Gender-Fair Post-Editing: A Case Study Beyond the Binary

Gender-Fair Post-Editing: A Case Study Beyond the Binary Manuel Lardelli University of Graz, Austria manuel.lardelli@uni-graz.at Dagmar Gromann University of Vienna, Austria dagmar.gromann@gmail.com Abstract Machine Translation (MT) models are well-known to suffer from gender bias, especially for gender beyond a binary conception. Due to the multiplicity of language-specific strategies for gender rep- resentation beyond the binary, debiasing MT is extremely challenging. As an alter- native, we propose a case study on gender- fair post-editing. In this study, six pro- fessional translators each post-edited three English to German machine translations. For each translation, participants were in- structed to use a different gender-fair lan- guage approach, that is, gender-neutral re- wording, gender-inclusive characters, and a neosystem. The focus of this study is not on translation quality but rather on the ease of integrating gender-fair language into the post-editing process. Findings from non-participant observation and in- terviews show clear differences in tempo- ral and cognitive effort between partici- pants and GFL approach as well as in the success of using gender-fair language. 1 Introduction Gender bias in Machine Translation (MT) has been studied from different angles and a lot of work has been published for debiasing MT, however, only recently from the perspective of bias beyond a binary conception of gender. Most approaches (Piergentili et al., 2023; Savoldi et al., 2021) dis- cuss it from a theoretical perspective. One ex- ception is Saunders et al. (2020), who proposed © 2023 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. a gender-tagging approach to translate from in- flected to gender-neutral language with moderate success. Gender-fair debiasing MT models is chal- lenging, since there is a lack of datasets and even human translators find it difficult to select and cor- rectly apply gender-fair language strategies. As a first step, we propose a case study to investigate gender-fair language within the context of post- editing. Post-editing approaches generally focus on speed, productivity, cognitive load, and quality (Jia et al., 2019; Toral et al., 2018). To the best of our knowledge, this is the first post-editing study to fo- cus on gender-fair language. Non-binary individ- uals have become increasingly visible, such as in TV series like One Day at a Time or Sex Education, and the number and type of strategies to linguis- tically represent them differs considerably across languages. Gender-fair language (GFL) subsumes gender-neutral, that is, omitting any gender refer- ences, and gender-inclusive language, that is, lin- guistically including all genders. With grammati- cal differences and a multitude of strategies across languages, gender-fair translation is challenging for machines and humans. Nevertheless, human and machine translators can act as ambassadors for gender equality beyond the binary by using gender-fair language. In this case study, we chose post-editing over translation from scratch to focus on the tempo- ral and cognitive effort required to revise an exist- ing translation in terms of gender references. Six professional translators post-edited three machine translations from English to German, which con- tain references to non-binary …

Content truncated to 500 words