There are plenty of articles, forum entries, Reddit threads and other player communities’ posts highlighting machine translation errors in video games. The critique? “Bad translations” (language and cultural accuracy) get in the way of delivering an engaging experience.
New developments address these issues, making machine translation (MT) more and more apt for creative industries such as game localization. However, it is also important to understand players’ wants and needs to know how to balance translation automation correctly.
What Players Want
According to expert game localizers, game localization for players is more than just being able to understand the game’s content and instructions. It’s about feeling represented in the market by having content in their native language, ultimately leading to more authentic experiences.
For players, gaming is also about belonging to a community, and language can be an obstacle to building rapport. Release simultaneity is also an important concern. Developers usually release games in English and localize them into FIGS (French, Italian, German, Spanish) first. This means that players outside these target groups must wait -sometimes a long time- to get localized content. However, for publishers, localizing games also depends on revenue and the number of users in that target language.
In many cases, players resort to doing and using fan translations because their games have not been localized in their native languages. Or because localization is not up to their standards. In some cases, companies incorporate these translations into the games. However, this also touches the subject of how ethical it is to benefit from fan translations and how much agency players have in game development and localization.
In summary, the expectations are simple. In theory: Having tailored experiences, on-time releases, and language accuracy.
Source: Fontolan et al., 2022
When to Use Automation in Game Localization
Automation through MT, in particular, targets speed and fluency. MT significantly increases localization times, allowing for the inclusion of more target languages and larger volumes of content. However, it’s crucial to understand when automation is appropriate and when human intervention is necessary.
The best candidates for raw or MT requiring light PE are the less culturally dependent and language-nuanced texts, such as:
- Non-Critical Texts: Text that doesn’t affect player’s interaction with the game, or poses a liability (such as legal text).
- High-volume and low-impact content: When time is a constraint, translating high volumes of content is possible with MT as long as it doesn’t require full PE. This depends on how well-trained engines are for the type of content and how culturally dependent the text is.
What content is safe to localize with MT then?
- Patch notes and updates: MT can efficiently handle rapid updates and patch notes, ensuring that players receive timely information in their native languages.
- UI-strings: They are repetitive and straightforward. This is usually maintained in TMs that can be used to leverage MT, making the process fast and efficient.
- Support documents: This type of content is ideal for MT since consistency and speed are key. Support documents include guides, FAQs, user-generated content, and website content.
- Real-time translation for multiplayer games: Real-time translation can enhance multiplayer experiences by allowing players from different linguistic backgrounds to communicate without an issue. See more about this here.
- Initial market testing: For indie developers, AI-driven translation can provide a cost-effective entry point for initial market testing, helping them gauge interest before investing in more comprehensive human-led localization.
Although we have talked about ready-MT for in-game text and high narrative games, they require more considerations for implementation that we won’t expand on here.
Doing Automation Right
Even with low-risk content, it is important to have tight QA checks to ensure fatal errors are not passed to localized content. Human expertise is indispensable for maintaining cultural sensitivity, creativity, and language nuances. A hybrid approach, combining AI-driven automation with human post-editing, offers the best of machine efficiency and human insight.
AI integration into game localization workflows is increasing with improved tools for quality estimation, analytics, and audio localization technologies. Automation implementation depends on the project’s needs, resources and other factors. Check for instance, this article for further information on how to implement automation in game localization.
