[skreutzer]

Whichever you prefer: XHTML, DocBook, TEI, ODT, or your own custom one. Anyway, in the meantime, jeffcb mentioned that the source format is *.doc. As those source documents only contain raw text when they arrive (or their formatting gets stripped), it would make sense to put that text into a layout-agnostic but semantic format as such work takes place anyway, so all target formats could be generated from that. Just to be very clear: if jeffcb would decide to construct the EPUBs as suggested, I don't see why one wouldn't generate a PDF from that very EPUB or XHTML files.