Getting Gender Right in Video Games, Simplifying for Inclusivity

Abstract

The video game industry’s meteoric rise since its birth in the 1970s has steadily increased the demand for professional localisers capable of dealing with “game crunch” (Brogan 2022) in what can only be described as “a double-blind process (no audiovisual context, no text linearity)” (Bernal-Merino 2013). Due to these working conditions, previous research in the field (Rivas Ginel 2021) has shown that the second most common error found by linguistic testers is mistranslations, a phenomenon that results from the above-mentioned constraints as well as the presence of variables. Variables, in the context of video games, are used to create the illusion of immersion by, for example, allowing the player to change the name (or gender) of the character. Therefore, translators are forced to find creative yet natural solutions to accommodate potential changes in the resulting sentence through simplification, neutralisation, and the use of controlled language⁠. This article, the last of a series of papers analysing gender bias caused by machine translation in video games, aims to provide an in-depth analysis of the above-mentioned techniques used by translators to deal with gender when translating from English into French. We begin by providing a brief overview of the characteristics of video game localisation and their impact on the translators’ work. Afterwards, we discuss some examples of mistranslations found in other games and the output generated by state-of-the-art baseline machine translation systems (Rivas Ginel and Theroine 2022). Subsequently, we will analyse neutralisation and simplification techniques extracted from a parallel corpus compiled from already translated video games that include non-binary characters. Finally, we present some of the results of an ongoing project that aims at conceiving a neural machine translation (NMT) tool for English into French video game localisation specialised in neutralisation techniques.