So I’ve been ignoring machine translated fan translation patches for a long time because they suck.
But it’s impossible to ignore how much better LLM generated translations have gotten in the past year or so. The suckage isn’t as much of an issue, like sure there’s nuance missing and clever jokes don’t get translated correctly but it’s like 98% of the way there. That’s great, exceeding many official translations.
Pair that with leaps in LLM code generation and reverse engineering = fan translations of games that nobody has paid much attention to. Sure many of the best games got labors of love that I would never want to replace but I never thought I’d see a translation of Torneko no Daibouken 3 for Game Boy Advance. But here it is:
I’m actually shocked at how good this is. Even a few months ago I saw AI slop jobs for Dragon Quest IV and Dragon Warrior Monsters 1+2 for PSX that, while appreciated, were still in the ignore category. But here’s a game where the gameplay is front-and-center, the dialog details aren’t really important, I can get by with my limited Japanese language skills even, but this translation is more than just serviceable everything seems fine? Ran into zero bugs or missed lines. Some odd verb choices like having to “push” a healing put to use it but I can accept that.
I’m a few hours in, and really enjoying the game as someone that had Torneko 2 for PS1 as a kid and occasionally dabble in assorted Mystery Dungeon games.
So I thought why not a thread to document cases where the AI generated fan translations are actually worth bothering with. Anything you’ve run into lately you want to share?
I’m not necessarily against machine translation as a placeholder provided it is checked by someone who speaks the original language. It’s absurd to me that you’d release something with no idea whether it is even adequate - much less accurate.
But I’m also concerned about the limited testing some of these translations get and the lack of emphasis on getting them to work on original hardware. I think clout chasers are starting to spoil the scene.
Yeah awesome. Tools being developed to make translations easier overall will be so useful. Some games already have fully translated scripts from 20 years ago but nobody has coded them into the game.
Hmm, well translations were not accurate in the 90s either, and honestly these RPG stories are often slop anyway. Oh yet another Chozo one who can weild the ancient power to stop etc.
For me a translation is helpful just to get through menus, understand objectives and actually be able to get through the game. So if a machine translation gets it to that level it is better than nothing? And certainly better than Castlevania II.
I was translating the unreleased Dreamcast game, Giest Force. I don’t know if I’ll ever bother to finish it now with all these machine translations out there with zero effort put in to them.
By the way, I should make it clear that I was only translating the text. My plan was to partner with someone who was capable of putting the English text back in to the game.
It’s more like you could put the effort into a good textual translation and rather than be beholden to someone very talented to put in many (free) hours of hacking the original code someone with moderate technical skills could help turn that into a working translation.
I don’t think it needs to be slop all the way down.
I mean sure - but why would I want to spend my time playing something made by someone who can’t even check their own work? Or has even bothered to test it?
Two interesting points, or should I say assumptions?
I understand checking to mean that the translation makes sense, and testing to mean that it doesn’t break the game.
Of course it’s possible that somebody prompting these things may not be checking or testing. But it’s also possible that they could be prompting more thorough checking and testing than is possible by a human in any reasonable time. How do we know which one describes their work? We don’t know.
Automated checking is probably the hardest of the two, but not impossible. With programming, itself a type of language, there are formal grammar rules and multiple methods of checking structure. With a translation of text you could use proven services like DeepL or Google Translate (both powered by “ai” and self-improving since they began, fwiw) as well as translations using the LLM itself, and compare and contrast. And of course there’s still room for a human to add their own or have final say what goes.
Automated testing, on the other hand, is easy. With software we can map the whole thing, every path that can be taken in the code, every result. With each change we reproduce the map and see if it is different to the baseline. None of this requires a machine, but would take time, so tools to do it have existed for decades. The ai agents could drive those tools more quickly and act on the results, by fixing and repeating, until it passes.
So I think what I’m happy to criticise, as you are, is lazy or inexperienced use of a tool. But how do we know if that applies to the person wielding the tool? We don’t. Same as with any game you might buy. Try it and review it, or don’t try it and read the reviews; not trying it and reviewing it seems unfair.
One more thing I thought of, I think we’re seeing lots of people suddenly become software engineers overnight whilst knowing little about the processes involved.
This one came a couple days ago, GBA Ace Attorney 1,2,3 with the script from the DS games:
There’s been a patch for Ace Attorney 3 for a while now but nice to see complete ports of the other two.
I think this is an interesting case of taking an official translation and backporting it to older releases. Would be nice if someone fixed up the broken MSX translation of Snatcher using this approach.
Full disclosure I am someone who has used LLMs at their job and for help on creative things so I for sure can grasp the excitement of maybe being able to tackle a project that seemed out of reach before - that’s my expectation for a lot of these translation projects. I do think the cream will always rise - slop will always be slop and good work will obviously stand out. I do not think it diminishes hard “real” work, but I do think there is a bigger opportunity for things to be translated that didn’t have much of a shot before - an AI assisted translated menu so someone can get through a game is a win imo I think.
Bums me out that it may disincline some good work from being done but I understand how it could be discouraging.
On a serious note, who doesn’t use machine translation for something? Let’s be real – it’s a standout achievement of the modern era. The improvements seen over the last couple of decades are astonishing.
It seems we’re getting this the wrong way round. People in the ‘developed’ world don’t want to work. This is why AI has become so dominant, so swiftly.
Yes, the technology has advanced, but the real story is our decline as a species. Look around – people can’t even tie their own shoelaces without an app. If your phone is on 24/7, you are the reason this happened. By this, I mean the rapid proliferation of AI in our day-to-day lives.
As someone who doesn’t live online, I can see this clearly and actually welcome the tech. @matt is spot on – it’s a tool. If it’s anything less than that, or more, you only have yourself to blame. Bleating about it after the fact won’t change anything.
