For some time now, I’ve worked to build an open-source JATS XML typesetter. It’s called meTypeset. It’s not by any means perfect and the approach it takes is unlikely to ever yield 100% good markup from Word input. It does, though, get it right a lot of the time in “basic” (in terms of underlying XML semantic complexity) rich-text documents, which is what most academics (especially in the humanities) are writing.