Fiction in Russian Translation: A Translationese Study

Proceedings of Recent Advances in Natural Language Processing, pages 734–743 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_084 734 Fiction in Russian Translation: A Translationese Study Maria Kunilovskaya1, Ekaterina Lapshinova-Koltunski2, and Ruslan Mitkov3 1,3University of Wolverhampton 2Saarland University 1,3maria.kunilovskaya@wlv.ac.uk, r.mitkov@wlv.ac.uk 2e.lapshinova@mx.uni-saarland.de Abstract This paper presents a translationese study based on the parallel data from the Russian Na- tional Corpus (RNC). We explored differences between literary texts originally authored in Russian and fiction translated into Russian from 11 languages. The texts are repre- sented with frequency-based features that cap- ture structural and lexical properties of lan- guage. Binary classification results indicate that literary translations can be distinguished from non-translations with an accuracy rang- ing from 82 to 92% depending on the source language and feature set. Multiclass classifica- tion confirms that translations from distant lan- guages are more distinct from non-translations than translations from languages that are typo- logically close to Russian. It also demonstrates that translations from same-family source lan- guages share translationese properties. Struc- tural features return more consistent results than features relying on external resources and capturing lexical properties of texts in both translationese detection and source language identification tasks. 1 Introduction This paper reports results of a translationese study based on literary texts translated from 11 source languages into Russian and included into the Rus- sian National Corpus1. Translationese is under- stood as specific linguistic properties of transla- tions distinguishing them from non-translated lan- guage. Most existing investigations into the prop- erties of translations have focused on text types other than fiction and, to the best of our knowl- edge, there were no large-scale investigations into translational Russian, especially based on literary texts. The research applications of the parallel multilingual corpus of translations into Russian 1See http://ruscorpora.ru/ are limited to constrastive studies such as (Dobro- vol’skij and Pöppel, 2017) and translation studies such as (Krasnopeyeva, 2016), and deal with a few individual items or constructions. We explore lin- guistic differences between fiction translated into Russian and originally-authored literature in Rus- sian with a focus on the impact of the source lan- guage (SL). The comparison between translated and non- translated Russian is based on an extended fea- ture set capturing structural and abstract lexical parameters of texts, including collocational prop- erties. This investigation mostly relies on the text classification approach and univariate statistical analyses following a methodology established in computational studies of translationese (Baroni and Bernardini, 2006; Volansky et al., 2015; Evert and Neumann, 2017, amongst others). Translationese indicators that cut across all translated subcorpora (if any) or are specific to a particular language pair are identified through feature selection. It does not come as a surprise that researchers are wary of using fiction for translationese studies. Fiction can be less homogeneous register because each literary work and author might display unique conceptual and linguistic properties reflecting a par- ticular manner of artistic expression and idiolect. With literary translation established as a form of creative endeavour of its own standing and with the covert ‘domesticating’ translation paradigm ac- cepted as a …

Content truncated to 500 words