I've been working on a FEP draft for the past few days: FEP-22cd Attributing translations.

Hackers' Pub already machine-translates articles, but I want authors to be able to publish their own translations too, and let organizations split the work across members. Once you do that, Activity Streams doesn't give you much to work with. attributedTo names an author, not a translator. Nothing distinguishes a language a human translated from one an AI produced. And nothing says whether a translation still matches the current text or was written against an older version of the article.

Schema.org already has translator, translationOfWork, and inLanguage for exactly this, so the FEP mostly reuses those instead of minting new vocabulary. I had to decide whether to give each language its own object, linked back to the original, or keep everything on one object the way contentMap already does. Splitting it looked cleaner at first, but it also means spreading replies and reactions across separate posts for each translation. I don't want adding a translation to split the conversation.

I kept everything on one object, using contentMap for the text in each language. The draft adds a translations property to record who translated each version and whether it was translated by a human, generated by a machine, or generated by a machine and reviewed by a human. A timestamp lets you tell whether a translation still matches the current text, without publishing the server's private edit history.

You can discuss it on SocialHub or in the draft PR. If you run anything that touches multilingual content, I'd like to hear where this breaks.