MT-Pese: Machine Translation and Post-Editese

MT-Pese: Machine Translation and Post-Editese Sheila Castilho ADAPT Centre School of Computing Dublin City University sheila.castilho@adaptcentre.ie Natália Resende ADAPT Centre School of Computing Dublin City University natalia.resende@adaptcentre.ie Abstract This paper1introduces the MT-Pese pro- ject which is an umbrella name for a series of experiment venues that started in 2019. The project aims at researching the post- editese phenomena in machine-translated texts. We describe a range of experiments performed in order to gauge the effect of post-editese in different domains, back- translation, and quality. 1 Translationese and Post-editese A number of studies (Volansky et al., 2013) have shown evidence of the so-called translationese phenomena (Gellerstam, 1986), that is, statistical differences between translated texts and non- translated texts. Recently, post-editing (PE) of machine-translated (MT) texts has secured its space in the translation workflow for a variety of domains, and consequently, the research interest for the typical features of human-translated texts has shifted for the typical features of post-edited texts. However, results of studies searching for typical features of post-edited texts - what has been called “post-editese - have presented mixed results, that is, while some studies found evidence for the existence of post-editese (e.g. Toral, 2019; Castilho et al. 2019), other studies did not find evidence of the phenomena (e.g. Daems et al. 2017). The aim of the MT-Pese project is to investi- gate the post-editese phenomena on MT PE texts, using the rationale behind the translationese fea- tures as proposed by Baker (1996): simplification, explicitation, normalisation (or conservatism) and 1 ©️ 2022 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. levelling out (or convergence). We define post-ed- itese as the difference between the characteristics of human-translated texts (HT) and the PE ver- sions, in relation to the raw MT output. MT-Pese has researched what influences the features of post-editese in two different textual domains, namely, news and literature (Castilho et al., 2019). We found that the literature domain contained more post-editese features. In a further study, we looked into the post-editese features in two differ- ent genres within the literature domain (Castilho and Resende, 2022). Currently, the project is fo- cused on investigating the features of Post-editese on backtranslations (BT), with the aim to identify, for instance, if BT of PE versions would still carry strong post-editese features. Finally, the project also aims at addressing the question of whether the features of post-editese could be related to MT quality (section 4). 2 What influences the features of post- editese? A preliminary study This study (Castilho et al 2019) investigated the presence of post-editese in a corpus composed by HT, MT and PE texts post-edited by either professional translators or student translators in two domains: news and literature. We also tested whether the PE level (light PE vs. full PE). Results showed evidence of post-editese features manifested as PE texts closer to the source texts and raw MT output rather than HT texts, and that the translators’ experience as well as the …

Content truncated to 500 words