For most of the twentieth century, literary translation occupied a peculiar professional niche: underpaid, underappreciated, and absolutely essential to global culture. Translators transformed Tolstoy into something English readers could weep over, carried García Márquez across the Atlantic, and smuggled dissident poetry out of regimes that would have preferred silence. The work demanded fluency in two languages and mastery of neither—because the real skill was something harder to name.

Then the machines arrived, and they were shockingly good at the nameable parts.

The fluency trap

Neural machine translation, the technology underlying services from Google to DeepL, has achieved a kind of competence that would have seemed miraculous two decades ago. Feed it a German novel and it will return English prose that flows, that respects grammar, that rarely stumbles over idiom. The output reads like something a human might have written—which is precisely the problem.

Fluency, it turns out, was always the easy part. A talented undergraduate with a dictionary can produce fluent translation. What distinguishes the great literary translators—figures like Gregory Rabassa, who rendered One Hundred Years of Solitude into English, or Constance Garnett, whose Dostoevsky shaped generations of Anglophone readers—is something else entirely: the capacity to hear what a sentence is doing beneath what it says, and to find a way to make it do that thing in another language.

Machine translation optimizes for equivalence. It seeks the statistically most probable rendering of each phrase. But literature often lives in the improbable—the word choice that surprises, the rhythm that unsettles, the ambiguity that the author planted like a seed.

What the job has become

The economics of translation have shifted accordingly. Publishers increasingly use machine translation as a first draft, hiring human translators to "post-edit" the output rather than translate from scratch. The fee structure reflects this: post-editing pays less than original translation, sometimes substantially so. For genre fiction, technical manuals, and commercial nonfiction, the arrangement works tolerably well. For literature, it produces a particular kind of mediocrity—prose that is correct and lifeless, like a photograph of a meal.

Some translators have adapted by specializing in what machines cannot do: poetry, experimental fiction, works where voice is everything. Others have left the field. A generation of mid-career professionals, trained before neural networks but too young to retire, finds itself in an uncomfortable position—possessing skills that remain valuable but are increasingly difficult to monetize.

The translators who thrive tend to be those who can articulate, to editors and publishers, exactly what they provide that a machine cannot. This requires a kind of professional self-awareness that the field historically never demanded. You used to prove your worth by producing good translations. Now you must also explain why your translations are good in ways that matter.

The deeper question

Beneath the economic disruption lies a more interesting problem. Machine translation forces us to ask what literary translation is for. If the goal is simply to convey information—to let an English reader know what happens in a French novel—then machines are adequate and improving. If the goal is to create a work of art that stands on its own in the target language, that captures not just meaning but experience, then we are dealing with something machines have not yet learned to want.

The word "yet" does a lot of work in that sentence. Large language models are improving at tasks that once seemed to require understanding. Perhaps they will eventually produce translations that move readers to tears. But the current generation of AI, for all its fluency, operates without any sense of what makes a sentence beautiful or why a particular word choice might matter. It has read everything and understood nothing.

Our take

The best literary translators have always been invisible artists—shaping how millions of readers experience foreign masterpieces while receiving little credit and less money. AI has made their invisibility harder to sustain. The machines have forced a reckoning with what translation actually is, and the answer turns out to be something closer to performance than transcription. A great translator does not convert a text; they interpret it, the way a pianist interprets a score. That interpretation cannot be automated, not because machines lack processing power, but because they lack the one thing interpretation requires: a self that has lived in language and knows what it means to be moved.