Doctoral researcher Brenner Judith

96000 €

AI-Powered Automatic Translation for Creative Video Games – Different Post-Editing Methods in the Translation Process and their Effects on Translations and Translators

Tieteellinen tutkimus / siihen pohjautuva työ | Kolmivuotinen

When AI-powered neural machine translation (NMT) systems are used to automatically translate creative texts in video games, these translations need to be corrected by video game translators, a process called post-editing of machine translation (MTPE). Contrary to the claim by NMT providers that MTPE is faster than translating from scratch, game translators perceive MTPE to be slower. They also voice concerns that it hinders them in their creative freedom and that it does not result in the same high-quality creative translations as without NMT. My doctoral research project aims to test these reservations and to identify MTPE methods that are faster, more satisfactory, and help achieve expected quality levels, and thus are better accepted by game translators for their creative process. So far, the common MTPE method is to edit texts entirely pre-translated by the machine. I compare this method with two innovative methods: 1) Translators can decide freely for each sentence whether to edit the NMT suggestion or to translate from scratch. 2) The NMT system learns and makes suggestions in real time based on the translator’s edits made and style used in previous sentences. For data collection, I cooperate with game companies interested in the results of this study. Keylogging and eye-tracking during the translation process will provide quantitative data, interviewing translators before and after using the MTPE methods will provide qualitative data, and assessing the errors found in the final, post-edited translations will provide additional data for a mixed-methods analysis. These analyses will uncover the impact of the three MTPE methods on translation productivity, translator satisfaction, and translation quality, both aggregated and per individual translators. These insights should support game translators in finding their preferred MTPE method so that they can translate faster, keep their creative freedom and be more satisfied using MTPE.