Will AI machine translation help fan translation game projects?

[rocking3]

I personally don't think AI is all there yet. It makes mistakes frequently, and even makes shit up. I do wonder how accurate it will get at translating human languages though. I don't personally know that much about what it takes to translate a game, and I imagine it's more than just being able to speak the two languages. What do you guys think?

 

[RageBurner]

I hope so, and in some cases it already has, though the key here is how much work the AI actually does. Without a second pass by a knowledgeable human translator, results will be spotty at best.

That said, models are growing in complexity very fast. I think they will be reliable tools for this kind of thing sooner than we thought possible, especially when quantum computing becomes commonplace.

 

[Zachery2112]

I've played a few games with machine translations and I can best describe these games in two words: they suck. AI does not understand language like a human can. It does not pick up on subtext, innuendos, irony, or history in conversation. As such, most AI translated crap I've seen is direct and bland.

That's not even getting into the problems with differences in grammar that AI has no idea how to account for. Japanese especially is known for the subjects of sentences being unspoken and implied. AI can't pick up on that and nearly every game I've played with AI translations have this problem.

I speak Spanish as well as English and I've noticed a similar problem with AI translations because Spanish also often leaves the subjects and objects to implication. That's two European languages struggling to cross the barrier with AI, let alone all the other tongues the world has to offer with their unique vocabulary and grammar.

For the time being, AI is crap and I believe if anything it will worsen the translation scene if it is used as a crutch. I'd rather wait half-a-decade to get a fan made project than a couple months for an AI translation.

 

[LiberoVulpes]

AI alone can't really do something good, far too many hurdles and nuances that i can't see it realistically being able to overcome for a while.

I have used it for RPG Maker games, and it really leaves a lot to be desired.

 

[UnDubious]

They would already be useful if we were trying to translate French games into English or something, but 99% of the time we want translations from Japanese, and they still suck. I mess with it quite a bit for non-game stuff. It makes tons of mistake - it still can't tell who's speaking or who or what is being talked about most of the time, and fails to recognize expressions and figures or speech. It still produces complete nonsense sentences regularly. I've done comparisons using different software, and for example, DeepL will produce more readable sentences, but it'll just drop essential information from the original text if it doesn't know what to do with it. Too much of Japanese is context-dependant, both grammatically and culturally.

As it is now, if a game requires English text to be appreciated - any RPG, adventure, or action-adventure game, anything with a real story or dialogue - AI will ruin your enjoyment of the game. I tried playing Zill O'Ill on the PS2 using that translation last year and it was horrible, I gave up just minutes in. Same with one of the PSP Shining games. Unreadable dialogue.

A lot of progress was made between, say, 2015 and the early 2020s, but I haven't noticed progress in the last couple years.

A proper translator going over stuff would have to essentially re-translate everything if they have standards. A person who doesn't understand Japanese going over stuff to make it readable will try to fix what they can, but they're working from a very flawed translation, so serious errors and issues will remain. It's not a solution at this point.

 

[Hotwer]

At the state that they are in right now, no. But perhaps one day it will be good enough so that people can use it as a tool to help them translate stuff. Even then, I still believe a human element is still needed as there will be lots of nuances that are lost in translation when AI translate them.

 

[What's His Face]

What I hate is that it's being used for things like auto clubs and crap. Imagine being on the side of the freeway with a busted car screaming into your phone because the AI is shit and screws-up your request for help that you pay good money for. I weep for the day AI is used for actual emergency services.

"Hello, 9-9-9? My neighbour is being murdered! ... MURDERED!! ... No, not neutered, MURDERED!!! No, I do not need a veterinarian, I need the police! POLICE!!! ...Who are you calling 'obese' you bloody wanker!? ... You bloody bolloxing gormless plank, get me a human being! HUMAN. BEING. ... GET ME THE BOBBIES YOU STUPID SACK OF PIG SHITE!! ... Wait, you're sending the police to arrest me?! What the bloody fook for??!! 'Racism?! What the flying...?"

 

[Aziamuth]

AI WILL NEVER REPLACE HUMANS GRAAAAAAAAAAAAAAAAAH!

 

[Bonkers225]

Give it about 2-3 more years.

 

[Nidstang]

I not only think it will be used for fan translations, it will be used for official translations.

I personally don't think AI does a great job when it comes to translation, beyond the basics. But localizers have developed such a terrible reputation in the last 10 years, I know many people would prefer finnicky AI translations than bastardized localizations at this point.

I personally think it's swapping one problem for another problem, but it's what is going to be used.

 

[CG4R12I4N]

I'd say yes.
Because ultimately it's just a tool. They're supposed to merely help the translator getting something faster done outputwise and not being a main job. It's only when misused to the point of almost leaving everything to the AI and not doing comparison quality check that it could ruin the fantranslation scenes, and that's more of the fault of the lazy and retarded translators themselves than the AI then.

 

[Chapu]

My two cents: as it stands now, AI is useful for doing the initial heavy lifting of translation, and barely even that. You can run a whole script through it and you'll get _something_, but if you want it to be anywhere near good a human has to proofread it. And this is not 100% the AI's fault: when translating a game the most common scenario is getting the script extracted without any context whatsoever. You *have* to play the game fully to check for consistency and such, make the necessary changes on the go.

Having said all that, whether AI will help fan translations is still up in the air for me. What do we have more availabilty of, hackers or translators? If the former, then I guess it could help streamline some projects (with mediocre results in the translation department, at best). But if there's more translators, then AI shouldn't make an impact in the amount of finalized projects itself, at best in the time they take.

 

[Leon]

AI translation on it's own, sucks, if someone after using a ai translator goes through and proof-reads it though to clean up the meaning, it could speed up fan translations significantly, problem is it seems like it's either all in or nothing.

 

[BassForever]

Give it about 2-3 more years.

Machine translation tools have existed for decades, they are absolutely not 2-3 years away from being anywhere near good enough.

 

[Zachery2112]

Machine translation tools have existed for decades, they are absolutely not 2-3 years away from being anywhere near good enough.

This is something worth noting that everyone seems to have forgotten. Translation tools have existed for a long time but for some reason people are realizing they exist now. I'm guessing that ChatGPT made everyone forget that Google Translate exists.

 

[RageBurner]

Even with its faults, AI is on a whole other level compared to MT, however. It is a form of MT, but not quite the same. That said, calling it an AI is overblowing its capabilities. It's closer to a trained language parser that happens to be able to use the weights (biases) it was trained with against the data it was fed to produce approximately what is being prompted; you could say it's just semantics, but it's not a true intelligence.

 

[Bonkers225]

Machine translation tools have existed for decades, they are absolutely not 2-3 years away from being anywhere near good enough.

Not the same as AI. In case of AI, you can train it to become better, which is what field techs are doing nowadays, and corporations hiring people to improve AI by engaging conversations with it every day. And as time goes on, AI will learn to adapt more and more for a certain scenario. "Google translate" is nothing like this. Comparing the two is like comparing a Tech Toy(Brazil manufacturer) emulation system to a Steamdeck in gamer terms.

 

[BassForever]

Not the same as AI. In case of AI, you can train it to become better, which is what field techs are doing nowadays, and corporations hiring people to improve AI by engaging conversations with it every day. And as time goes on, AI will learn to adapt more and more for a certain scenario. "Google translate" is nothing like this. Comparing the two is like comparing a Tech Toy(Brazil manufacturer) emulation system to a Steamdeck in gamer terms.

You know who I trust to train AI to be good at translating a complicated language like Japanese into English? Field techs who can barely help these AI learn basic English. There are far simpler tasks that AI continue to struggle with despite years of training.

 

[Bonkers225]

You know who I trust to train AI to be good at translating a complicated language like Japanese into English? Field techs who can barely help these AI learn basic English. There are far simpler tasks that AI continue to struggle with despite years of training.

Do you have any idea how long it took computers to get to where it is now? Things take time, but it will eventually get there. And once it does, it will make quality of life so much easier.

 

[fake-kun]

Why do people insist on making AI-related threads? They always descend into shit.

 

[Egg (McMuffin) Man]

Most fan translations use some form of machine translation, usually DeepL or Google Translate. Hell, even a lot of official translations use machine translations. After the machine translation the human "translator" will go through the English output and correct instances of incorrect grammar or change things that don't make sense.

AI translation will simply replace existing machine translation platforms. Anyone saying that AI will do a bad job is probably unaware that they have already played many machine translated games already.

That being said, I'm quite confident that the AI will make many mistakes just as current machine translation platforms make lots of mistakes. But you would never know unless you learn Japanese and play the game in Japanese.

I don't personally know that much about what it takes to translate a game, and I imagine it's more than just being able to speak the two languages

A lot. Languages aren't 1:1, there's lots of stuff in Language A that doesn't make sense in Language B and those things will need to be changed. And not just linguistic differences, but also cultural differences; it's not uncommon to make references to something that someone from another country wouldn't understand, and that would need to be changed also. And of course, the actual content of the game may be considered offensive by someone of a different culture background, so things of that nature would also need to be changed. So not only do you need to be able to speak two languages, but you need to have a knowledge of both the culture of the base language and the culture of the target language. And in the case of being an English translator in the video game or anime/manga industry, you also need too have a raging hate boner against basically everything about the material that you're being paid to translate, and you will need to utilize that in order to censor the content and then brag about it to your followers on Twitter.

 

[rocking3]

Languages aren't 1:1, there's lots of stuff in Language A that doesn't make sense in Language B and those things will need to be changed. And not just linguistic differences, but also cultural differences; it's not uncommon to make references to something that someone from another country wouldn't understand, and that would need to be changed also. And of course, the actual content of the game may be considered offensive by someone of a different culture background, so things of that nature would also need to be changed. So not only do you need to be able to speak two languages, but you need to have a knowledge of both the culture of the base language and the culture of the target language. And in the case of being an English translator in the video game or anime/manga industry, you also need too have a raging hate boner against basically everything about the material that you're being paid to translate, and you will need to utilize that in order to censor the content and then brag about it to your followers on Twitter.

I've been teaching myself Japanese over the last year, and totally know what you're talking about. It's opened my perspective such that I've gained a lot of respect for those in the community who are translating these games by hand.

 

[SpaceCadetGenocide]

I bet it'll likely eclipse machine translations but it won't replace translators completely because somebody will always have to be around to polish things that the machine didn't catch

 

[skagskagskag]

putting aside AI being as costly as it is for the environment, from my understanding a lot of language model AI create their output as an aggregate of their training data. I feel like this fact inherently blocks it off from creating a proper translation of anything. Sure, you can get so far, like I don't get too bothered machine translating map directions if I need to get around. But as video games are written art pieces, like a book being translated there needs to be creative interpretation on the part of the translator. There are many statements out there that could be written in one single precise way, but have many interpretations depending on the context. There could be many contexts throughout a text that effect a statement's meaning.

Another thought I'd be concerned with is favoritism among languages. Sure, japanese/english/chinese/french/etc may have decent enough models for translation (ignoring why i said i dont think theyd be ideal). But relying on AI for the translation of lesser and lesser popular languages will have diminishing returns based on the people working on the models and their specific biases. It's kinda like how a lot of image models have had issues with primarily presenting the users with white people when given prompts describing a person or subject neutrally in respect to race.

 

[Valcrist]

No. Everything that comes from it is trash.