Will humanity get lost in translation?

Unlock the Editor’s Digest for free

Anyone who grew up in a multilingual society intuitively understands the value of translation, and translators. The gorgeous, flamboyant fiction of Gabriel García Márquez and Julio Cortázar came to Calcutta street bookshops via the late Gregory Rabassa’s superb translations from Spanish to English; my grandmother first read out the Mahabharata to me in Kaliprasanna Singha’s Bengali translation from the original Sanskrit.

I played hopscotch with languages as a child, skipping between Bengali, English, Hindi, grazing alongside French, Spanish, Sanskrit. Translators ferried us from the banks of one language to another, bringing Malayalam writers into Hindi, Pushkin or Isabel Allende into Bengali. A poor translation could kill your love for a book stone-dead; a good translation breathed fresh life into the works of beloved writers. “To translate a book is to enter into a relationship with it, to approach and accompany it, to know it intimately, word by word, and to enjoy the comfort of its company in return,” author and translator Jhumpa Lahiri wrote in Translating Myself and Others (2022).

But will translation remain human for much longer? The debates over AI and machine translation have been fiery this year. A month ago, Veen Bosch & Keuning — the largest publisher in the Netherlands — announced plans to experiment with AI translation for a limited number of commercial fiction titles. Industry giant HarperCollins asked authors for permission to use “select nonfiction backlist titles” to train AI Language Learning Models (LLMs) for Microsoft. There is a widespread fear that these AI models may one day make authors — and translators — obsolete.

My own preference is strongly for human translations, but AI enthusiasts argue that not all human translations are phenomenal, and that a poor translation can kill a novel or a collection of poems. Human translation takes time and can be expensive for publishers; the promise of AI is that it might — if LLM models advance in the right direction — open up a whole new world of literature to readers, instantly. Even for publishers who prefer to keep more creative writing in the hands of a human expert, the temptation to use AI translations for textbooks and bare-bones non-fiction is likely to be irresistible.

In January 2024, the Society of Authors ran a survey, asking their 11,500 members (authors, illustrators and translators) for their views on the impact of generative AI on creative careers. The results speak to both the promise and threat of AI translations: more than a third of translators (37 per cent) polled said they had experimented with generative AI, 36 per cent of translators had already lost work because of generative AI, and 77 per cent believed that machine translation would negatively impact their income.

Can LLMs and machines translate as intuitively as a human? Despite machine “hallucinations”, where generative AI produces false or misleading responses, and other early glitches, AI translation will improve, and will probably be adopted across the corporate world. This threat to the livelihood of creative translators comes at a time when there are more translation prizes, and greater appreciation of the work that they do — visible, and invisible. AI treats translation like a crossword or logic puzzle, looking for one correct answer. Some of the skills that translators possess — creativity, deft selection, an understanding of authorial intent — are purely human acts. And as ambassadors for authors and books, no AI model could hope to replace the figure of the translator.

The translator’s role so often extends beyond the text. Jennifer Croft, who translates works from Polish, Ukrainian and Argentine Spanish into English, spent almost a decade pitching Olga Tokarczuk’s 2007 novel, the Man Booker International winner Flights, to publishers. “I am often the person who also sells those books, acting as a kind of agent for the authors I’m translating,” she said in an interview. “That’s an aspect of translation that I think most people are not aware of.”

Translators are writers, and translation is an act intimately tied to emotion, imagination, bodily experience. Charles Simic, the Serbian-American poet who died in 2023, told the Library of Congress that translation was “an act of love, an act of supreme empathy”; to Daisy Rockwell, another star translator of authors such as Upendranath Ashk and Geetanjali Shree, the translator and the author are “ballroom dancers”. And Idra Novey felt, when she was translating the Brazilian writer Clarice Lispector, “as if I were placing my palms against her palms and could hear the beat of her sentences matching my own”.

None of us can wish generative AI away, and there are plenty of fields where it might be a timesaver, a boon. But publishers must reflect on whether AI translations have any place in our creative lives. Back in 2011, the legendary translator Margaret Jull Costa said, “Every translator will produce a different version, because every reader or listener reads and listens differently.” The miracle of a good translation, to her, was that it was “as seductively fresh and original as the original.” That is what I and other readers respond to so strongly; the dance between author and translator, one palm pressed against the other.

Join our online book group on Facebook at FT Books Café and subscribe to our podcast Life and Art wherever you listen

#humanity #lost #translation