All Your Base

It was not easy localizing CRUFT, and though we've done it we know there will be some issues to sort out. The game was not built to be easily localized. It started is a little side project, then snowballed into a product. We have to keep things moving and we are currently working on about 1.5 new projects (one's still a prototype). Still, we always wanted to get CRUFT out to a wider audience and loc is an important part of any business model, so we leaned on AI to do some awkward lifting.

The results are mixed, but it was an interesting ride.

We have a precedent behind us to live up to.

First off, there was a lot of architectural shenanigans to sort out. Tagging every player-facing string uniquely took some doing. There was also a lot of coaching and reinforcing for the AI to understand the scope of work. I had to repeatedly remind the agent that it would eventually be doing the translation content. It kept trying different methods that assumed we had any budget whatsoever to send this to human beings to actually translate. Sorry friendo, this is on you.

Next, I really had to spend a lot of quality time with the agent teaching it about CRUFT as a game. We had long discussions about how the game was played. We watched some videos. Context is everything. After a bunch of chit-chat and some watching of gameplay footage, I set the agent to the task of actually translating the strings.

Mixed results.

Some really challenging conceptual stuff got translated really well (as far as I can tell by back translating with google translate). Some simple stuff, like the "Resume" button getting translated to "CV" in French, despite having the context of translation in the other languages to help with understanding, was puzzling. It is a mixed bag, but has been combed over and refined a bunch and has improved with each pass. Still, I know there are strings in there that just aren't going to hit the nail on the head.

There was, after the initial implementation, a lot of review work and changes that Will shouldered. Not fun stuff. Following up on an AI-agent-implemented and designer-driven technical task that touches the whole game is not an easy ride. The PRs were pretty massive in some spots and I'm sure the technological leaps in logic were a pain to retrace.

And yet... We're up on steam in French, Italian, German, Spanish, Brazilian Portuguese, Simplified Chinese, and Japanese.

Role model and spirit guide... the man himself.

In the past, I have had the esteemed privilege, and by that I mean I have been cursed, to be involved in a bunch of localization efforts at a variety of studios. It is pretty rough stuff that most devs would rather not touch. There is a ton of back-and-forth massaging of strings and a heaping pile of bugs that come out of the process. Quite often translation is happening when you are sleeping, so you show up to work with a bunch of emails and updates that need to be sorted out. I have nothing but the deepest respect for localization producers. I cannot imagine living that life but I'm glad it is a fit for some folks.

Where I'm going with this is that localization is hard/messy/annoying work. It has never, in my experience at least, gone smoothly... even when your localization partners are amazing. It is taxing. It creates problems where there weren't any. It is overhead. It is uncomfortable work. I do not speak many languages. I want my game to be represented well. I cannot in any way shape or form guarantee that representation in other languages. Even when an excellent translator goes back and forth with you on idiomatic stuff you are still in a bazaar stall on the other side of the world trying not to offend but still get the shiny at a reasonable price.

This has been the least painful round of localization I have ever been a part of.

How do I feel about all this? I settle in at a rating of OK. We use AI for some stuff and not other stuff. I feel like human language is well within the bounds of machine learnable without stealing. The amount of coaching and lifting needed to get the AI to translate even close to the target validated that it was not just a fire-and-forget effort. No jobs were stolen as there would be no translation at all without being able to do it this way and frankly I feel better about this than petitioning someone to do it for free.

It still leaves a lot to ponder.

There are a lot of out of work devs and support folks out there. I'm technically one of them myself. A huge part of the situation in the job market is due to how crazily many companies have invested in AI rather than humans to make games. So as I sit here at my desk, AI tools at the ready, plying my trade with the same tools that make it really hard to find a job right now, I'm a little mixed. I'm certainly not Robin Hood. I also certainly feel a bit penned in as an indie when I know that I can at least marginally translate our tiny game for no additional cost beyond just the time to shepherd the process through.

It would please me if the same technology that has left many of us out of work for the time being turned out to enable us to work on our own and not return at all. It is a big ol' shifty time right now. Everyone has their own take on if/how/where/when AI is cool. I still think it is terrible and unethical for art. I'm fine with it for code and apparently localization. I'm ok with it for music, but not singing. We live in interesting times.

Anyhoo, if you would prefer to play CRUFT in French, you may now do so. Thanks for following along!