I didn't get the mic, so you get an article
By Marc Eybert-Guillon
Wednesday, March 20, 2024. The IGDA hosts a roundtable at GDC on localization, AI and machine translation. A great topical initiative and a very good conversation among professionals. A Capcom representative talks about how AI can be used to streamline internal processes, but should not be used for player-facing materials. Alexander Radkevich from CD Projekt Red highlights how they use MT and text-to-speech solutions for temp PL-EN translation and VO placeholders, but again avoid it for anything user-facing in their final product.
It gets pointed out that machine translation keeps improving at a rapid rate, hinting at a possible future where it’s “as good” as humans. The counter-argument I give is that it doesn’t matter how good it gets. Even if the surface quality of AI gets to near perfection, it cannot match what makes the human brain special, simply because of how it works. LLMs are aggregators. When you ask an MT engine to translate, you are asking it for the most probable translation for that source text according to its database. In other words, it always spits out the most common denominator. The most expected answer. The most boring option. To me, in creative fields such as games, movies or literature, the magic happens when people get creative. When they have that one special idea, or take some time to research a more obscure or unique term, or come up with a fun reference, joke or pun. The magic happens when people get to inject their personalities, quirks and sensibilities into the work. The machine will never do that. It simply cannot.
But I get to say all that at the roundtable, so all is good.
… Then a Keywords representative seemingly cosplaying as Tony Stark emerges from the shadows like a cartoon villain to give his take: AI is evolving at an incredible rate. AI is inevitable. AI should be embraced. Keywords can and is running projects where a large percentage of the output is handled by AI. It’s so much more efficient. So much more cost-effective. Many clients do not care about creativity and superior quality. They only want “good enough”.
During his speech, he says “you’re all creatives”. I think it’s very telling he did not include himself in that.
Alexander Radkevich replies that we should be trying to raise the bar, not lower it. I could not agree more. However, I had already talked quite a bit, so I couldn’t get a hold of the mic, which is fair enough, but now you’re getting this article.
Story time! A familiar story, but my story nonetheless:
I’ve wanted to be involved with games ever since I was a kid. I started playing on the Nintendo 64, but where my love of games really took hold was on the Gamecube with titles such as Metroid Prime, Super Mario Sunshine, Resident Evil 4, Paper Mario: TTYD, etc. From there, that love only grew with each gaming generation. As the fields that attracted me changed over time, the only constant remained that I would one day work in games. No matter how, I wanted to be one of the people contributing to that magic that had had such an impact on me. I wanted to have a hand in recreating this feeling for other kids, teens and adults… It turned out I enjoyed writing and translating… so I became a game localizer.
It doesn’t matter how good AI gets. It doesn’t matter how efficient it is. How cost-effective. I won’t let machine translation work in my stead because I want to do the work. I want it to be my words, not a machine’s. I’m a creative. I want to create. If a kid falls in love with games because of one I worked on, I want it to be my actual work that contributed to that feeling. So I will continue to create with my human brain, typing with my human fingers… until my business collapses, if that’s what it comes down to.
I don’t care if I can produce so much more and make that much more money by heavily relying on MT, I simply do not want to do it. I want to make money because I’m an adult with bills to pay, but what I really want deep down is to contribute to the making of games. Actually contribute, not press a button and have an algorithm do it for me. If it eventually becomes impossible for me to do my job as I intend because a bunch of Tony Stark wannabes have convinced enough people that MT is “good enough”, then I’d rather fuck off and do something else.
But I do believe there will always be people who see the difference and will be seeking that authenticity. People who want to create products that speak to others in a conversation between humans. For those seeking that care, that creativity, that honesty, that authenticity, they can stick with us. For those who don’t care about art, who want cheap and “good enough”, there’s always Keywords.
Edited on 21.03.24 to add that CDPR uses MT & text-to-speech for temp work, not other AI tools.