"Every week, our colleagues (translators) quit. It's a massacre." In November 2022, ChatGPT, OpenAI's conversational agent, was launched in the United States, marking the beginning of the artificial intelligence (AI) wave. While several studies have predicted the end of certain professions, the tsunami has indeed swept over the translation sector. And nearly three years later: the translation professionals we interviewed are unanimous. There has definitely been a before and after ChatGPT.
"Over the past two years, the deterioration of our working conditions has really accelerated, especially for (freelance) people like me who work for intermediaries, such as translation agencies," says Julie, who wished to remain anonymous. "We're no longer dealing with translation orders, but exclusively with post-editing," laments this freelancer who has been translating technical and marketing texts for several years.
For the uninitiated, "'Post-editing' is the word we used to use for generative AI," she explains: it involves working no longer exclusively on the original text, but on a first version "translated" by machine translation (NAT) or generative AI (GAI) tools. "Translation agencies have long been trying to force us to use these systems. And until 2023, we could still say no. But since ChatGPT, that's over," regrets the thirty-year-old.
A salary halved
"We've really seen a shift from the discourse of: 'it's inevitable, now you use machine translation/generative AI, and you'll be paid half as much'," confirms Laura, another freelancer, member of the translators' collective AI-lerte générale. Result: "My salary has been halved. Some clients have replaced me with Deepl" (a general machine translation platform, editor's note), emphasizes the woman who translates technical and marketing texts. The trend is seen in all translation professions, although some areas are more affected than others.
A survey conducted in March 2025 by ATAA, the association of audiovisual translators and adapters, and to which nearly 450 translators responded, shows a particularly marked decline in activity between the first quarter of 2024 and the beginning of this year. Of all the activities in the sector (dubbing, subtitling, voice-over, audio description, script and video game adaptation), voice-over and subtitling for the deaf and hard of hearing are increasingly being translated automatically or via AGI. The first dams to jump before the others?
"We asked a colleague, who had worked on adapting a foreign series into French, to take care of the new season, but for the first time, with post-editing. She said no, I don't do post-editing. She was told: "Understood, you won't be adapting this series anymore," reports Stéphanie Lenoir, translator, adapter, and president of ATAA.
Another study, this time conducted in April 2025 by the ATLF, the association of literary translators of France on the working conditions of professionals in the sector, and which will be published in early July, shows a real drop in the remuneration of publishing translators, "fluctuating between 30 and 64%," the organization explains.
"It's as if someone who knows nothing about plumbing were imposing a wrench on the plumbers."
For Laura, "there really is a battle that was lost or avoided, we don't really know. But we didn't subscribe to that, we don't agree with being presented with a fait accompli" – understand: being forced to go through post-editing, and therefore through generative AI (GAI), to translate a text. Faced with this upheaval, some have grouped together in collectives like IA-lerte générale, or En chair et en os. Their goal is to raise public awareness of the reality of AI's impact on their professions, but also of the reality of translation work.
It must be said that in this field, misunderstandings are legion, the translators we interviewed unanimously emphasize. Starting with the very idea of post-editing. Its premise is to say: these AGI and machine translation tools will save translators time by "pre-chewing" a first draft, Julie explains. The problem is that "these tools are imposed on us in 99% of cases by people who are not translators, and who will say: "you have ChatGPT, it's great"," she continues. For Valentine, another freelance professional who has been translating two foreign languages into French for about ten years, "it's as if someone who knows nothing about plumbing were imposing a wrench on the plumbers."
It must be said that at first glance, "if you take a text and put it into AI, the sentences that come out are, let's say, grammatical," she acknowledges. "It's not gibberish. People are impressed, they say to themselves: I pressed 'enter,' I have my translation, so what's the point of asking someone to do it, and paying them?" continues Julie.
"But translation isn't that. It's about meaning, and meaning isn't just about words," she adds. Words are not neat, distinct, and cut-out things. A word is an infinite number of possibilities, which depend on several cultural, emotional, unconscious, and interpersonal factors. What AI does is put words one after the other. And since it's probabilistic, from time to time, it works, but only on the surface," she believes.
Same story from Stéphanie Lenoir, an audiovisual translator and adapter, for whom: "artificial intelligence and machine translation tools consider language as a system for coding thought. Translating with AGI means going from code A to code B.But you can't reduce a text to a code,” she counters.
“In the end, we degrade ourselves”
However, if the translators interviewed took the time to describe to us what a (human) translation is, it was to explain to us to what extent having machine translation or generative AI tools imposed on them affects their work itself. Imagine: “I have the source text (in its original language) and I have the machine translation,” Julie explains. The problem is that “LLMs are not capable of doing all the analysis of the text. There are a lot of mistranslations, words for which the LLM will choose the wrong meaning,” not to mention “the repetitions, and the clumsiness of style,” she lists.
In other words, “I find myself correcting a text that is full of errors that I would not have made in advance, because I know there are plenty of pitfalls to avoid. But the machine doesn’t know that. And it’s extremely tiring to do that. I can’t even say that it’s the equivalent of redoing a beginner’s work, because even a beginner translator wouldn’t make such mistakes,” she continues.
“Having discussed this with other colleagues, redoing post-editing takes us as much time, if not more, than translating a text from scratch. We are immediately polluted by the suggestions of machine translation. And it's tiring to spend your time trying to get away from this first impression, which is riddled with mistakes and errors," Julie laments.
According to the ATLF study to be published next July, AGI and machine translation tools are rarely used by literary translators today. 93% of the professionals surveyed said they do not use such systems to translate their text.
But some, like Valentine, were economically forced to do so. The thirty-year-old explains that she spent several weeks doing post-editing, which was "not well paid and required a fast pace." Her observation: in the end, "I had lost all sense of quality, I had lost the compass inside me that told me: is this a good text or not? I said to myself: 'I have to stop right away.' We're actually degrading our skills, even without wanting to. We're degrading ourselves. And we can't write afterward," she laments.
While the freelancer currently has commissions that allow her to do without post-editing, she doesn't know how long this will last. "If I were completely committed to my principles, I wouldn't have a job," she admits. For Laura, who has chosen to refuse all post-editing, commissions have become rare. "If I wasn't in a relationship, I would have already gone to work at McDonald's," she confides, before adding: "Many people don't have a choice; those who have children or are single accept being paid half as much to pay the bills and eat." Security issues, theft, click workers... For the translators we interviewed, the deterioration in the quality of the texts, their work, and their salary is far from being the only problem. First, particularly sensitive or confidential texts can end up in the mill of AI tools, such as instructions for medication, defibrillators, or nuclear power. "We risk having gross errors that can be serious. "Besides the risk of making a fool of yourself for marketing texts, for anything medical, nuclear, or highly technical, people's lives are at stake. What are we waiting for, for something to happen for people to start reacting?", protests Laura. Some professionals interviewed, hired by translation agencies, even wonder if their end client really knows that their text has gone through "post-editing."
Another major problem: the entire system is based on "theft, disastrous consequences for the planet, and invisible click workers," laments Laura, who wonders: "ultimately, it doesn't bother people too much that our work has been stolen." Generative AI tools were trained on copyright-protected texts, without the permission of their authors and rights holders.
Authors struggle to oppose their refusal to use their works in these training phases. "When you think that before, we threatened teenagers who downloaded films," says Valentine, also a member of the AI-lerte générale collective, referring to Hadopi, which sent letters and sometimes issued fines in cases of illegal downloading.
But for Stéphanie Lenoir, who considers translators to be whistleblowers, affected on the front lines by generative AI, the problem is much deeper. "It's not so much because our professions will be the first to disappear. In our sector, we've been dealing with these issues for five, even ten years. We have perspective and are able to fully understand the abuses that this leads to. And the essential risk against which public authorities and the general public must mobilize is the standardization of our language at high speed, the standardization of thought and all modes of expression."
Because if our language system is put through a machine translation mill (and AGI), "we will always fall back into the mode of operation of the thought of the majority model (the Anglo-Saxon model, editor's note)," she explains. "It no longer allows for the expression of nuances, of divergent positions. (We see this with) the algorithm, designed to work on the most frequent occurrence. So it closes the field of discovery, it closes the field of expression, it closes the field of reflection," she insists.
At university, "we're already learning that the future of translation is post-editing. When they leave university, that's all young people do. So, in fact, they don't know how to translate. What scares me is that one day, no one will know how to translate, write for themselves, think for themselves," Valentine sighs. The battle isn't lost yet, however, Laura believes. "If I have a message for translators who are struggling and want to give up everything, it's: keep hope. The more of us there are who raise awareness, defend our profession, and speak out, the more impact we'll have. Translation is a profession that has existed for thousands of years. It's not possible for it to disappear like that, overnight."
0 Comments