• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Trails series developer Falcom eyes using AI translation to shorten localization process to 6 months, or maybe even make the releases simultaneous

Thick Thighs Save Lives

NeoGAF's Physical Games Advocate Extraordinaire
20241022-38784-001.jpg


In a recent interview with 4Gamer, Nihon Falcom’s CEO Toshihiro Kondo shared his views on using AI to translate games. According to the creator, the company is open to potentially introducing AI into the initial stages of their localization process to make global rollout more efficient.

“When developing a game, speed is an important factor. Until about 20 years ago, 80 to 90 percent of our players were Japanese. However, recently, the ratio of overseas users – especially from Asia, has been increasing. It has come to the point that if our games don’t sell overseas, we could end up not being able to release games in Japan either.” Against this background, Kondo considers that AI translation could be a way to shorten the time needed to get Nihon Falcom’s games to audiences outside of Japan.

This includes the Trails series, according to Kondo. However, the CEO appears aware that AI translation would need to be supervised and edited by humans. “The Trails series is all about enjoying the story, so human effort would still be needed to make final adjustments to the translated words and lines.”

While it has been disputed that editing AI translations can be more time-consuming for staff than working from scratch, Kondo believes that using AI could help Nihon Falcom optimize their localization process. “From a management perspective, I think that by leaving the initial translation work up to AI, we can compress the entire process and ultimately speed up development.”

By using AI translation, Nihon Falcom also hopes to shorten the gaps between Japanese and overseas releases of future Trails games. “We can’t release Trails titles simultaneously worldwide because the amount of text in them is several times larger than that of a typical RPG. Translation begins after we’re done working on the Japanese version, so the overseas versions come out a year later at best. Hence, if we could use AI translation to reduce this to six months, or maybe even make the releases simultaneous, it would be a blessing for us as a company, and it wouldn’t be a bad thing for the fans either.”

 

yazenov

Gold Member
Good. Fuck those woke translators adding their shit takes and agendas to the translation instead of what the source material intended to convey.

This is a win-win for us consumers. Less bulshit translations and faster time.
 
Last edited:

NanaMiku

Gold Member
So this is why they still haven't confirmed the English version is by XSEED or NISA.

I hope if they chose to use AI, they really handled it well. I've seen a lot of bad MTL manga scanlation
 
I'm fine if they use AI for small things such as localization.
I do hope they'll proofread it though. Ready Or Not used AI for some localization (ie. subtitles) and they barely made any sense.
 

Pejo

Member
I'm fine if they use AI for small things such as localization.
I do hope they'll proofread it though. Ready Or Not used AI for some localization (ie. subtitles) and they barely made any sense.
I'm unfamiliar with Ready or Not, did they use AI localization software or did they just use Machine Translation (running the text through a translator). The AI stuff shows capabilities to identify stiff/incorrect wording and to keep character personalities consistent throughout the story. This is a pretty new technology for localization.
 

LiquidMetal14

hide your water-based mammals
These companies just need to do some QA and this could be a great idea. Not all these companies are made of money. I approve.
 

Madjaba

Member
It has already been done on some fan translation (Trails into Daybreak before the official announcement and release for example) by using AI to translate the script and then rewriting by a human to make sure everything is making sense.

As far as I could see for the brief hours I played the Fan Translated games, it was even with the official translation that I played after that.

As a JRPG fans it means having every games without waiting too much for translation (and not localization...)
But as a person I'm not for people losing their jobs though so I hope that it will be an evolution of the job instead of a pure replacement.
 

MMaRsu

Member
Fucking trash

On HBO I was watching Curb Your Enthusiasm, and where Larry talked about a stop & chat, the subtitles talked about "setting up a tree together"

Like what the actual fuck?

Fuck you AI shit titles
 

FingerBang

Member
Honestly, this is unavoidable and, in the long run, it's a good thing. I'm not happy with AI "taking jobs from people" but I'm now realizing it's pointless to fight progress. All these tools make us work faster. Fewer people will be needed to create stuff. Supervision will ALWAYS be required, and more products will be able to be created quickly.
 

mèx

Member
As AI gets better, it makes sense for translators to leverage it as a tool to provide better and faster translations. I think this is inevitable.

But doing an automatic translation in bulk and then edit the whole thing? Sounds dumb as hell.
 

NeoIkaruGAF

Gold Member
Is AI capable enough to translate efficiently from a language such as Japanese? The rare times I use Google Translate for some Japanese titles, the results are pretty hit and miss. The mass of AI-translated text would still have to be proofread by humans, and during that phase nothing would prevent some people to change things to inject… pretty much what we’re seeing now, anyway. If anything, the risk is that the machine translation would be proofread by less experienced people than those usually involved in game localization. Things could actually get worse.
 

IFireflyl

Gold Member
Is AI capable enough to translate efficiently from a language such as Japanese? The rare times I use Google Translate for some Japanese titles, the results are pretty hit and miss. The mass of AI-translated text would still have to be proofread by humans, and during that phase nothing would prevent some people to change things to inject… pretty much what we’re seeing now, anyway. If anything, the risk is that the machine translation would be proofread by less experienced people than those usually involved in game localization. Things could actually get worse.

It would be cheaper to use AI to translate and then hire a team to proofread than it would be to hire a team to handle the translation.
 

NanaMiku

Gold Member
Is AI capable enough to translate efficiently from a language such as Japanese? The rare times I use Google Translate for some Japanese titles, the results are pretty hit and miss. The mass of AI-translated text would still have to be proofread by humans, and during that phase nothing would prevent some people to change things to inject… pretty much what we’re seeing now, anyway. If anything, the risk is that the machine translation would be proofread by less experienced people than those usually involved in game localization. Things could actually get worse.
Not just that, Japanese also like to cheap out. Sometimes they only give all texts in Excel without any accompanying material.
 
I'm unfamiliar with Ready or Not, did they use AI localization software or did they just use Machine Translation (running the text through a translator). The AI stuff shows capabilities to identify stiff/incorrect wording and to keep character personalities consistent throughout the story. This is a pretty new technology for localization.
Not sure but let me explain;
Your character says 'Lead to TOC'.
Subtitles will display it as 'Lead to Talk'.
And bunch other mistaken subtitles.
 

YeulEmeralda

Linux User
Localisation is so weird. Imagine localising Breaking Bad to Dutch:

Hello I have cancer can I have million eurodollars chemo?
Sure thing sir take a seat.
End credits.
 

DonkeyPunchJr

World’s Biggest Weeb
The bar for quality writing is REALLY LOW.

Correct me if I’m wrong but I suspect that right now, the localization process is something like:
1. Start with some rough Google Translate style machine translation
2. Have a human writer take that and (in the best case) word it in a way that sounds natural in English and remains true to the character + the writer’s original intent, or (in the worst case) take lots of liberties, injecting it with their own biases and “witty” attempts at humor

The new way could be like:
1. Start with an AI that’s fluent in both languages and trained on some great works of fantasy/sci fi/anime/etc
2. Prompt the AI with whatever backing documentation, writers’ notes, concepts, character bios, etc that the original writers want to provide it
3. Have the AI produce a localized script
4. Localization team just gives it the final overview/sanity check (and could even use the AI to assist with that e.g. “what was this character’s motivation for saying X?”)


I think there’s a lot of potential to make localizations not just faster, but better. This could be really interesting.
 

Ridicululzz

Member
Yeah this could work if they edit and clean up the translation afterwards. Obviously AI translations aren’t going to be that great, and will definitely mess up but I think if used properly as a tool it could for sure speed things up
 
Last edited:

Pelao

Member
AI translation would need to be supervised and edited by humans.
I translate movies and TV shows into my native language for a living, and this is literally how I've been working for over a year now. It's ridiculous how much faster I get my work done because of it. Correcting mistakes here and there is way quicker than typing everything from scratch. A movie used to take me a whole day. Now I have it done by lunch and get paid the same.
 
I don't have a problem with this as long as the AI's translation is trained on their own data and they still have a couple people curating & overseeing the process to make sure the translations are coherent.

Whether we like it or not, AI is here to stay, so the best thing to do is determine what ethical boundaries there should be. If Falcom aren't stealing some translator's work to train their model (or stealing English literature works to do the same) without compensating them, then fine. If the data sets are in-house or under license from some company providing theirs at a good price, fine. If they still have some actual people to curate the results and make sure they're accurate & sound fluent, great.

They save a lot of time and money, and fans outside of Japan get the games faster. That's a win-win.

You realize professional translation is rarely literal, especially from a language like Japanese?

Some people are just obsessed with "woke this, woke that". They can't help it. I just hope they're no older than teenagers 'cuz otherwise, hooo boy :/....
 
Last edited:

Kagoshima_Luke

Gold Member
Option #1: Delay the release for 1 year while everything is translated by hand, yet still ends up being garbage.

Option #2: Completely AI translate everything in 15 minutes and spend 1 week going over the results and correcting issues. Simultaneous worldwide release.
 

ReyBrujo

Member
1. Start with some rough Google Translate style machine translation

I doubt nowadays they do that, they probably have a localization team with the original script and start working looking at some video for context. After all, they will have the VA read those translations and they need to match the context.
 
Things like this is why I think we are at the peak of how expensive game development is, and it'll actually start to go down.
People think it's just going to keep increasing, but with better tools and better AI you shorten the dev time.

Same there is going to be some issues for a good 5-10 more years but I do agree AI will start bringing cost back down.
 
Same there is going to be some issues for a good 5-10 more years but I do agree AI will start bringing cost back down.

Yea, people think AI will lead to everything being generic, but I think it'll have a lot of use cases. Like if AI could have a hi res scan of 100,000 trees and it can instantly give 10 games hundreds of trees that never repeat, can't even find the same tree in any of those 10 games. What's the problem again with using it?
It could have so many assets, it's far from generic. Especially if you still just do the stories 100% manually.
 

Saber

Member
Good tidings I say. With this games not only can arrive faster, but they also can respect the original intent of dialogue without worries of the west.
 
Best news I've read all day, honestly

e0wdp8.jpeg
Speaking of, I fed the upper pick into ChatGPT. It said

The text in the image is in Japanese and translates roughly as:

"I'll make sure to easily, quickly, and thoroughly defeat someone as senile as you."

The name "Erika" appears above the dialogue, indicating that this is the character speaking.

It already deserves more of a salary than those translators
 
I honestly look forward to AI translations of Japanese games. They'll probably produce more interesting and faithful dialogue which in the end will retain much of the original meaning and intent. Plus, make for more memorable/quotable catch phrases and lines again.

This is one instance where I see AI improving and upholding artistic visions. Especially, if its in-house trained models.
 
Last edited:
They'll still need a human proof reader to check what the AI generated, and that step could still result in the human inserting their woke agenda. But at least it could be faster than the current eternal wait for Falcom games I guess lol
 

NeoIkaruGAF

Gold Member
Best news I've read all day, honestly

e0wdp8.jpeg
All of these are bad, honestly.
The fan translation delivers exactly as expected from you average weeb who studied some Japanese. It’s a bland, almost literal rendition of Japanese syntax that sounds uncanny even to someone who doesn’t understand English perfectly. Nobody says “against me as I am now” in English.
The machine translation does the job. It’s clear-cut, grammatically impeccable, and again, not a sentence you’ll find in any English script ever.
The localization is a disaster of self-indulgence, but toning it down a little and doing it in a single sentence would beat the previous two any day of the week.
 
All of these are bad, honestly.
The fan translation delivers exactly as expected from you average weeb who studied some Japanese. It’s a bland, almost literal rendition of Japanese syntax that sounds uncanny even to someone who doesn’t understand English perfectly. Nobody says “against me as I am now” in English.
The machine translation does the job. It’s clear-cut, grammatically impeccable, and again, not a sentence you’ll find in any English script ever.
The localization is a disaster of self-indulgence, but toning it down a little and doing it in a single sentence would beat the previous two any day of the week.
While you're correct that all three are lacking, if those are the choices I'm given I'll choose the fan translation over the MTL over the localization. Simply because I do not like when localizators believe themselves to be better writers than whoever wrote the work they're localizing.

For MTL, it's just a matter of some curation. I gave ChatGPT her personality outline according to the wiki and I was given this as a result.

"Taking down a useless old fool like you? Oh, that'll be easy. Quick, too. And believe me, I'll make sure you feel every second of it."

Which I believe beats the official TL work by any available metric.
 
Yea, people think AI will lead to everything being generic, but I think it'll have a lot of use cases. Like if AI could have a hi res scan of 100,000 trees and it can instantly give 10 games hundreds of trees that never repeat, can't even find the same tree in any of those 10 games. What's the problem again with using it?
It could have so many assets, it's far from generic. Especially if you still just do the stories 100% manually.

I'm sure at first it will be sort of generic depending on the uses but we are just seeing the start of AI. This is like looking at the Alpha of a game and saying it's crap and will never be good. AI will continue to grow and get better and better. Unfortunately it will lead to people losing their jobs but there's nothing we can really do about it anymore than people could stop the car industry from taking over the horse and carriage industry.
 

Thick Thighs Save Lives

NeoGAF's Physical Games Advocate Extraordinaire
I'm fine with this. If they go this way it will be better in the long run and we won't get agenda fueled English Translations. These LOLcalizers can go fuck off finally
LOLcalizers? That's a clever one dude! :p I wouldn't be too surprised if the Kai no Kiseki Clouded Leopard PC release includes some form of ML/AI translation for English when it eventually comes to Steam next year.
 

NeoIkaruGAF

Gold Member
For MTL, it's just a matter of some curation. I gave ChatGPT her personality outline according to the wiki and I was given this as a result.

"Taking down a useless old fool like you? Oh, that'll be easy. Quick, too. And believe me, I'll make sure you feel every second of it."

Which I believe beats the official TL work by any available metric.
I guarantee you, you’d still have a small army of weebs asking for your head on social media if you put that in the final game.
Some people are fiercely particular about translations from a language they don’t even understand. They’d rather have a completely literal translation than a good one.
 
Top Bottom