From W3C Wiki
The application/rdf+xml media type is currently being registered. It is already seeing widespread use.
The application/owl+xml media type was proposed by Jonathan Borden, but rejected by the OWL WG. It's only mentioned here for its relevance to the issue below, about whether application/rdf+xml suffices for all RDF.
Is the RDF media type sufficient for all RDF?
The "No" Argument
Take any RDF Schema document for which there exist triples which can be inferred. If it is the intent of the sender to communicate those inferred triples, then either;
- it should explicitly mention the triples in the message and use the RDF media type, or
- it should not mention the triples, and use an RDF Schema media type to indicate that inference is expected (if the RDF Schema media type were so defined)
"application/owl+xml" appears to have been proposed for this reason.
The "Yes" Argument
"it's a zero-probability event that you used rdfs:domain on accident, without ever hearing of the RDFS spec. (and without copying from somebody who has, etc.)"
If the "Yes" argument is to be believed, then that suggests that either a) some namespaces are special such that their use implies automatic import of any associated axiomatic triples, or b) the use of any namespace implies automatic import. If a), who decides which ones are special, and what if, after deploying software, we decide that another one is special? If b), then existing RDF agents would be unable to extract all the triples in an RDF message if the graph includes namespace they aren't familiar with.
MarkBaker believes that a new media type is required for any specification declaring axiomatic triples. As a result, he also believes that axiomatic triples are bad practice and should be avoided, and instead, that those triples be explicitly included in any transferred serialization of the graph. This gives existing agents superior visibility into those graphs.