Take the stress out of video game localization with neural machine translation
You’ve made a great game, but, when it comes to localization and translation, a lot of developers feel a little lost. In this blog entry, we’ll demystify part of the translation process by focusing on the newest tool at game developers’ disposal: machine translation.
Did you know that the machine translation market is expected to grow by over 19% between 2017 and 2024? There are lots of mixed messages about the quality and usefulness of this approach, but we’re going to explain how machine translation works and its advantages.
You’ll also learn about the limitations to this option, as well as what kinds of content are best suited to this approach. At the end, you’ll have a better understanding of how to translate video games via this process and expand into new markets. Your work deserves to reach as many people as possible and machine translation is just the tool to make that happen!
Introducing neural machine translation
In the past, machine translation technology was notoriously unreliable, since word-for-word translations often resulted in strange, incomprehensible, and even humorous word choices. Translators and linguists warned clients about the inherent dangers of leaving the vital task of conveying a brand message to a machine, one that cannot understand human thoughts and emotions. That all changed with the advent of neural machine translation.
Today, the best machine learning tools employ a deep-learning mechanism to improve the quality of translation outputs. Machine translation using deep learning has become much more common over the past few years, thus significantly expanding the machine translation market size. In simple terms, the tools learn from a huge corpus of data and then performs a given task over and over – in this case, translating texts – improving the quality of their output with each new iteration. Machine translation technology is fed massive amounts of previously translated documents, which it processes. By analysing the source and target documents, which have been translated or corrected by human translators, it is able to continuously improve its own output. Recently, additional efforts have been put into researching attention-based neural machine translation. This follows a similar process, but focuses on particular parts of the source sentence rather than the phrase as a whole.
Over time, as an increasing number of translations are processed by the tool, it learns how to recognise what is correct and what is incorrect. This means that machine translation deep learning improves with frequent use. That means there are major long-term benefits to choosing this translation approach. Similarly, effective approaches to attention-based neural machine translation also involve a long-term commitment, since, over time, the machine translation tool learns what elements are particularly worthy of attention.
Can neural machine translation really replace a translator?
While machine translation deep learning has been a huge step forward, where machine translation really comes into its own is when it is paired with a human contribution. This leads us to machine translation post editing, or MTPE.
How does this work? Well, once your content has been translated by machine translation software, your output will be worked on by a translator. They can take the output and polish it, correcting any mistranslations, grammatical errors, or disfluencies that may appear. A human translator can also fix elements that a computer is simply unable to handle: humour, emotions, cultural references, etc. This means that the previously translated content is polished to a higher standard, resulting in a high-quality translation that retains a human touch, even after it has passed through the machine translation process.
There are major advantages to machine translation post editing for game developers and players. Firstly, the entire process is now much quicker, because machine translation is almost instantaneous. Since much of the translation will be usable – especially if you have a well-trained neural machine translation engine at your disposal – the human translator will be able to polish the final product far more easily, cutting down on the time needed for localization.
Secondly, machine translation post editing can drastically reduce costs. Many localization service providers have started to offer MTPE as a service, with their rates being significantly lower for this service than full translation. This is largely based on the increased speed and throughput of this process, which allows the translator to optimise their time. Furthermore, machine translation post editing rates decrease over time, since the quality of the machine translation output improves with each new iteration. In effect, translators not only polish your content, they also train your tool to perform even better when you next need to call on it.
It’s clear, then, that machine translation post editing is not about replacing a human translator, but rather about combining the best parts of both approaches. That said, you still need to carefully consider what method is best for your game and your content. Machine translation post editing is a fantastic service, but it’s important to know exactly when to use it. But don’t worry – more on that below!
Machine translation and human translation: choosing the right approach at the right time
We’ve already discussed the ways in which machine translation can increase throughputs and lower costs, so it’s natural to assume that post editing is most effective for projects with tight turn-around times and restricted budgets. This is true, in theory, but there are other considerations that you need to keep in mind.
Computers – and, in turn, machine translation technology – have come a long way, but machines still can’t think like people. If you’ve spent weeks or months crafting the perfect ad campaign for your game, then you want your creativity to be conveyed in every market. Unfortunately, even the best machine translation tools aren’t creative, so human translation is often the best option. Even machine translation post editing is of little value in these cases, since the output is often completely unnatural and needs to be reworked entirely.
Alternatively, if your content is less emotive, MTPE is a viable option. This is especially true if you are planning continuous updates to your content, since the tool will become increasingly accurate over time, further lowering costs and shortening delivery times. That’s particularly good news for developers working on mobile games, MMOs and MOBAs. Good neural machine translation is like an investment; the more you put into it now, the greater the dividends for future projects.
Leveraging your assets for more efficient neural machine translation
Every developer knows that their team is their most valuable asset and that the success of their game relies on their hard work and diligence. The same is true with machine translation, since the development team can really make a big difference upstream.
A good development team is aware of what video game localization entails and, therefore, takes the relevant factors into account when creating the software. For example, if you’re opting to use neural machine translation without any subsequent human input (that is, no post editing), it’s vital to leave space for expanding and contracting strings, since no human will be there to ensure that character limitations are adhered to. If this is not done, display errors will occur.
The development team isn’t the only asset at your disposal, however. Today, translation memories are one of the most valuable tools employed by human translators, so why wouldn’t you take advantage of them? If you are using the same language services provider for each job, ask them to create a translation memory in which each translation unit (that is, each sentence or phrase that has been translated) is stored.
When a new translation project arises, any sentences that have previously been translated can be automatically translated at a significantly reduced cost, since they will be automatically handled by the translation memory. Not only does this lower costs – and shorten deadlines, since some content will be automatically translated with a human-approved translation from a prior task – it also guarantees consistency between projects. If a previous game opts for a specific translation for a term, for example, this term can be carried forward to the sequel title, lending the player an immediate sense of familiarity. Your memories are also portable, which means you can send them to a new translation vendor if you decide to change video game localization services. With all of these assets already available to you, it only makes sense to leverage them in the translation process!
A new approach to neural machine translation
Altagram has developed a cutting-edge machine translation tool. This tool for machine translation, using deep-learning methods, has been fed with content from numerous games, making it an accurate and powerful tool in your translation arsenal. Our new machine translation tool will be available alongside our upcoming localization platform. If you would like to learn more, or to find out how you can become an early adopter, fill in this form with your contact information and we will get in touch with you.