The translator the writer would have hired.

There's a quiet violence in a bad translation. The book ships, the cover looks the same, the name on the spine is the writer's. But the voice on the page belongs to whoever was cheap enough to take the job.

We didn't start here because we were excited about translation as a market. We started because every writer we know who isn't writing in English already lost something they cared about, in a translation they couldn't afford to fix.

The catch is old. Good literary translators are rare and expensive. Cheap translators are everywhere and reckless. Most writers' books fall through the gap. The few that don't are the ones whose publishers paid for the translator the writer would have hired.

What we're trying to build is a tool that gives that translator's hands to the writer — not the translator's judgment, not the translator's taste, not anything that should remain a person's choice. Just hands. So a writer can move paragraph by paragraph, keep her voice, ship the book.

The temptation, with this kind of work, is to promise the moon. Replace the translator entirely. Push the button, get the book. We have no interest in that pitch. The reason the cheap translation flattens the voice is the same reason an automatic one does: nobody on the other side cares about this writer's specific decisions, this writer's specific rhythm, this writer's specific reasons for the line break in the third paragraph of chapter four. Caring is the part you can't automate. So we don't try.

The bar we're holding ourselves to is simple: the writer should finish a session more capable than she started. If she has to fight the tool to keep her voice, the tool failed. If the tool helps her keep her voice and ship faster, it earned its place.