von Philipp Techen

Everyone is talking about it: Artificial intelligence – creator of a new paradigm or harbinger of doom. This text will assume you have heard a little about it already, specifically about the uproar regarding potential damage done to artists due to copyright infringement, or more simply put, the copying of art.

Why is this relevant for translation? Do translators even have copyrights? Yes, they do. While the industry at large would rather sweep this knowledge under the rug, translators are creators of a unique product, often creative, and as such they do retain the copyright to their creation.

A great example for this is books in the public domain. As you might know, if no one extends a copyright on a very old piece of writing, or after a certain number of decades (varying in different countries), it becomes part of the public domain and may be freely used. An example would be classics such as Jane Austen’s books and even some newer ones like The Great Gatsby. But even if a book’s original version enters the public domain, translations of it were often done much later and will likely still be copyrighted.

Things get even more interesting when we look at our main topic: copyright in video game translation.

In theory, the translator of a video game translates an often unique and creative version and legally holds the copyright to their translation. In practice, this is almost never taken into account. The legal situation is obscured by companies and their elaborate contracts, often stipulating that translators waive all rights to their creations. According to German law, it is not even legally possible to pass on a copyright but only to grant usage rights, which in other professional areas would usually come with a fee for the client. These mandatory NDAs can become rather fantastical, containing paragraphs about “ignoring existing copyright regulations” and “passing all rights on in perpetuity”. They know they’re on legally weak ground and are trying really hard to convince us they’re not. Even though a lot of legal mumbo jumbo would likely not hold up in any court, clients keep using these scare tactics; they rely on the fact that most translators would never dare challenge them, given their weaker position as individuals versus companies with large legal budgets.

So why don’t translators just say „No, thank you?“ Where are the protest signs? Why do we let them get away with it? Why don’t we ignore or sue these clients? The sad truth is, it’s not that easy because the seeds of confusion were planted so long ago.

The translation industry started to create precursors to the technology long before the current concept of large language model AI systems arose. Translation memory systems (see below) save a piece of text and its translation so they can be reused at will (1). This is great for consistency and can really make complicated texts with fairly clear repetitions easier to translate. Unfortunately, as these systems have become more and more server-based and are held by the big companies, the translations end up in the clients’ hands to reuse however they want. That’s right: any translator’s work can and certainly will be reused by other translators or automated systems for updates on an existing title, but also for entirely new projects that the translator might never even work on, while in most cases translators themselves are not allowed to reuse their own work independently.

And while this is an accepted practice (as translators do typically also benefit from it in the workflow), it plays right into the larger issue at hand. Filling AI systems with existing translation memories is easy, and as some big companies hold translation memories for thousands of titles, they can create a generative artificial intelligence with ease, neatly bypassing (or rather ignoring) the issue of the titles and translators’ rights to existing texts.

Fortunately, this doesn’t happen on too large a scale (yet), but rather for each project (as consistency in style within a project is always a major factor). But looking at a very large project with millions of lines of text – for example a large annually updated massive multiplayer game – you can imagine how an automated system could bypass translators and copyright for fast and cheap results. And with all of this, I’m only talking about the companies’ internal AI engines and not public systems such as Google Translate or DeepL, which stir up a much larger hornet’s nest of copyright infringement and confidentiality. This trigger word, ‘confidentiality’, still keeps most large companies from dabbling with these tools. However, cost reduction is always alluring, and it wouldn’t be the first legally ambiguous shortcut taken by players in the industry for the sake of smoothing a budget and meeting a deadline.

So where are we at? While the video game industry flourishes, translators working within it have rarely received recognition (#translatorsinthecredits) or appropriate pay for their works. With large providers buying up all the small ones, what little leverage individual translators have is increasingly diminished in the general market. I’m not saying that AI is creating a new crisis for translators, it’s merely pouring fuel on a very healthy fire that’s been burning for many years. And unless everyone really wants AI to take the translation of novels, TV, and video games out of human hands (trust me: even if you can’t imagine it yet, you really don’t want it), things will have to change.

Right now no one has the perfect answer, but that shouldn’t stop us from reflecting and trying to shift the stubborn building blocks of a near-monopolistic localisation industry. Should we try to push for clearer laws on AI worldwide and include machine translation functionalities and translation memories to leave no wiggle room for abuse? Yes! Should we try to establish new rules and standards for the game industry (and other fields) on what constitutes the acceptable use of our copyrighted translations? Absolutely! Should we make clear that copyright is not a joke that can be disregarded at will and investigate the right steps to ensure that our hard work won’t become the thing that’s used to replace us? Hell yeah! The bottom line is: We’ve stayed silent long enough. And the most dangerous thing we can do right now is to remain passive and watch as technology evolves around us without sufficient input by the people deeply affected by it.

So, this sounds ambiguous. Am I calling for you to pick up pitchforks? No. A revolution is rarely a realistic goal, and improving our peaceful cooperation with companies that dominate the market is in everyone’s best interest. But if we want to put a stop to the steadily decreasing quality standards in our industry for the sake of art – for our love of language – then we must stick together and become strong enough to occasionally just say NO. If you’re part of the AVÜ or a different association, get involved and find out how to get your voice heard. And if you’re not involved yet, join us and work with us for a sustainable future.


I am not a legal professional and while this is well-researched (most relevant parts are in §§ 3, 29 and 31 of the German Copyright Act (UrhG) and there is supporting European law), I’m sure the big companies still have some loopholes that I don’t know about. The legal information refers to German law and can differ wildly throughout Europe and the world, so you may have to look into your countries’ laws.

Lektorat:  Brenda Benthien

TRANSLATION MEMORIES: Translation memories save texts in multiple languages. The text might be separated by sentences or paragraphs or simply the content of an Excel cell. When the software recognises a similar text, it displays the previous one and highlights the changes between the texts. For instance, if “He petted a dog” will be followed by “He petted a cat”, the translation memory will offer the previous translation and highlight the word "cat" as the change that needs to be considered.

He petted a dog – Er streichelte einen Hund. [Translator's confirmed translation]

He petted a cat – Er streichelte einen Hund. [Automatically inserted text 90% match in which the word for dog has to be replaced according to the software recommendation. Actually though, 2 words have to be changed.]

Newer software uses algorithms to cut these texts into smaller snippets and can automatically put them back together in different patterns to offer machine translation, for instance by recognising that the word "cat" was at the end of “I once had a cat” and then taking the word from this translation and adding it to “He petted a cat”.

I once had a cat – Ich hatte mal eine Katze. [Translator's confirmed translation]

He petted a cat – Er streichelte einen Katze. [Badly auto combined from smaller snippets. Doesn’t take into account the switch from masculine (for dog) to feminine (for cat). More advanced systems might get this right.]

He petted a cat – Er streichelte eine Katze. [Adjusted by the translator]

And all of this could similarly happen with a 500-word text that only contains minor changes (the larger the amount of text, the more changes are allowed since it’s all about statistical percentages. See my article on the fuzzy grid for more on how this is abused to pay translators less.)