Make all the comments I18N tokens, get people to produce comments in every language needed and populate the needed libraries, run the code through an XSLT transform to produce the appropriate doc for any given language as needed via an script at editor load time, with conversion back to canonical format after the editor is done and appropriate separation of any new comments back into the appropriate library via some other script (obviously an XSLT 2 app)....

Umm, yeah, right.  It would be simpler to teach all the developers English from scratch...