Abstract :
[en] Video game localisation presents a medium often divorced from visual context and texts arranged based on technical criteria rather than a narrative flow, operating under a "double-blind process" (Bernal-Merino, 2013). The field's dynamic nature situates it at the intersection of technology, linguistics, and social progress, putting pressure on localisers to create an immersive experience while reflecting the developers’ intentions. As such, the increasing inclusion of non-binary characters in video games poses new challenges for localisers dealing with gendered languages that lack established non-binary pronouns. Current translation strategies involve the use of either “direct or indirect non-binary language” (López, 2021; Lardelli & Gromann, 2023), where the former employs neologisms and the latter eliminates all gendered words. The complexity of the task coupled with the well-known crunch times that tend to define the video game industry call for the creation of resources to support localisers and reduce their cognitive load.
Building upon All-inGMT (Rivas Ginel & Theroine, 2022), a specialised Neural Machine Translation (NMT) system for non-binary language in video game localisation, we aim to develop a second tool specialised in neutralisation techniques. Similarly to the Dodiom initiative (Eryiğit et al., 2023) and given the substantial data required to train these systems, crowdsourcing and gamification arise as key approaches to compile our corpus. First, crowdsourcing methodologies facilitate the generation of considerable scientific knowledge and integrate citizens into processes with significant societal implications. Second, the gamification of crowdsourcing initiatives has demonstrated high efficacy in sustaining participants’ engagement and motivation (Morschheuser et al., 2019).
In this presentation, we will introduce NB: Automata, a gamified crowdsourcing initiative revolving around indirect non-binary translation based on Dodiom (Eryiğit et al., 2023). Using telegram to send the competing teams sentences that need to be rephrased to avoid gender markers, the participants will play two roles: neutralisers and reviewers. As a result, we will obtain a ranking of the most idiomatic solutions that will constitute the dataset used to train All-inGMT+. The Fun4All conference will provide the ideal opportunity to launch our gamified corpus project officially.
