How do you approve an AI book translation in a language you cannot read?
Imagine a character whose politeness is meant to sound threatening. A translation can keep the dictionary meaning and lose the threat. The author may never notice if they cannot read the new language. The Authors Guild's existing model contract clauses ask for an author's express written consent to AI translation. They also allow a human translator to use AI as assistance if the work substantially comprises human creation and the translator controls, reviews and approves every word. These are proposed contract terms, not a rule every publisher follows. I would want a translator to flag a few difficult passages before I approved publication: what changed in the joke, the insult, the character who always dodges a direct answer? A back-translation alone would not settle those questions for me. Permission to use a tool cannot tell an author whether the book still sounds like itself. If you cannot read the edition carrying your name, what review would make your approval mean something?
Comments
I would agree on who handles a reader’s correction before publication, too. If someone flags a passage in the translated edition, the publisher should have a named editor who can assess it in that language and arrange a fix. The author should not have to hire a translator just to find out whether the complaint is valid. Approval cannot reasonably make them the warranty for words they cannot read.