How will artificial intelligence change game translation?
Localization is the process ofadapting your content tomeet the needs and preferences of individuals in different cultures and countries. In the gaming industry, localization is key to opening up different international markets and making your game more accessible to a wider audience. Previously , game localization was usually handled by the company that made the game, a localization agency, or a local distributor . With thehelp of artificial intelligence (AI), localization isnoweasier and more straightforward than ever before ,and today we’re going to take a look at how AI is changing game localization. The traditional and typical game localization process involves of translators and editors(reviewers, checkers, etc.). This method includes translating text from the source language to the target language, correcting translations to account for cultural differences, and manually checking and correcting text embedded in images or graphics in your game. The people involved in localizationneeded to understand the context of the text as well as the game’s world , so they could capture cultural nuancesand translatethings like jokes andslang into the target language. Generally speaking, localization in the eyes of the expertsmeans going from initialtranslation, to expertreview, to finalproofreading by a nativespeaker. Traditional localization is time-consuming, quite expensive, and involves multiple people , so maintaining consistencyandminimizing human error iskey. AI-powered game localization uses machine learning (LM)and natural language processing (NLP) to increase the speed and efficiency of translation. AI translation systems have the ability to learn from large amounts of data to generate translations for specific game genres or styles. These systems can build an automated a Translation Memory (TM) to provide consistent translations, and translate in real time to support development quickly. It isalso possible to recognize various elements in the game and suggest contextualized translations, albeit not completely . In addition, translation software(computer-assistedtranslation tools, or CATtools) used for game localizationareincreasingly equipped with AI translation features, so both professional companies and individual translators can easily use AI translation without having to set up any difficult settings for a fee. Localization with AI continues to evolve, driven by advances in AI technology, and the industry expects it to eventually surpass humans in every way. However, it’s not quite there yet, so let’s take a look at the advantages and disadvantages of game AI localization to help you be selective . Pros: Disadvantages: Regardless of the field , it’sclear that AI is atrend that can’t be resisted. And, as mentioned earlier, it will eventually surpass human work in every way. In the real world, AI localization is already producing significantly better quality for IT and general documents. However, in the case of today’s games, the nature of the text means that the AIleaves a lot to be desired in terms of quality. If your focus is on relatively fast turnaround times and cost reduction, AIgame localization is a good choice. But if you’re prioritizing localization quality, it’s pretty clear that traditional localization is still the better choice. Since AItranslation is free for anyone regardless of the amount ofwork, it may behelpful to run thetext of the game you want to localize through an AI tosee the quality level. Today, most localization companies offer translation services using AI. AI localization produces very different results depending on the level and style of text, as well as the genre of the game, so it’s important to consult with your localization partner and test samples before choosing the right service.
Is LQAT Needed When Localizing a Game?
Game localization is divided into text localization and LQAT. In the past, it was common to only localize the text and not consider LQAT. This was in part because securing a development schedule before the release can be difficult, as well as less emphasis was placed on the importance of the quality of the translated text. However, as time went on, there were significant statistics that showed that a well-localized game increased sales. And as users began to directly request better localization quality, or even just localization, the demand for localization testing, or LQAT, grew. Today, we explore the specific reasons to conduct LQAT. LQAT, or Linguistic Quality Assurance Test, involves playing (testing) the game in order to find and fix linguistic errors. Games have special characteristics that make them different from other types of media (regular documents, video subtitles, webcomics, books, etc.). These are known as “sporadic in-game texts“. Other types of media usually have the text appearing sequentially. By looking at the surrounding sentences for context, the translation and tone can be easily determined. Games, however, are different. But for genres like stories are straightforward or visual novels, most text will appear sporadically. For extremely complex games like open-world games, this can be even more challenging. In these games, the text does not appear sequentially, due to the nature of the game medium. With a few examples, we look at this issue in more detail. First, let’s look at the text that appears on the screen of an open-world game. When launching an open-world game and looking at the screen, a lot of the text can be seen in multiple places. The text may be on a single screen, but if the assembled text does not have any references, it will be difficult to understand the correct context due to the lack of perspective. This is particularly apparent when it comes to text or a variable tag in the UI, as they are often very short, with one or two words. And, depending on where the text is being used, it’s often natural to translate it differently. Let’s take a look at a specific example of this. If, say, there are a lot of characters, you might want to create a variable tag that outputs a rotating list of character names, which then creates a sentence with that variable. This is not limited to character names; it is often used whenever a word is changed in the same type of sentence, such as an item, number, or region. <Example 1> There were no actual issues when solely translating the text. Issues arose when playing the actual game. This was in part because the text translated as “Mr. Gates” was an output of a tag with the variable <A>, something that would be realized during the translation process. Let’s look at another example. As you know, consistency in terminology is crucial when it comes to game localization. Displaying the same term with multiple versions of the translation may confuse the user, which may result in a lower rating for your game. This leads us to wonder: Are there times when consistency does more harm than good? <Example 2> In this example, like the previous one, it all comes down to context. Knowing where, and in what context, the text is being used helps in registering that the same word needs to be translated differently. Unlike other mediums in media, game text often needs to be translated differently depending on where it will be used. And, oftentimes, with the text being in an Excel file without any frame of reference, localization and contextualization is difficult. Therefore, solely translating game text may not be enough for the quality to be considered of good quality. Lastly, any number of technical issues can arise in a game. Even when the translation is good, the translated content may not be correctly displayed (i.e. errors where there is no text or in a different language.) When it comes to UI, there is a limit on the amount of text it can hold, resulting in texts that spills past the UI (aka Overflowing UI.) These are the types of issues that are unknown until you actually play the game. So far, we’ve discussed the nature of texts in the game and why it is hard to achieve solid quality with text localization alone, using specific errors as examples. For the sake of brevity, we’ve only scratched the surface, but there are many more types of errors that LQAT can catch. Of course, there are some in-game texts that are well-crafted in the development stage. In those cases, more contextual information is given than one might think. They are most likely unique values associated with each text, called “string information” or “key values”, or columns of information to be referenced during the translation process. In practice, it is extremely rare for text to have all this information, and even if it did, it would be nearly impossible for a text-only localization to be produced without a single error in the actual game. When planning to release a game for global markets, it is known that localization is no longer an option, but a necessity. Users are increasingly demanding better localization quality, and it is statistically noted that localization quality plays a very significant impact on revenue. When aiming to provide the best localization quality, we recommend not stopping at just text localization, but rather, to end strong by using a professional LQAT service . Remember: As the quality of your localization increases, so will your game’s ratings.
