So… O’Reilly sent me email today hyping up how my books (really, just the one, I assume) is going to be AI-translated into Spanish and German, with other languages to follow. This was probably inevitable, but I still have concerns.
First: are there no human translators of these languages?
Second: who’s going to proof-read all 1,126 pages to make sure nothing got botched, especially given the technical nature of the content? The readers? Which isn’t even crowd-sourcing: it’s customer-sourcing.
@Meyerweb are you allowed to say "no"? Or does the contract say they can do that without your approval?
@subtl I’m pretty sure translation rights (specifically, the rights to sell translation rights) rest with the publisher, which made sense when I first signed a contract 25 years ago.
@Lapizistik @Meyerweb @subtl good point: who will the translated book be attributed to? For transparency, should AI be listed as a co-author?
@Meyerweb Human translators cost money. AI can do a bad job for more money (though it's probably OpenAI's money or something... so less money to the publisher), so why not?
Do you give up your control over translations when you sign the publishing deal?
@mathaetaes I’m pretty sure translation rights (specifically, the rights to sell translation rights) rest with the publisher, which made sense when I first signed a contract 25 years ago.
@Meyerweb Bummer. Hopefully you at least get some royalties out of whatever they end up selling, since it may or may not be a good representation of your writing or technical ability, depending on how bad the AI fubars the situation, and any blame for that is unlikely to be properly attributed to the publisher. :(
@Meyerweb @mathaetaes So they can *sell* translated copies, but I wouldn't read that as having rights to *do* the translating. My instinct would be to push back and insist on human translators, but what do I know?!
@bytebro @Meyerweb When my aunt's book was translated, the publisher called to let her know... not to ask... so that's another data point that seems to point to not having rights.
At the time my aunt was stoked - it was a big deal to have a book worth translating. This was pre-AI though, so human translators did the work... which has its own complexities to it (like, what happens if the human translators just suck - her name is still on the book, etc..)
@mathaetaes @Meyerweb Good points!
@Meyerweb We’ve done a lot of work on AI-assisted translation at U.S. Digital Response for the past year. It can make a good first draft for a translator, but it is in no way a replacement for a human translator.
@Meyerweb can you refuse?
@fabienmarry I believe translation rights rest with the publisher, which made sense when I first signed a contract 25 years ago, but maybe not as much now.
@Meyerweb I wonder if it could be argued that ai translation and human translation are different things, and that you couldn’t possibly have agreed to the former 25y ago.
(jgg) I barely buy technical books nowadays, but I have always prefered O'Reilly's because they had wonderful Spanish translations.
It is really sad to see O'Reilly doesn't care about quality anymore.
@Meyerweb That’s unsurprising, but still disappointing to see from O’Reilly.
This reminds me of @wordridden’s explanation from years back, of the very creative, very human work that goes into translation.
https://wordridden.com/post/720
So I would add a third question: how much of the authors’ personality is going to be ground away and lost, even if the translations are technically correct?
@Meyerweb won’t it just translate all the CSS properties and values? Rendering it completely useless.
@dletorey ¯\_(ツ)_/¯
@adamrice @dletorey @Meyerweb I expect this gets complicated though, since you'd want comments in the source translated, except for where the comments mention CSS properties.
That said, as much as I hate the current climate-and-logic destroying AI hype cycle, LLMs have taken Searl's Chinese Room further than I thought possible. The real problem is if there's no humans checking the work.
And there might be, they might hide that to build their "AI" cred. Sucks that we can't trust anyone.
@Meyerweb reading that AI was going to translate a book, I'd presume the workflow ends with people editing it.
Did the announcement suggest otherwise? Seems reasonable to pipe a book into some Spanishification function to save some amount of manual labor typing... then let people focus on the more interesting task of proof reading.
Too naive?
@RyanParsley Probably too naïve.
@RyanParsley @Meyerweb oops, did you misspell "Spanishittification"? 
@Meyerweb I was the one who proof read your first or second edition's translation in German.
It was not that easy for this trained translator, he needed my help for not so clear cases and when he lacked understanding. But exactly this guy has translated some books for O'Reilly. He knows the topics, I guess.
I don't think that the outcome will be very good. Proof reading is in my view necessary.
@Meyerweb how extremely disappointing. And you’re not the only person I’ve seen saying this. I’m not competent enough at non-English languages to say whether having a computer-generated and possibly completely misleading translation of an Eric book is better than no translation at all, but… this feels like a very worrying thing to normalise. I’m sure it’s cheaper for the publisher, even given the disastrous energy, water, and money cost of doing it, but nonetheless… :-(
First: there are, but they cost more and take more time.
Second: machine translation should really be accompanied by a post-edit step by a human.
@Meyerweb That is extremely no bueno or nicht gut depending on the book/translation....
I mean come on O'Reilly .... that's very disheartening. I would think the very technical nature of some subjects would be why you would want a human translator fluent in those subjects.
I mean I took a couple years of German in high school and I wouldn't think of trying to translate anything....
@Meyerweb
You realize, of course, OR is simply using translation as an excuse to steal, without scraping, your IP?
#badcapitalism
@Meyerweb Speaking as a technical translator, I sympathize. I encounter stuff I don’t understand in depth often enough (please don’t ask me to explain fast Fourier transforms), and I can flag the parts I’m unsure about for follow-up with the author or publisher.
I have to admit that machine translation has gotten surprisingly good (setting aside problems of consistent authorial voice and stuff like that), but it’s not always right, and it never knows when it’s wrong.
@adamrice @Meyerweb yikes: "AI" as "smart" machines with Dunning-Kruger effect ... making mistakes much, much faster than humans could 
https://en.m.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect
@Meyerweb O reilly translating books into spanish with AI??? Interesting, I will spread the word, and start boycotting O Reilly. Also will make sure no one I know buys anything from them.
Yes, there is Spanish translators, or were, because the AI has left the majority without a job.
If they don,t pay translators, we are not paying for their books anymore.
@Meyerweb I put a translation approval clause into all my contracts in 1996. I suggest all authors put in such a clause on their side. I know O'Reilly doesn't adjust paper...which is why I published with Wiley and Addison-Wesley.
I did have to scramble to find a japanese geek to read Java Security as a sanity check in 1997.
@Meyerweb I am hoping they send out copies to native language speakers to "proof"... would be interested in finding out if they do..
@Meyerweb I used to consider O'Reilly the top tech publisher, where one could confidently buy their books sight unseen. I think they lost the plot a while ago, this probably is the icing on the cake.
I'm sorry they are trying to do that to you.
@Meyerweb when are they publishing the AI generated sequel?
@Meyerweb As an author I’m partially horrified, but one of the things we saw with Google translate was that even bad translation was helpful to a lot of folks. Maybe with a clear label, it’s a win?
@Meyerweb To be clear, I shudder at the thought of it being done to my books.
@Meyerweb Wish I could say I was surprised.
@Meyerweb you have away the rights for derivative works ? I don’t think my publisher would have been able to do this without my consent
@Meyerweb don’t worry multiple AIs will proofread it ;)
@Meyerweb I had a bit of a whinge on LinkedIn about this. Very disappointed indeed. https://www.linkedin.com/posts/parisba_publications-activity-7249244992496361472-4pLj?utm_source=share&utm_medium=member_ios
𝚂𝗆𝖺𝗋𝗍𝗆𝖺𝗇 𝙰𝗉𝗉𝗌
@Meyerweb @sandyarmstrong
And even human translators aren't necessarily all that great. The manual for my fridge, which I bought many years ago, well before AI, says "do not store documents inside the fridge". Wait what?! Why on Earth would anyone be putting documents into a fridge, and what on Earth was the actual original meaning that somehow got turned into "document" by the translator??
@Meyerweb even if they have the translation rights I hope you can sue them for image damages when the translation turns out like shit
@Meyerweb asking people to pay for a product that nobody bothered to develop seems like a good way to ensure people avoid paying for it in the future
@Meyerweb will they compare with the original version?
@Meyerweb why specify ai ? all machine translations are ai. did they get hit by the current ai hype too ?
@Meyerweb I wonder what the environmental cost of using power-hungry GPUs to achieve a passable translation is? It can’t ever just be about money (although I also wonder if the true financial cost is not as far from paying a human given that currently companies like OpenAI are operating at huge losses to get a foothold)
@Meyerweb That seems like a terrible idea, I would expect machine translation + human post-edit to cut costs.
Also, are they going to only provide electronic versions, which could be technically fixed quickly (not to the advantage of the reader though)? The risk of printing poor or factually wrong text is way too high to make sense economically.
@Meyerweb I didn't expect that even O'Reilly would go this way.
Will it be placed on noticable page like back cover that it has been translated by a machine?
@Meyerweb works created by a human are protected by copyright. Works created by not-a-human (including AI) are not. But I don't think I understand what the copyright regime of derivative work created by not-a-human is. Your original text will still be copyrighted, but somehow the translation will not? Doesn't seem right. But the translation, not being the creative work of a human, cannot be. So what gives? 
@Meyerweb We have machine translations for web.dev and developer.chrome.com and the results are often really bad.
The Dutch translation of my recent “Animate to height: auto in CSS” article translates back into something along the lines of “Animate on altitude: automatic with CSS” 
@sebastianlaube @Meyerweb IIIRC that recently changed and your language choice is now persisted.
(Looking at @tunetheweb to confirm)