Incorporating Human Translator Style into English-Turkish Literary Machine Translation

Incorporating Human Translator Style into English-Turkish Literary Machine Translation Zeynep Yirmibes¸o˘glu, Olgun Dursun, Harun Dallı, Mehmet S¸ahin, Ena Hodzik, Sabri G¨urses, Tunga G¨ung¨or Bo˘gazic¸i University Istanbul, T¨urkiye, 34342 {zeynep.yirmibesoglu, olgun.dursun, harun.dalli, mehmet.sahin5, ena.hodzik, sabri.gurses, gungort} @boun.edu.tr Abstract Although machine translation systems are mostly designed to serve in the general do- main, there is a growing tendency to adapt these systems to other domains like liter- ary translation. In this paper, we focus on English-Turkish literary translation and develop machine translation models that take into account the stylistic features of translators. We fine-tune a pre-trained ma- chine translation model by the manually- aligned works of a particular translator. We make a detailed analysis of the effects of manual and automatic alignments, data augmentation methods, and corpus size on the translations. We propose an approach based on stylistic features to evaluate the style of a translator in the output transla- tions. We show that the human translator style can be highly recreated in the target machine translations by adapting the mod- els to the style of the translator. 1 Introduction Machine translation (MT) work has included lit- erary texts in its agenda in the last decade and recent studies have shown some evidence for the possible contribution of machine translation in lit- erary translation (Toral and Way, 2015; Toral and Way, 2018). A few studies focused on the trans- lator style in relation to machine translation (e.g., Kenny and Winters (2020)), but to the best of our knowledge no research has embarked on building customized machine translation models evaluated on style metrics in literary texts. © 2023 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. In this paper, we aim at creating machine trans- lation models which could generate outputs with literary style, particularly with the style of a trans- lator. As a case study, we focus on the English- Turkish language pair. We make an analysis of lit- erary style by following a hybrid methodology and identify the lexical and syntactic features that can reflect the translator’s style. We compile and man- ually align a corpus comprised of the works of a literary translator. By fine-tuning a pre-trained ma- chine translation model on the corpus, we analyze in depth the effects of manual and automatic align- ments, data augmentation techniques, and corpus size on both the translation quality and the style of the translations. We show that a machine transla- tion system can be adapted to the style of a trans- lator to obtain literary translations with that partic- ular style. The contributions in this paper are as follows: • We introduce the first study in Turkish lit- erary machine translation that trains models specific to a translator’s works • We make a detailed analysis of literary style by following a hybrid methodology • We build a manually-aligned corpus of a dis- tinguished Turkish literary translator • We devise a method that filters the alignments made by automatic alignment tools • …

Content truncated to 500 words