The bots are here. Should we panic or pop the champagne? Neural networks are already changing the way we localize games—but are they helping or replacing us? We asked game localization pros how AI is impacting their work. Let’s see what they revealed.
How long have you been working with Allcorrect?
- Alyssa Kollgaard, Head of Production and Operations
Nearly 5 years now!
- Valentina Ambrogio, Freelance Linguist
8 years.
- Stephanie Zhang, Freelance Linguist
4 years.
- Vadim Kiselev, Linguist
5+ years.
- Elizaveta Shevchenkoa, Localization Team Lead
6 years.
What was your first reaction when neural networks started knocking on the door of our industry?
Tell us how you felt and what was going through your mind when you saw AI tools creeping into the game localization space.
- Alyssa Kollgaard, Head of Production and Operations
In general, I am against tools that shortcut creativity and circumvent human experiences. I do think they can have their place to help reduce overhead or friction, but AI tools are not a replacement for people when it comes to creative roles (which localization absolutely is).
- Valentina Ambrogio, Freelance Linguist
Just like with every innovation that comes knocking at our door, I felt both curious and worried. Curious to see how this modern technology was going to evolve, but also worried because I felt that, eventually, we translators would risk being considered redundant and replaced rather than aided by this new tool.
- Stephanie Zhang, Freelance Linguist
I’ve never been enthusiastic about neural networks, and my initial reaction to seeing them spread into the game localization space wasn’t very positive, either. After learning about how large corporations developed LLMs—with stolen or pirated work—and following the actors’ and writers’ strikes in Hollywood, I could only respond to their proliferation in our industry with wariness. I actually started my translation career doing MTPE (machine translation post-editing), but after a few months, I did my best to transition away from that work. It truly requires a different set of skills than directly translating a text does, and I enjoyed the latter much more. Alas, here they are again, and their reach seems even broader this time. I did my best to avoid them for as long as I could last year, but this year, I’ve already had to work on a few games that were largely AI-translated.
- Vadim Kiselev, Linguist
I was okay with it, as it was rather expected. I’d already worked on AI-translated texts and thus had no major objections or prejudices against AI. However, I still caught myself thinking something like, “Well, that MIGHT be the end of your career,” but just for a second.
- Elizaveta Shevchenkoa, Localization Team Lead
A bit of panic at first. It felt like AI had come to take our jobs. No one really knew how serious it was or how fast things would change.
But then came the realization that AI isn’t the enemy; it’s a tool. When used wisely, it doesn’t replace us; it enhances us. Now, we see AI as a new teammate: helpful, but still very much in need of human supervision.
In your day-to-day work, have neural networks been more of a headache or a helping hand?
Maybe they save you tons of time, or maybe you spend more time fixing their “creative” translations than doing the job yourself. What’s your honest take?
- Alyssa Kollgaard, Head of Production and Operations
I haven’t (and wouldn’t) use them in the context of localization. The only time I have shortcut localization is to use Google Translate when I need to submit a page into certification and am short on time, but I ALWAYS come back and do a proper translation pass after that. I have, however, used fan translations and can say with 100% certainty that you spend more time and money for worse results doing LQA, retranslating, reintegrating, and retesting when you could have just done it right using a professional localizer from the beginning.
Localization isn’t just translating words 1:1. It has a lot of nuance! AI doesn’t understand the context of culture, relationships, gender, hierarchy, humor, tone, or gameplay intent. Localization isn’t just about making the words make sense; it needs all the rest of that context to work. Do the jokes work in the native language? Does the wording allow the right hints for solving puzzles? Is it conveying the emotion, the narrative you want it to? AI lacks the human experience to be able to accomplish any of what a native speaker (and gamer) can.
- Valentina Ambrogio, Freelance Linguist
In all honesty, I think that only 10% of the translations that I post-edit require small tweaks and amendments. I often find myself having to rephrase and retranslate good chunks of texts to add fluency and the idiomatic expressions that make even the simplest texts pleasant to read. I strongly believe that AI is being misused in fields that require a high level of creativity, such as game localization and subtitling, to name a few, where you need to have a good dose of wit, humor, sensibility, and a whole range of emotions to be able to deliver translations that make you just “feel.” It’s the human touch that I see lacking, the creativity, the feelings. Most texts are plain, follow the source’s structure, repetitive, and not necessarily consistent (we didn’t need AI to be consistent with terminology—we already had CAT tools for that). They just lack the human touch, which is what truly makes this work a soul-nurturing experience for both translators and end users.
- Stephanie Zhang, Freelance Linguist
Personally, I find neural networks to be quite a headache. Maybe this is because the languages I work with (Mandarin Chinese and English) are extremely different from each other, or because the history of ZH>EN translation is shorter than that of European languages, but I frequently encounter MT that just doesn’t quite work. For example, the resulting translation might have lost the rich language that made the source text sparkle, or it ends up being so literal that you have to edit it significantly (or give up on it if the deadline is looming). More than once, I’ve encountered ZH>EN MT files where the source text has been translated word-for-word, and the target text ends up not making much sense. I suppose, to be fair, MTPE works just fine for translating engine and system text, like menus and item descriptions, but world building, dialogue, and narrative text in particular suffer when run through AI, in my opinion. It’s not just the produced text that impacts a translator’s workflow either—the sped-up process of MTPE tasks does, too. I’ve always loved the written word and wanted to play a role in conveying the beauty found in the Chinese language to a wider audience. But the tighter time constraints and higher word counts of MTPE tasks frequently limit my ability to do that.
- Vadim Kiselev, Linguist
It depends. AI-assisted loc can definitely save lots of time and money if the use of AI tools is proper. I’d say the key factor here is to implement AI. Sometimes, there might be little slips, but it’s still more of a helping hand for me personally.
- Elizaveta Shevchenkoa, Localization Team Lead
AI has definitely become a helping hand. It speeds up routine tasks, offers quick analytics, and helps with first drafts. That kind of support can be a real timesaver.
But it can also be a headache. Sometimes, explaining to AI exactly what you want to convey in a client email—the tone, the key points to emphasize, etc.—takes so much time that it’s just easier to write it yourself and ask AI to check for typos afterward
Have you ever had a moment when an AI tool really impressed you or when it totally missed the mark?
Share a story! Whether it was a surprisingly spot-on localization or a total translation fail that made you laugh (or cry), we’d love to hear it.
- Alyssa Kollgaard, Head of Production and Operations
The only use case I have found in game development where an AI tool was very helpful was for live issue intake and triage. This is still a lot of clunky manual overhead between a producer, QA, community manager, marketing, engineers, and users, and it’s always been a struggle to build a proper pipeline that doesn’t require a huge amount of overhead. Streamlining that process actually does reduce time and produce better results. That’s solving a pain point and freeing up people to work on higher-priority and higher-impact tasks.
I am curious to see if AI tools are better or worse than Google Translate when it comes to localization—it seems like they are solving the same problem, and we didn’t need AI tools to begin with. I wouldn’t be able to infer the quality of the translation from either, though, as I’m not a native speaker of any language I’d translate to! That’s why we delegate to subject-matter experts. The only time you really will know if your translation failed is when it goes public and you get feedback from a user about the quality of your translation. Why risk it? It makes you look sloppy, cheap, and unprofessional to put out subpar work.
- Valentina Ambrogio, Freelance Linguist
Last year, I was assigned an MT evaluation task for a gardening game based on a popular comedian personality. I was really impressed by the general fluency of the text and how the humor was delivered, especially one joke based on the song “No Scrubs” by TLC and the wordplay of “No Shrubs,” beautifully translated with a perfect pun in Italian. At the end of the evaluation, it was utterly clear that it was not a raw MT output like the client had suggested.
- Stephanie Zhang, Freelance Linguist
I don’t have any specific stories I can think of at the moment. I do remember a file where one line of MT impressed me, but I can’t recall the specifics. I’ve dealt with far more examples of bad AI text, like the word-for-word translations I mentioned earlier.
- Vadim Kiselev, Linguist
Sometimes, the AI output might be rather creative. Every time the AI catches puns, wordplay, or some connotation hidden within a line, I think, “Kudos.”
- Elizaveta Shevchenkoa, Localization Team Lead
In one of the games we were localizing, there was a quest called “UNprotected.” In English, it can imply both “protected” and “unprotected,” making it quite tricky to reflect in translations. However, AI successfully detected the wordplay and highlighted the unique nature of the phrase. This helped the linguists, who might have missed the double meaning, to get a clear hint on how to localize it properly, capturing the wordplay and the intended meaning.
Looking ahead, do you see neural networks as partners we’ll learn to work with or competitors we’ll have to outsmart?
How do you think this technology will shape the future of localization in games? Should we be worried or just ready to adapt?
- Alyssa Kollgaard, Head of Production and Operations
It depends on the localizers’ preferred workflow. Is it better to do an AI pass and then edit it to be more correct, or does that add in unnecessary complexity, and it’s better to work from scratch? I can’t speak to that part. Does it allow you to achieve the same quality of results but faster and more efficiently? There are always going to be people looking to cut corners and costs, but there are just as many people advocating for the human element to remain in games who won’t compromise on translation services. It might be nice for solo teams who can’t afford translation services but still want to reach a broader audience.
- Valentina Ambrogio, Freelance Linguist
Ideally, the future will see us translators having the chance to choose whether to implement these tools into our workflow and assess whether a particular text can be expedited with machine translation or requires translation from scratch. More realistically, we will have to co-exist for the sake of faster turnarounds and reduced costs at the expense of creativity.
- Stephanie Zhang, Freelance Linguist
In 2025, AI translation is more prominent than ever. While it helps reduce costs and can support smaller studios, I have serious concerns about its long-term effects on the localization industry.
The pressure on localizers has grown: machine translation demands faster turnaround and higher volumes. Without regulations, workloads may become overwhelming, or jobs may be at risk.
Quality is another concern. AI often generates text based on previous AI output, not human writing, leading to errors and unnatural phrasing. If AI dominates a language’s corpus, we may see a decline in clarity, coherence, and cultural accuracy—especially in games rooted in myths or history.
For ZH>EN translation, the challenge is even greater. These languages differ vastly in grammar and culture. Despite recent improvements through collaboration, mistranslations are still common. I fear neural networks could normalize these issues, leading to dull or inaccurate English versions.
I’m not against AI, but I believe it should assist, not replace, human translators. Let AI handle repetitive tasks or provide research support, while humans focus on storytelling, nuance, and quality.
Right now, I’m more concerned about the consequences of AI translation than excited by its promise.
- Vadim Kiselev, Linguist
I could only guess, but I generally tend to think they’re the future, so we should start getting used to it. In a few years, I believe, MTPE won’t be an option but a standard.
- Elizaveta Shevchenkoa, Localization Team Lead
I see neural networks as partners we’ll learn to work with, not competitors we have to outsmart. It’s in our power to keep AI tools under control.
It’s actually a relief to delegate all the manual, repetitive, and boring tasks to AI. That gives us more time and mental space to focus on creative, meaningful, and value-adding work, especially in game localization, where nuance and emotional aspects really matter.
Of course, with any new technology, there’s always a bit of worry That’s natural. But what’s more important is to not worry too much and instead focus on how to adapt and extract the most value from it. With the right mindset, AI can help us work smarter, not harder.
Do you see any progress in the quality of text generation and translation from the early days of LLMs to today?
- Alyssa Kollgaard, Head of Production and Operations
Don’t know—haven’t used it, don’t plan to.
- Valentina Ambrogio, Freelance Linguist
Yes, especially in plain texts that are not very nuanced and idiomatic. But this was, I think, to be expected, considering they’ve been fed human translations over the last years.
- Stephanie Zhang, Freelance Linguist
I personally haven’t noticed a major change from the early days of LLMs to today. My first encounter with LLMs/MTPE was in 2021, and I think the quality of ZH>EN text remains pretty similar in 2025. I’ll acknowledge that, since I don’t use AI apps, my exposure to their translations and text generation is limited to videos I watch on the topic and the files I post-edit. But I did work on one game this year that was predominantly translated by AI, and I found the localization of its story, which was deeply rooted in traditional Chinese philosophy, quite awkward, even after several rounds of AI prompting and post-editing. One could argue that ZH>EN LLM won’t improve until it’s given more high-quality input, but as mentioned above, I have reservations about moving forward with that process.
- Vadim Kiselev, Linguist
Sure thing, they’ve definitely learned a lot! If previously one often had to translate AI output from scratch, now it often requires just some polishing. For some texts, AI translation engines have reached the quality comparable with human translation, and that’s impressive.
- Elizaveta Shevchenkoa, Localization Team Lead
I see a big improvement in the quality of text generation and translation compared to the early days of LLMs. It’s really inspiring to see how fast this technology is growing, especially thanks to feedback from real users like us.
We’re not just watching it happen; we’re actually part of the process. And that makes it even more exciting.
AI isn’t going away — but neither are we.
What’s clear from these conversations is that neural networks are neither the ultimate enemy nor the flawless savior. They’re tools — powerful ones — but tools nonetheless. And like any tool, they’re only as good as the hands that use them.
Whether you’re embracing the tech, cautiously experimenting, or side-eyeing it from a distance, one thing’s for sure: the future of game localization will be shaped by humans and machines — not one instead of the other.
