Specialized Knowledge

Abstract

The recent arrival on the market of high-performing neural MT engines will likely lead to a profound transformation of the translation profession. The purpose of this study is to explore how this paradigm change impacts the post-editing process, with a focus on lexico-grammatical patterns that are used in the communication of specialized knowledge. A corpus of 109 medical abstracts pre-translated from English into French by the neural MT engine DeepL and post-edited by master’s students in translation was used to study potential distortions in the translation of lexico-grammatical patterns. The results suggest that neural MT leads to specific sources of distortion in the translation of these patterns, not unlike what has previously been observed in human translation. These observations highlight the need to pay particular attention to lexico-grammatical patterns when post-editing neural MT in order to achieve functional equivalence in the translation of specialized texts.

Exploring undergraduate students’ experiences in dealing with post-editing of machine translation

“…In addition to grammar and vocabulary, the above findings, as presented in Figure 3, also indicated that undergraduate students were faced with these other particular problems identified by some researchers in the relevant fields, i.e. grammar, vocabularies, and technical terms (Harto et al, 2020a;Yamada, 2019), lexico-grammatical patterns (Martikainen, 2019), complex sentences, strange phrases, slang words, and idioms (Quah, 2006), and lexico-semantic (Sycz-Opon & Galuskina, 2017). The students found problems with English tenses and collocations, for instance, particularly when they translated texts from Indonesian into English.…”

Assessing Human Post-Editing Efforts to Compare the Performance of Three Machine Translation Engines for English to Russian Translation of Cochrane Plain Language Health Information: Results of a Randomised Comparison

“…A researcher affiliated with the Cochrane France team (who is leading Cochrane's French translation activities) pre-translated a corpus of Cochrane review abstracts with DeepL and had them post-edited by translation master students to assess lexicogrammatical patterns for potential distortions. She showed that DeepL created specific sources of distortion of the translated patterns and argued for the need for special focus on these lexicogrammatical patterns in the post-editing efforts, in order to improve the quality of machine translation of medical texts [13].…”

Students’ Linguistic Challenges in Post-Editing of Machine Translation

“…In a postediting process, student-3 (S3) should be able to correctly choose these affixes to make the passive sentences relevant to the intended meaning. As future post-editors, the students should also be trained on these particular lexico-grammatical patterns (Martikainen, 2019). These differences could lead to errors potentially made by the students in post-editing practice as indicated in Quotation [10].…”