‘Being human helps’: despite rise of AI is there still hope for Europe’s translators? | AI (artificial intelligence)

IIn February 2022, literary translator Yoann Gentric decided he needed some relief while trying to translate US author Dana Spiotta’s novel Wayward into French. He was going to test whether artificial intelligence could put him out of business.

Gentric was struggling with a short wordless sentence that described the emotions the book’s protagonist feels when she opens a window: “Bright, crisp night air, invigorating.” He put the prompt into DeepL, a neural network-powered machine translation engine. regularly It outperforms Google Translate in accuracy assessments.

The suggested translation was reassuring considering job security: L’air de la nuit, lively and lively, lively lively (The crisp, vibrant night air was refreshing.) The AI ​​had translated the meaning of the sentence but apparently didn’t realize that the repetitions made the line nonsensical. It was of much poorer quality than his own translation, which was published in the book a year later: L’air pur et piquant de la nuit, live.

But when Gentric repeated the experiment this spring, he was less comforted by the result: L’air nocturne était vif, pure and livelyDeepL suggested this time. The online translator still lost the style of the sentence by adding a verb, but he learned to use three different words that had a musical connotation. “I don’t know if this is due to chance or the operation of a finely tuned algorithm, but night And pure “Not bad,” Gentric said.

Chatbots running on large language models (LLMs)—neural networks trained on large amounts of text to create natural-sounding language—are rapidly infiltrating every aspect of our work and leisure lives. But few professional sectors are being impacted by technology as rapidly as the translation sector in Europe, which is home to more than 200 languages ​​and a thriving technology sector.

According to a last joint survey According to a survey by French writers’ collective ADAGP and Société des Gens de Lettres, 79% of translators believe that the rise of artificial intelligence “threatens to change all or part of their work.” one in britain 2025 survey It found that 84% of translators questioned the expected lower demand for human translation, leading to lower pay.

These fears are future-oriented, but for many translators the nature of their work has already changed. Laura Radosh, a Berlin-based German-to-English translator, was receiving about four job offers a month from clients including universities, professors and museums. Last year, the number of offers dropped to one per month.

Most of these were “post-editing” jobs that required him to proofread text that had previously gone through a machine translation engine. “The final edit took as much time as translating from scratch,” Radosh said.

Post-editing, which is far less creatively rewarding than translation from scratch, also pays less: it is often paid per hour rather than per page or per book, and is paid at “unacceptable rates considering the work involved,” according to the French translators’ association. In Germany, publishers were found to offer typical rates of two to eight euros per page; This is one quarter of the average fee paid to translate a page from scratch.

But rates for regular technical translations have also fallen. “I got a job offer for 60 cents[s] a line,” Radosh said. “Before that, 80 cents[s] “It was the lowest rate I have ever encountered.”

Even before the advent of the Master’s degree, translation was a precarious profession: a recent study by the German translators’ association VdÜ found that the average income of literary translators (traditionally at the lowest-paid end of the sector) was: as little as 20,363 euros per year before tax. But recent changes in the industry mean that the numbers no longer add up for many translators; Radosh recently took a part-time job as an accountant for an NGO.

Marco Trombetti, co-founder and CEO of machine translation company Translated, said: “Without assistance, the human brain can basically produce translations of around 3,000 words per day. Beginners can manage around 1,500 words, the best translator in the world can manage 6,000 words, but the difference is not that big.”

He suggested that the cost of human translation has so far been determined by the number of neurons in the brain. “That’s about $100 billion,” Trombetti said. “But if we change this, we also change the unit economics of translation.”

But the pace of technological change is also revealing what human translators still do best. First, many machine translators still struggle with context. German-British academic publisher Springer Nature offers its authors the option of having their books automatically translated into other languages ​​for free, but despite assurances of subsequent “human checking”, the process has led to hilarious results in the past.

In 2024, Springer Nature machine translated an English book by a group of Indian academics into German. ‘Capital’ in the East: Reflections on Marx. inside chapter titlesbut the machine translator DeepL did not translate “capital” as follows: Capital in the intended sense, but HauptstadtIt means “capital city”.

A spokesperson for Springer Nature said in a statement: “Our AI-assisted translation is human-led and reviewed by professional editors. Errors like this are rare and regrettable, and this example relates to a limited pilot that has since ended.”

Jörn Cambreleng, director of Atlas, a French organization that promotes literary translation, said: “Machine translation is not creative. These systems are built to produce sentences that are generic, that have been said before, or sound like they have been said before. Whereas good human translators try to put into words something that has never been said before.”

One of the ironies of this upheaval is that literary translation now seems like a relatively safer career option than technical translation.

HarperCollins-owned Harlequin France brand confirmed it was working with Fluent Planet, a French communications agency, to produce translations created by AI software and then post-edited by humans; but such trial runs are for now confined to cheaper areas of the market: Harlequin’s books include A Mistress’s Confession and The Embrace of a Prince.

In Germany, where the total number of newly published books has been decreasing from year to year, literature in translation has performed quite well; In 2024, 8,765 translated books were published. accounts for a historically high 15% of total production. Marieke Heimburger, a Danish-German translator who heads the VdU, says authors are increasingly contractually obligated to their publishers not to use artificial intelligence in the translation process.

“AI can’t really do dialogue,” said Katy Derbyshire, a Berlin-based translator who has translated English-language novels by Clemens Meyer, Christa Wolf and others. “When you’re translating from scratch, you learn to understand the characters and their motivations, and you’re constantly adjusting them in your head to individual situations and also to the genre. The dialogue the AI ​​came up with didn’t fit the character description at all.”

He added that being human helps the translation process. “My body has experienced all the pain and joy that literature tries to convey. I understand how much someone can scream when they stub their toe on the bed frame, but an algorithm can’t do that.”

Fernando Prieto Ramos, of the Faculty of Translation and Interpreting at the University of Geneva, said his center noticed a decline in applications for translation courses three years ago, when the rise of generative artificial intelligence fueled excitement around machine translation. “But this trend is slowly reversing again with a more diversified educational offer,” he said.

Even people who develop machine translation software admit that there are tasks their products can’t achieve. “If I say Italian Solo three parole: not alone“Just three words: you are not alone,” said Trombetti, who founded Translated in 1999. “But you ended up with four words, not three. This is something machine translation is still struggling with.”

Heimburger said: “I’m not actually afraid of AI because I know it can’t do what I can do. What I’m afraid of is people who think AI can do my job.”