Nicole Vulser has a horrifying piece in Le Monde (archived; en français) about what AI is doing to a valuable profession:
The artificial intelligence (AI) revolution has already claimed its first victims in the publishing world. Literary translators − the most fragile link and the most exposed to the AI tsunami − are witnessing their working conditions worsen by the day and their orders dwindle. As the use of automatic translation programs like DeepL becomes increasingly widespread, the job of a translator is increasingly reduced to post-editing contracts (using a text pre-translated by a machine).
According to Jörn Cambreleng, the director of Atlas, an organization promoting literary translation, this practice is still considered “shameful” among publishers, who never mention AI use on book covers, but also among translators, who accept this type of cheaper contract only due to a lack of other options.
The latest survey on machine translation and post-editing conducted by the French Literary Translators’ Association (ATLF) in December 2022 among 400 people already showed a “strong lack of transparency from publishers” on AI use and “lower compensation” (lower than average translation rates in 68% of cases). […]
ATLF secretary Peggy Rolland is concerned about the arrival of AI and fears a chain reaction of challenges, starting with legal ones. “Translators are authors and must receive royalties on each book sale (usually between 1% and 2%). “However, publishers who use AI want to pay us as self-employed contractors, which is not legal,” Rolland pointed out.
Unfortunately, I don’t see a way to fight this other than an unlikely multinational legal ban — translators are self-employed (as far as I know) and inherently hard to organize. But it’s a wretched development.
This is the sort of thing the EU can and does regulate, isn’t it? If they can tame Apple and Google and regulate appelations and such?
I hope so, but that will only help a smallish subset of translators. Maybe it will start the ball rolling, though…
Not to mention that literary translation is one of the most subtle and nuanced of the various type of translation. It is probably the least amenable to AI translation.
“Literary” here means something like “sold in a bookshop” as opposed to “assigned in a literature class”. Is it too optimistic to hope that computers translating Dan Brown, sports reports, and recipes will free up human translators for belles lettres? Yes. Yes it is.
Maybe AI will come for ghost writers next?
I live in between the worlds of literary and commercial translation, and post-editing is decimating the latter. In the former, for the moment, in my part of it—the Italian and American literary translation communities—it’s just making everyone extremely nervous. Literary translators in the US are, somewhat sadly, “helped” by the fact that translation is a negligible part of American publishing, and thus is seen as a niche market where churning things out is kind of beside the point. In Italy, though, translation is even more essential to the industry than it is in France. And translators are already well aware that many publishers, all too often, do not give a good goddamn about quality—otherwise they would be paying decently, valuing experience, giving people time to do their job right, and editing properly.
Translators can and do organize, though. I belong to the Italian union of translators who work in publishing (literary or otherwise), STRADE, which in turn belongs to CEATL, the European Council of Associations of Literary Translators. These groups are small but very active, and they’ve won some extremely important battles in the last few years. We’ll see. But I’m increasingly grateful that my main income is from art gobbledygook that even most humans have trouble with, and that I’m not at the beginning of my career.
Translators can and do organize, though.
That’s great to know; thanks for damping down my pessimism!
my main income is from art gobbledygook that even most humans have trouble with
Call me cynical, but isn’t that exactly where AI could be used without anyone noticing? Like in translating Zizek, Hegel or Heidegger? Noone can understand what they write anyway…
Let me throw out one maybe fruitful question. Usually a translation of a literary work is considered “original” enough that it’s entitled to its own copyright (although if the work being translated is not yet in the public domain you need permission, of course). Right now, the better argument (although the law could differ in different countries of course) is probably that a translation not created by a human being can’t be copyrighted. Is there enough incremental economic value to e.g. the Italian publisher from owning an independent copyright in the Italian version to incentivize that publisher to make sure they have a human translator?
@Hans: yeah, in my darker moments I’m not thrilled about having spent half my life sweating over essays that almost no one reads or understands, and I sometimes have to remind myself that even on the worst days it’s better than all the jobs I had in my youth. But I can assure you that curators and artists and editors often notice and argue over every last word. And some critics actually can write and have interesting things to say, thank God, it’s just that it takes years to get to the point where you’re able to pick and choose a little.
@mollymooly: I wish it worked that way. But honestly, that other stuff helps a lot of people make a living and develop important translation skills that can be devoted, among other things, to belles lettres. Take it all away, and only academics or independently wealthy folks will be able to translate (as is pretty much the case already in the US). Academics and independently wealthy folks can be great translators, mind you! But I think people like me also have something to contribute to the profession, and I would never have been able to stay in it were it not for a certain percentage of crap. Also, you learn a lot from translating recipes, and shoe catalogues, and guidebooks. Things that can be incredibly useful when translating fiction, for instance, since characters often eat food and wear shoes and go places.
Dan Brown was translated into Italian by the wonderful and much mourned Riccardo Valla, who was frustrated by the awfulness but also managed to have fun, and even wrote a parody of it later. I don’t think the problem was the job, the problem was that Italian translators don’t get royalties (there’s a law that should change this, but getting it implemented is a struggle). Had that been different, I’m betting he would have judged it time very well spent. Instead, with AI, the poor schmucks post-editing the Dan Browns of the future will be probably paid a third as much and still have to go over every word, in about two-thirds the time, with all the fun taken out.
@JWB: translations are derivative works that can only be published by whoever’s purchased the foreign rights to the original; unless you’re talking about a classic, it’s not as if anybody else can just machine translate and publish the same thing. And with classics, publishers want to be able to push them as an amazing new translation—otherwise, for cheaper editions, they’ll have someone dust off an old version a little. In Italy, translators always retain the moral rights to their work and are simply selling the commercial rights for ten or twenty years. If anything, many publishers would jump at the chance not only to pay less, but to not have to renew anyone’s contract if the book is still selling when it runs out. And to never be bugged about citing the translator’s name and so on.
Mollymooly: Maybe AI will come for ghost writers next?
The New York Times (Feb. 18 2024):
AI might be hitting hardest translators into European languages, who are working for publishers that are trying to maintain a for-profit business. In the anglophone world, so much foreign-language literature is translated through arts grants, and while the translator’s remuneration is still usually poverty-level, at least the cost of a real human translator is covered.
@Trond: aha, AI has somewhat misunderstood the “ghost” part
And to never be bugged about citing the translator’s name […]
Why don’t they want to cite a translator’s name?
Books — often riddled with gross grammatical and factual errors — are appearing for sale online soon after the death of well-known people.
Aren’t those prewritten anyway?
@D.O.: laziness, mostly. And sometimes, in the anglophone world, reluctance to highlight the fact that it’s a translation, if it’s not meant for that niche market. Crediting the translator somewhere in the book itself is pretty much a given, of course, although the degree of visibility varies. But publishers are also supposed to make sure that they’re credited in publicity materials and when lengthy extracts are reprinted elsewhere, and not all of them can be bothered. It really is disconcerting for a translator to see a few thousand words of their work show up in a newspaper with their name nowhere in sight, despite the little box specifying the price of the book and the number of pages. And it happens all the time.
A couple of random points.
For almost half of those surveyed, post-editing requires “more time than a conventional translation.
I don’t know if this is surprising or not. It suggests that a) translators are surprisingly fast at transforming text in language A into output in language B, or b) the AI-generated translations are surprisingly bad and require a lot of time and effort to rectify. If it takes more time to post-edit than to translate, wouldn’t it be better to just translate it and send it off? Would the publisher even notice? Would they come back with “We noticed that you didn’t follow the AI-generated translation we sent you and we frankly think it’s a bad translation”? Or would they accept it and publish it? (Of course, this would just normalise lower rates, but handing in a superior translation might get publishers to realise that their AI-generated tosh is going to result in inferior work.)
In fact, I find it hard to believe that AI could create any kind of acceptable literary prose. No matter how hard it tries, AI is only going to create “understandable” prose, certainly nothing that is likely to pass muster with readers of fiction. (Wait, I’m going to qualify that. I recently had occasion to browse a few modern fiction books in the local library. I can’t tell you how ungripping the prose was. I was bored by the opening parts of the opening chapters and couldn’t wait to put them down. Perhaps I’ve lost the knack of reading made-up stories and blatantly hard-boiled attempts to “draw me in” to some kind of fictional world. It didn’t seem worth my time.)
I actually have a lot of experience in doing this kind of post-editing, although not for literary translation. Faced with an English-language informational text that needed to be translated into Chinese, it was far easier to give it to Google Translate and then go through and fix the errors. It was certainly not sparkling Chinese prose (especially since I would have had considerable difficulty producing sparkling Chinese prose), but since it was merely for informational purposes that didn’t matter. (Doing it this way meant, unfortunately, that my ability to write Chinese never improved; you only develop that ability if you are translating from Chinese).
This reminds me of a previous example of technologically-induced sector-specific cataclysm in the humanities. I read years ago that when LPs were replaced by CDs, the number of people making a living designing album covers dropped by some huge percentage, as did the average income of the survivors.
Having only ~16% as much space to work with (I just measured) necessarily made album covers less interesting, therefore not worth the expense they had been worth.
Made sense to me.
If it takes more time to post-edit than to translate, wouldn’t it be better to just translate it and send it off?
Not if they’re paying you half as much, giving you half the time, and not crediting you. At that point, your standards change. And I have no trouble believing that in many cases achieving the same results could take even longer, because in my experience that’s certainly true a lot of the time when a non-native author has written something directly in English and asks you to clean it up. At least in post-editing there is an original and you don’t have to waste time puzzling out what they were trying to say. But still, the process of changing the syntax and half the words can be more complicated than if you just wrote them that way to begin with. Because, as you say, a machine can’t do acceptable literary prose. Even when it spits out a sentence that’s grammatical and technically means the same thing, with no obvious mistakes, it’s not thinking about style, voice, pace, intratextual references, humor, all the things that a translator actually spends time on. Even for books without enormous literary pretensions, if you want them to be an enjoyable read or at least not a slog.
Actually, it can be a lot harder to translate a so-so writer than a good one. Flaws get amplified, clichés no longer work the same way as filler, and you can’t simply trust that every strange choice was intentional, the way you can with a great author. And the so-so writers and books that were poorly edited in the original will be the first ones to get the post-editing treatment, if it becomes standard. It’s not as if publishers don’t know that human translators will produce a superior translation. They also know that copywriters, copyeditors, graphic designers, and translators – hell, even authors – will make better books if they’re not stressed by impossible deadlines and lack of job security and the humiliation of scrabbling for pennies without even knowing if those pennies will be paid on time or six months late. But publishers, especially the big conglomerates, are selling a product, and many are not concerned with making the best possible product.
Reminds me of self checkouts at supermarkets, which are neither cheaper nor faster than human cashiers, but took hold because Tech
handing in a superior translation might get publishers to realise that their AI-generated tosh is going to result in inferior work.
It’s not as if publishers don’t know that human translators will produce a superior translation. …. But publishers, especially the big conglomerates, are selling a product, and many are not concerned with making the best possible product.
It was a silly surmise. Viewed in that light, reading fiction for any kind of edification is starting to sound like an even greater waste of time and literacy than I was beginning to suspect it was. Crime fiction and sci-fi, mindless as they can be if they are badly done, are sounding like a more attractive option.
Reminds me of self checkouts at supermarkets, which are neither cheaper nor faster than human cashiers, but took hold because Tech
Don’t get me started…
I read a great deal of “crime and sci-fi” for edification, actually.
Is it safe to assume that Le Monde used AI to translate this article into English?
Speaking of undervaluing translators, this NYTimes opinion piece from 2022 brings up something I hadn’t realized:
AI’s not going to do that. (The quoted link goes to an interview with Hur, who translates Korean to English; there are quite a few more interviews with literary translators at the same site.)