Literary translation as a three-stage process: machine translation, post-editing and revision

Literary translation as a three-stage process: machine translation, post-editing and revision Lieve Macken, Bram Vanroy, Luca Desmet and Arda Tezcan LT3, Language and Translation Technology Team Ghent University Belgium {firstname.lastname}@ugent.be Abstract This study focuses on English-Dutch lit- erary translations that were created in a professional environment using an MT- enhanced workflow consisting of a three- stage process of automatic translation fol- lowed by post-editing and (mainly) mono- lingual revision. We compare the three successive versions of the target texts. We used different automatic metrics to mea- sure the (dis)similarity between the con- secutive versions and analyzed the linguis- tic characteristics of the three translation variants. Additionally, on a subset of 200 segments, we manually annotated all er- rors in the machine translation output and classified the different editing actions that were carried out. The results show that more editing occurred during revision than during post-editing and that the types of editing actions were different. 1 Introduction With the current quality of neural machine trans- lation (NMT) systems, the question arises whether post-editing NMT output is a viable alternative to human translation for real large-scale translation tasks. In this paper, we present the results of a case study on literary translations. We collabo- rated with Nuanxed, a book translation company, which uses an MT-enhanced workflow consisting of a three-stage process of automatic translation followed by post-editing and revision. In this case study, we compare three successive versions of the target texts as they proceed through © 2022 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. the translation process: the machine translation, the post-edited and the (mainly) monolingually re- vised translation. We used different automatic metrics to measure the (dis)similarity between the consecutive versions and to analyze the linguistic characteristics of the three translation variants. To assess the quality of the MT output and to get an insight into the editing actions that were carried out, a fine-grained manual annotation was carried out on a subset of 200 segments. 2 Related research Although employing Machine Translation (MT) for more creative text types such as literature may not seem to be a natural fit, several researchers looked into the feasibility of using MT for literary texts, first with statistical (Besacier and Schwartz, 2015; Toral and Way, 2015) and later with neural machine translation systems (Toral and Way, 2018; Kuzman et al., 2019; Toral et al., 2020). To assess the usefulness of MT for literary texts, researchers often compare raw (unedited) machine translations of literary texts with their human- translated (HT) counterparts. Three successive studies were conducted to assess the quality of generic NMT systems for English-Dutch literary texts, the language pair we also focus on in this study (Tezcan et al., 2019; Fonteyne et al., 2020; Webster et al., 2020). According to these studies, the main issues found in literary NMT are differ- ent types of mistranslations, coherence issues, and style & register problems. The percentage of NMT sentences that were free of …

Content truncated to 500 words