No, using machine translation as a replacement for paid or volunteer localizers is not an ethical use of machine translation.

Someone who actually knows both languages can use machine translation as a first step, but it is never okay to pass off machine translated text as something someone wrote.

Oprindeligt skrevet af rawWwRrr:

The use of an AI to produce commercial content requires payment. We are able to use AI to provide translation for free because we're doing so for personal use. The moment you acquire the generated content to make money, that generator deserves a share of that profit.

That's not the question they asked. They're asking about localizing a game during its development.

Oprindeligt skrevet af Ben Lubar:

Oprindeligt skrevet af rawWwRrr:

The use of an AI to produce commercial content requires payment. We are able to use AI to provide translation for free because we're doing so for personal use. The moment you acquire the generated content to make money, that generator deserves a share of that profit.

That's not the question they asked. They're asking about localizing a game during its development.

No but there's an assumption in the question that a proper translator can't be hired due to cost. Using AI instead isn't a free option either. There's going to be a cost regardless.

Oprindeligt skrevet af Ben Lubar:

No, using machine translation as a replacement for paid or volunteer localizers is not an ethical use of machine translation.

Someone who actually knows both languages can use machine translation as a first step, but it is never okay to pass off machine translated text as something someone wrote.

In this scenario it is disclosed (hence me writing “expressly telling people”) to the consumer and the “first step” is being made available until a proper translation is paid for. Therefore it would not be passing off machine translated text as something someone wrote.

It is improbable for a single developer to know every language asked for.

Oprindeligt skrevet af rawWwRrr:

The use of an AI to produce commercial content requires payment. We are able to use AI to provide translation for free because we're doing so for personal use. The moment you acquire the generated content to make money, that generator deserves a share of that profit.

I see. That would complicate the matter. Thank you.

Oprindeligt skrevet af Sciencemile:

Would you consider this an ethical use of AI?

Professional translators are already using LLM generated translations for a first pass. This is simply the next evolution from older forms of machine translation which have been in use since the 70s/80s.

Oprindeligt skrevet af Sciencemile:

Would a LLM-quality translation of a game turn you off personally, even if you know that it's just the "early access" version?

Depending on language pair and other specifics, the quality of a first pass LLM translation will range from unreadable to nearly perfect.

Oprindeligt skrevet af Sciencemile:

Do you think this would increase sales in those linguistic regions?

Again, it all depends on specifics. Are we talking a major publisher with an established localisation pipeline, or an indie with a budget that is akin to a packet of crisps and a pint?

Oprindeligt skrevet af Sciencemile:

How much does it cost to hire a translator? Is cost not actually the major factor for an indie game to have a given translation?

I can only speak for Japanese - English, as that's my field. Here's the rough industry standard for estimating cost:

Source character divided by 2.5 gives you rough output word count.
Source word multiplied by 2.5 gives you rough output character count.
While billing could be based on source or output, the general standard is source.

So, let's say a game contains 1,000,000 Japanese characters, that's an estimated 400,000 English words. Rates vary wildly. But the bottom of the scale is between 0.80 to 1 JPY per source character or 1.5 to 2 per output word. For this example, we'll assume 1 JPY (source) and 2 JPY (output). If you bill per source character, that's 1,000,000 JPY. If by output, it's 800,000 JPY.

Toward the high end, rates can reach around 15 JPY per source character or 30 JPY per output word. In that extreme, the above would bill at 15,000,000 JPY (source) or 12,000,000 JPY (output). But that's an extreme that is fast fading with the wider adoption of machine translation and post editing.

Most J > E translations will be billed at the low end, so expect 0.80 JPY as the base pay per source character. As a more concrete example, the original Steins;Gate visual novel contains approximately 569,763 characters. With today's lowest rates and the industry standard of billing by source, that would cost around 455,810 JPY (approx 2,914 USD) to translate, minimum.

Keep in mind also that you pretty much get what you pay for. Higher rate + plenty of time = a high quality translation. The industry trend of lower rate + short deadline = poorer translation. There is no such thing as cheap, quick, perfect translation. You can only ever have two of the three.

Oprindeligt skrevet af Chika Ogiue:

Again, it all depends on specifics. Are we talking a major publisher with an established localisation pipeline, or an indie with a budget that is akin to a packet of crisps and a pint?

It becomes more cut and dry the bigger they are so I was thinking mostly a small dev team with a game that blows up. They can guarantee a proper translation eventually but wanted to provide something as quickly as possible to the people asking for it and so are tempted to, with disclosure, stop-gap with a machine translation in the interim.

Oprindeligt skrevet af Sciencemile:

How much does it cost to hire a translator? Is cost not actually the major factor for an indie game to have a given translation?

I can only speak for Japanese - English, as that's my field. Here's the rough industry standard for estimating cost:

Source character divided by 2.5 gives you rough output word count.
Source word multiplied by 2.5 gives you rough output character count.
While billing could be based on source or output, the general standard is source.

So, let's say a game contains 1,000,000 Japanese characters, that's an estimated 400,000 English words. Rates vary wildly. But the bottom of the scale is between 0.80 to 1 JPY per source character or 1.5 to 2 per output word. For this example, we'll assume 1 JPY (source) and 2 JPY (output). If you bill per source character, that's 1,000,000 JPY. If by output, it's 800,000 JPY.

Toward the high end, rates can reach around 15 JPY per source character or 30 JPY per output word. In that extreme, the above would bill at 15,000,000 JPY (source) or 12,000,000 JPY (output). But that's an extreme that is fast fading, with the wider adoption of machine translation and post editing.

Most J > E translations will be billed at the low end, so expect 0.80 JPY as the base pay per source character. As a more concrete example, the original Steins;Gate visual novel contains approximately 569,763 character. With today's lowest rates and the industry standard of billing by source, that would cost around 455,810 JPY (approx 2,914 USD) to translate, minimum.

Keep in mind also that you pretty much get what you pay for. Higher rate + plenty of time = a high quality translation. The industry trend of lower rate + short deadline = poorer translation. There is no such thing as cheap, quick, perfect translation. You can only ever have two of the three.

It seems based on the fact that the official translators are already doing this at least cost-wise there is less utility to this possibility than I thought, especially since they would have more experience even with the machine translation.

Oprindeligt skrevet af Sciencemile:

Oprindeligt skrevet af Ben Lubar:

No, using machine translation as a replacement for paid or volunteer localizers is not an ethical use of machine translation.

Someone who actually knows both languages can use machine translation as a first step, but it is never okay to pass off machine translated text as something someone wrote.

In this scenario it is disclosed (hence me writing “expressly telling people”) to the consumer and the “first step” is being made available until a proper translation is paid for. Therefore it would not be passing off machine translated text as something someone wrote.

It is improbable for a single developer to know every language asked for.

It would be much better to simply not offer something you are not capable of supplying to your customers.

If you don't have access to a French localizer, your game will simply not have a French localization until you find one.

In many cases, community members are willing to do volunteer localization, especially if you make your game moddable, and if your game has a community in the first place.

Oprindeligt skrevet af Sciencemile:

I was thinking mostly a small dev team with a game that blows up. They can guarantee a proper translation eventually but wanted to provide something as quickly as possible to the people asking for it and so are tempted to, with disclosure, stop-gap with a machine translation in the interim.

It would still depend. If the AI generated translation wasn't post-edited by a professional translator, it carries a huge risk of being an absolute disaster. And it can be hard to recover from that even if the translation is fixed later. My personal opinion is that if a developer is serious about providing a translation, they should do it properly from the get-go.

Oprindeligt skrevet af Chika Ogiue:

Oprindeligt skrevet af Sciencemile:

I was thinking mostly a small dev team with a game that blows up. They can guarantee a proper translation eventually but wanted to provide something as quickly as possible to the people asking for it and so are tempted to, with disclosure, stop-gap with a machine translation in the interim.

It would still depend. If the AI generated translation wasn't post-edited by a professional translator, it carries a huge risk of being an absolute disaster.

...to the point where running AI generated translations repeatedly in order to intentionally make an absolute disaster is a popular genre of YouTube video and video game mod.

Sounds like between Ben, Rawr, and Chika this thought has been cross-examined very satisfactorily. Looks like it is a bad idea.

Thank you Chika for the figures on English to Japanese translation costs.