To get expressiveness in describing the semantics of the domain of the XML I think OWL 2 DL might be best, if querying technology can handle it. It is not as overwhelming to learn as all those W3C references to OWL standards make it appear. It just comes down to a trade-off between expressiveness and query/inference engine performance. If my memory serves me correctly. It is a long time since I was involved in studying this. I looked into it as part of my work with OASIS UBL and the UBL/ebXML ecosystem.