Human fallout may include being replaced by LLMs, diminished skills, and fewer career options for all but the elite scholars.
Turkiye’s sophisticated translators are moonlighting as trainers of artificial intelligence models, even as their profession shrinks with the rise of machine translations. As the models improve, these training jobs, too, may disappear.
“Translation has become a job for a limited number of translators who are really good at it. Machines translate the rest,” Mehmet Şahin, head of the translation department at Boğaziçi University in Istanbul, told Rest of World. Translators will become less skilled in the future as AI eats up entry-level work, he said.
As a teenager, Pelin Türkmen dreamed of becoming an interpreter, translating English into Turkish, and vice versa, in real time. She imagined jet-setting around the world with diplomats and scholars, and participating in history-making events.
Her tasks one recent January morning didn’t figure in her dreams. The 28-year-old translator’s computer displayed a dashboard for AI training provided by Outlier, a San Francisco-based company that hires contractors to train large language models. Outlier’s clients include OpenAI,Microsoft, and Meta, among others.
The dashboard displayed a prompt for training an LLM powering either ChatGPT, Gemini, or Preplexity AI.
The AI chatbot provided three menus. Türkmen rated them on accuracy, ethics, and relevance. She checked for grammar, fluency, tone, and structure. She looked for language that would reveal an AI author — phrases like “As an AI assistant …” or “Certainly!” She explained her reasoning so that the machine could learn.
Türkmen has earned several thousand dollars over the past nine months training AI. In the tech industry, this kind of work is known as reinforcement learning from human feedback, and it helps LLMs respond with intuition and context.
“The answers they [LLMs] provide to Turkish prompts often sound machine-like,” Zeynep Kırıcı, another translator who contracts for Outlier, told Rest of World. “We’re trying to fine-tune the mechanical tone of AI. We’re trying to humanize it.”
The job offers Türkmen a reliable income at a time when her career path looks like it’s disappearing, she said.
Translators and scholars in Turkiye told Rest of World the nature of translation work has changed. Before AI, young translators worked in translation bureaus tackling everything from administrative documents to trade reports and literary classics.
Today, most entry-level positions involve editing AI-generated content as a machine translation “post editor,” they said. Others involve training AI.
The shift is profound in Turkiye, where translators have helped shape the nation’s culture and politics. After Turkiye became a European-style republic a century ago, a government-sponsored translation bureau produced thousands of Western classics between 1940 and 1966, helping spread European ideas in society. Translators are highly esteemed and influential, and there were about 10,000 translators and 2,900 interpreters in 2015, according to the most recent data available from the Turkish government.
Book publishing, too, is transforming. Turkish publisher Dedalus announced in 2023 that it had machine-translated nine books. In 2022, Agora Books, helmed by translator Osman Akınhay, released a Turkish edition of Jean-Dominique Brierre’s Milan Kundera, une vie d’écrivain, a biography of the Czech-French novelist Milan Kundera. Akınhay, who does not know French, used Google Translate to help him in the translation, to much criticism from the industry.
“Nobody criticizes doctors who use Google Translate when talking to their patients,” Akınhay told Rest of World. “Nobody complains while using self-checkout desks in supermarkets.”
The trends are global. In November, the Netherlands’ largest publisher announced it would use AI to translate certain books, with humans only involved in light edits. In the U.K., over a third of 12,500 translators said they have lost work to machines.
“Frankly, we are at a stage where the profession of translator has lost its immunity,” Akınhay said.
Outlier and its parent company, Scale AI, did not respond to a request for comment about their AI training operations in Turkiye.
The new roles require much less skill and effort than translation, Türkmen said. For instance, she spent a year on her master’s thesis studying Samuel Beckett’s self-translation of his play Endgame from French to English. More recently, for her Ph.D. in translation studies, she studied for more than two years about the anti-feminist discourse in the Turkish translation of French author Pierre Loti’s 1906 novel, Les Désenchantées.
In contrast, working on an AI prompt takes about 20 minutes.
For every hour Türkmen works, she gets $20. That’s as much as a translator earns after spending roughly four hours converting two pages of densely written text into Turkish, according to Çevbir, the association of Turkish translators.
Outlier pays weekly, which is crucial as the Turkish lira fluctuates wildly. For book translations, workers are compensated only after publication — it can take months during which the payment can lose value.
Globally, Outlier has a mixed reputation and Scale AI faces a class action lawsuit in the U.S. alleging non-payment by contractors. The company has responded to the suit, saying it complies with all laws and regulations and has safeguards to protect its workers.
Erdem Hürer, 32, a graduate student in Boğaziçi University’s translation department and an Outlier contractor, told Rest of World he’d prefer to make his money entirely through AI training, while doing translations as a labor of love.
“Translation is an arduous job with so many details and labor involved that it drains away your time,” he said.
The contractors said they feared that AI training jobs, which are plentiful now, will disappear once LLMs improve.
But that future is a few years away for Turkish-language LLMs, which don’t have much training data and rely on humans for fine-tuning, said Şahin.
Yiğit Bener, former director of the Conference Interpreters Association of Turkiye, believes AI will catch up and displace human interpreters in all except the most sensitive diplomatic meetings.
Bener, 66, recounted live-translating a speech by the late Cuban President Fidel Castro when he visited Istanbul in 1996 for a United Nations conference. Castro spoke in a roaring tone, Bener recalled, then wondered whether AI could handle such a challenging meeting.
He answered his own question.
“In a year or two, as AI will offer flawless simultaneous translations, mediocre translators will be eliminated from the market,” he said. “And that is a good thing.”
Source: restofworld