Stylistic adequacy in neural machine translation of Russian scientific texts into English: opportunities and limitations

this study evaluates the ability of neural machine translation (NMT) systems to adapt scientific texts to the stylistic conventions of the target language. The research analyzes the translation of a Russian-language economics article produced by DeepL, focusing on its compliance with the norms of English academic style. Key criteria include syntactic structure, voice usage, lexical specificity, and logical cohesion. The findings indicate that while NMT accurately conveys content, it often fails to fully adapt the text to English academic discourse. Specifically, the translations retain excessive syntactic complexity, literal renderings of Russian impersonal constructions and genitive chains, reducing the naturalness of the English output. However, the system successfully converts reflexive verbs into passive constructions and partially adjusts lexical choices. The practical significance of this work lies in identifying specific areas for improving NMT systems in scientific translation. The result can be applied to develop novel neural network training approaches, including creating specialized language models trained on annotated scientific corpora, developing automated stylistic analysis mechanisms, and implementing context-dependent text transformation algorithms. These advancements are particularly crucial given the increasing volume of international scholarly publications and the need to overcome linguistic disparities in academic communication.

machine translation, neural network technologies, academic style, stylistic adaptation, Russian-English translation, translation transformations