The etymology of the word translation—“to carry across”—conjures an image of physical labor. It is deeply relational, requiring at least two bodies, those of an author and of the person who carries the author’s words to a previously unvisited place. Let’s say we removed the laborer and replaced that person with a car. Or a train. Suddenly there is a feeling of weight lifted, certainly ease, and perhaps a little relief. But the intimacy of the earlier image no longer holds. Whether this matters has been the subject of recent debate as some publishers consider using machines to replace human translators and what that decision might mean for an ancient art.
In November, Dutch publisher Veen Bosch & Keuning (VBK), a subsidiary of Simon & Schuster, announced that it would trial the use of artificial intelligence (AI) “to assist the translation of a limited number of books.” Reactions rose in a flurry: Writers, publishers, and translators contended that AI would produce “bland” work. They lamented the possibility of lost jobs. The European Council of Literary Translators’ Associations resisted the standardization of an idiosyncratic process, stating that the imagination, understanding, and creativity that translation demands are “intrinsically human.”
VBK’s decision to incorporate AI into the editorial process may shock some but is not unprecedented. With a broad range of AI tools now available on the market, an increasing number of writers and publishers have turned to large language models (LLMs) to assist in, or contribute to, the production of creative work. As of February 2023, there were more than two hundred e-books in Amazon’s Kindle store that listed ChatGPT as an author or coauthor, according to Reuters. Maverick publishers like Spines, although small players in the global book market, plan to publish thousands of AI-generated books next year.
AI isn’t new to translation either. Literary translators sometimes input segments of their source text into AI-based technologies like Google Translate and DeepL to generate ideas for particularly thorny passages. But these tools have to be used “very carefully,” warns Seattle-based Finnish-to-English translator Lola Rogers, “because the translations it produces are error-ridden and devoid of flow or beauty.” Edward Tian, a cofounder of AI-detecting start-up GPTZero, adds that current LLMs not only do “a mediocre job at translations,” but also reflect the “majority white, English-dominated” nature of their source texts. Reiterating such worldviews and their biases runs contrary to the aim of much literary translation: to expose audiences to new perspectives. And Rogers, who was recently commissioned to use a translation tool to expedite a months-long translation process to five or six weeks, says that from her brief experiments, the time saved with machine assistance was “minimal.” French-to-English translator Louise Rogers Lalaurie shared a similarly underwhelmed account of editing poor machine-led translations.
So what’s the threat?
One area where translators are feeling the pinch is in creating samples, book excerpts translated to give general impressions of a text to potential publishers. Some publishers have been considering using AI to do this work instead. Though she is unsure whether this is because samples are being automated, Rogers says, “The number of samples I’m asked to translate has fallen precipitously in the past couple of years, making it much harder to earn a living.” A 2024 survey of Society of Authors members found that over a third of translators have lost work due to generative AI. Close to half of translators surveyed said that income from their work has decreased.
To illustrate how AI might ease the time and cost pressures inherent to translation from a publisher’s perspective, Ilan Stavans, the publisher and cofounder of Restless Books, an independent press in Amherst, Massachusetts, gives the example of a recently acquired eight-hundred-page book. To translate it, “substantial investment” would be necessary: Not only are “first-rate translators” for the source language scarce, he says, but the project would also require at least two years of dedicated work. By the time the book is translated and published, the demand the publisher once saw for the title might easily have changed. Meanwhile, the publisher would have incurred a cost much greater than if it had used LLMs, the most expensive of which—such as the premium version of ChatGPT, which costs $200 a month—is a tenth of the average cost of publishing a translation.
“It would be fast and easy,” Stavans admits, “but it would not be the right move.” Though Stavans is enthusiastic about AI’s potential and sees the value of using AI to translate samples, he emphasizes that he would never condone translating an entire book using a machine. The key to the heart of translation is “that intimate, subjective relationship between a text and the translator,” he says—the nebulous yet nonetheless living connection that translator Kate Briggs describes as the “uniquely relational, lived-out practice” of “this little art.”
Will Evans, founder of independent publisher Deep Vellum in Dallas, does not see a future in which machine-led translations supersede the human. “I do not believe AI-led translation will be competitive for works of the literary caliber we are interested in any time before the AI bubble bursts,” he says, “though I have no doubts the corporate publishers who are interested in serving the same books to the same readers over and over again will have no such qualms.”
In the realm of literature, there is still a sanctity around “the human and the humane,” as Stavans puts it. “Machines can’t read a book or experience any of the personal connections to language that give a book life,” adds Rogers, who became a translator after translating Finnish song lyrics for friends. “Machines don’t find themselves unexpectedly chuckling at a phrase, or repeating a string of words because its sounds are satisfying, or remembering being in a place like the place described in a book.” Though a cliché, it nevertheless rings true: The destination might pale in comparison to the joy of the journey, something a machine might never know.
Jimin Kang is a Seoul-born, Hong Kong–raised, and England-based journalist and writer. Her work has appeared in the New York Times, the Nation, the Kenyon Review and the Los Angeles Review of Books, among other publications.
Thumbnail credit: Ricardo Resende