Machine translation is a powerful ally that can escalate translation solutions in record time. However, despite its many advantages, it is still not ready for implementation in video game localization. Why? Because it still misses the spark of creativity, cultural references, and expressions that resonate with target audiences. And players do know the difference.
Now let’s talk about games with interactive storylines. They present even a bigger challenge due to their creative and dynamic nature, and a myriad of storytelling that require context-dependent translation. We have talked about how MT is very useful for non-context-dependent game elements, but here, we will focus on how we can make this possible thanks to the latest advances in terms of MT and AI.
Understanding Interactive Storylines in Games
Video games need stories, they keep players hooked and excited. In that way, video games are very much like books or movies. They all have some degree of narrative that explains or dictates the rules of the game. However, in gaming, players can experience games in active ways. Now, interactive storylines take the matter of agency a bit further by giving players more control over their characters and the gaming worlds in general (Lebowitz & Klug, 2011).
However, the creative nature of these types of games makes localization, and machine translation, especially difficult. Interactive storylines are characterized by being:
- Dynamic.
- Context-dependent.
- Narrative tends to be unstructured (which poses an issue when localization is involved).
As such, localization must rely heavily on contextual information and adopt humour and tone to give target audiences the same experience.
Tailoring MT for Interactive Storylines
Considering the essence of interactive storylines, is it possible to use MT to localize these video games? Yes, but systems must be customized and maintained for it.
Research and advocacy on using MT in video gaming localization is growing. The efforts are aimed at reducing MT’s shortcomings and adapting it to translate creative content. In that regard, there are already some proposals to address issues with the translation of idiomatic expressions and context-dependent information:
- Using Interactive Neural Machine Translation (INMT): INMT involves using neural networks to provide on-the-fly hints and suggestions to human translators, enhancing the translation process by making it faster and more efficient. Although it requires a lot of human and development effort, it can improve MT outputs’ quality over time.
- Using Human-in-the-Loop approaches: Combining AI with human oversight allows for high-quality translations that capture nuances and context, which is particularly important in interactive storylines. Different from INMT, HITL approaches use linguists’ feedback to retrain engines in real time instead of having the engines predict outcomes from the linguists’ edits.
Implementing Game-Ready MT
Whatever the approach to adapt MT to game content is, it is essential to incorporate good data management and post-editing practices to make the most out of MT outputs and linguists’ efforts. Here are some basic steps to make sure outputs are suitable for translating creative content:
- Train engines right: Do not expect engines to perform well for a game if it contains only generic data. To enhance systems capabilities, engines must be trained in the domain they’re intended for. This means not only including gaming data, but genre-specific data. Here is also the tricky part, since games with a high narrative component, it might tend to include inappropriate language that must be curated before the data is used to train the engines.
- Prepare the content: In general terms, this means following the same practices as for localization. This involves extracting and formatting game text for MT, getting rid of images, codes, or dates, and ensuring the formats’ compatibility with MT tools.
- Choose the right MT tool: Selecting an MT provider that supports dynamic content, offers control over MT parameters, and allows post-editing. Flexibility is key here, since you want to be able to tweak raw outputs, introduce rules, exceptions and/or lower MT thresholds.
- Aim for integration: Ensuring seamless integration of MT into post-editing and localization pipelines will reduce time and back and forth between teams. One of the advantages of MT is timesaving, so you want to maximize this by reducing the steps needed to have content ready for development teams.
As an additional consideration, when post-editing, make sure to have clear standards regarding light and full-post editing. As with localization, best practice is having style guides, glossaries and notes available for linguists so they don’t spend more time than necessary editing raw MT outputs. Laying this out for experts doing post-editing will help them focus on the creative aspect of translations that comes with narrative games. Also, good post-edited content can be used to improve engines’ performance.
Future of Game-Ready MT
We’re approaching a time when machine translation can be used at a bigger scale in video game localization. Developers were already experimenting with it before the LLMs advent, and now the landscape looks even more promising with:
Advancements in AI and Generative Models
Future advancements in AI, more specifically in generative models, will likely enhance MT capabilities in gaming. The expectation is in improving contextual understanding and reducing the need for human intervention for minor errors to let linguists focus on the more creative aspects of translation.
Potential Challenges and Opportunities
Challenges remain on mainly two fronts: The language professional side, and the quality of MT outputs for gaming content.
Regarding linguists’ position in front of the localization process. We have talked before about linguists’ concerns regarding compensation for their work, and power over the creative process of translation. As hybrid and AI-workflows solidify in the industry, this is something that needs to be addressed by LSPs.
As for MT quality, despite advancements, the challenge of maintaining cultural sensitivity and handling complex narratives remains. Players are very wary of poor machine translation outputs, and with all due reason. A bad translation gets in the way of delivering an immersive experience. Again, here the key is to keep working to ensure good MT outputs + post-editing practices to make the best out of MT and language professionals’ expertise. We’re getting closer and closer to this goal though.
