HarperCollins Will Use AI To Translate Harlequin Romance Novels (404media.co)
- Reference: 0180527731
- News link: https://news.slashdot.org/story/26/01/06/1624238/harpercollins-will-use-ai-to-translate-harlequin-romance-novels
- Source link: https://www.404media.co/harpercollins-will-use-ai-to-translate-harlequin-romance-novels/
> Publisher's Weekly broke the news in English after French outlets reported on the story in December. According to a joint statement from French Association of Literary Translators (ATFL) and En Chair et en Os (In Flesh and Bone) -- an anti-AI activist group of French translators -- HarperCollins France has been contacting its translators to tell them they're being replaced with machines in 2026.
>
> The ATFL/ En Chair et en Os statement explained that HarperCollins France would use a third party company called Fluent Planet to run Harlequin romance novels through a machine translation system. The books would then be checked for errors and finalized by a team of freelancers. The ATFL and En Chair et en Os called on writers, book workers, and readers to refuse this machine translated future. They begged people to "reaffirm our unconditional commitment to human texts, created by human beings, in dignified working conditions."
[1] https://www.404media.co/harpercollins-will-use-ai-to-translate-harlequin-romance-novels/
Oh. (Score:2)
I don't bother with romance novels (they're usually about abusers being rewarded for being abusers, and not really my cup of tea even when they aren't), but AI is not great at translation, is terrible at metaphor, and is horrific at writing.
If they're going to use AI for auto-translation, then I think the best thing they can do is pay for the first 30 sessions of therapy needed afterwards.
Re: (Score:2)
This was a case a couple of years ago.
About a year ago, LLMs got really good at translations. They get most metaphors, and they're decent at writing.
Lately, it's probably on par with average human translator and way better than best "localizers". This is why you now find a metric shit ton of Japanese and Chinese games on Steam, and you can track translation quality vs time scale and LLM used. ChatGPT v3 were meh. V4 was ok. V5 is getting pretty good. And that's the general model for very different languages
Weird to be doing it in France (Score:2)
They're usually pretty uptight about language. Romance novels are basically girls version of porn though so I doubt the context and writing matter all that much.
Still sucks that it's another career path that's basically gone now.
Re: (Score:2)
I man with Google translate being like 20 years old ... it's nothing translators weren't warned about. I think in 2019 Google started using LLM for translate and the translation (and grammar checking) services have seen people leaving since ChatGPT got popular.
There is real value in translators (Score:3)
I wouldn't call myself bilingual, but my experience is that there are lots of nuances that an author imbues their work with, based on choosing various turns of phrase. Learning to see, understand and translate the author's intent is a learned skill. I question whether AI is yet capable of discerning the author's intent.
Re: (Score:2)
Last versions of ChatGPT are pretty good at it even for Japanese and Chinese to English translations.
It also is worth mentioning that ChatGPT V4 was so good at generating "women's romance novels" on the fly with the AI boyfriends, that when V5 came out and was way less flowery and affirming in its language, there was a massive movement of women to bring V4 back. They got it too.
I.e. V4 was very, very good at not just translating, but writing this sort of stuff from the scratch already.
and--enternity in an hour (Score:3)
Oh, here we go. Honestly, romance books were already promoting unrealistic expectations among women for em-dash length and frequency.
Re: (Score:1)
these novels and soap operas are porn for women, they get wetter than than otter's pocket reading them.
So they want to replace human slop with ai slop.
AI translation is bad (Score:2)
My work uses AI to transcribe meetings. It has frequent mistakes, which nobody notices because nobody that attended the meeting looks at the minutes immediately after the meeting.
And we see similar problems with translation software. It makes an error, and with no human willing to proofread it, errors slip in. And it's not like a simple typo, it can be totally strange nonsense.
When you use AI to drive both the cost and quality of your products to zero. Don't be surprised if consumers stop showing up to buy
Re: (Score:2)
That is why they will employ a proof reader as the story mentions.
Re: (Score:2)
I covered that in my post. Won't work.
Cuts out the middlemen (Score:3)
A lot of the time when you try to get "professionals" translate stuff, you get machine translation anyway. Had been a problem for more than a decade. It's all pretranslation and translation memory with the CAT tools. Even with the same agencies you can get different quality results from one day to the next. And I don't even blame them. Most stuff that gets translated is boring and repetitive. It makes complete sense to do a first quick pass with a machine and then do a lot of QA and polish to make it good. Style guides, character profiles, plot notes. I think it's all mostly to skip the step where you pay someone a lot to pretend to do translation by hand. Copyright might be an interesting angle though. I believe book translations run a separate copyright from the original. People got in trouble over using a still copyrighted translation, assuming it's fine because original is public domain. What machine translation does here might be interesting. Certainly no less "transformative" than all the AI companies stealing content to train their models.
I've found AI to be good at translation (Score:4, Informative)
Not only is AI pretty good at translation, but romance novels don't require exceptional translations because they are written at a junior high school level, so that the kind of people who buy them can read them. (Americans, that is.)
Seriously though, they really aren't complex, so AI translation will work fine. It's not like you're talking about poetry, it's just soft core porn with a stupid story.
The sort of people that read Harlequin books (Score:2)
I don't think they'll notice. The books are effectively machine generated anyways.
this calls for a bodice riot (Score:2)
En Chair et en Os agents should start riots, flip cars, set them on fire. Wearing ripped bodices, of course.
Re: (Score:1)
The bodices will be ripped by a muscular beefcake in 1940s stylish business attire named Brett. The start of rioting will commence with Brett ripping the bodices saying "Oh Hillary!" to which the bodice rippie will respond, "On Brett!"
What about lost context? (Score:2)
So I don't read romance novels... but I know enough to know that you can't just do a word-for-word translation in many cases. Granted both Google's and Apple's real time translation apps are pretty good. But even they make mistakes when doing literal translations vs what is meant in context. So unless they plan to keep a few people around to proof-read the translations and provide contextual translation updates, I see this being prone to challenges... the kind of challenges where people read them and quote
Re: (Score:2)
> So unless they plan to keep a few people around to proof-read the translations and provide contextual translation updates, I see this being prone to challenges
The summary says the translations will be checked and finalized by humans.
Hell, next year they'll write a story framing the human part as a big scandal proving that the AI is dumb and the company is trying to trick everybody into being impressed its technology when it's actually part mechanical turk.
Re: (Score:2)
One open is simply proofreading.
If you want to automate more, you can try multiple things. First, modern LLM have quite large context, so your problem is only if some in-joke from long before gets lost. One can already see this in human translated media. I've seen for example horrible translations of jokes from the Simpsons. Second, you could try doing several passes. First process it chapter by chapter letting the agent create a markdown document containing summaries, persons, in-jokes, etc. and then run t