Hmm†¦ While I agree with the idea of separating schema from data, I have to admit I am confused by this statement.

I’m not sure which of the many statements above to which you are referring. If you point one out, I’ll be happy to comment.

How to differentiate T and A box can be a bit mysterious, but it seems to me that owl:sameAs talks about instances, so it is Abox. The fact that it is transitive does not impact this determination. A custom property (such as partOf, for example) can also be transitive.

Yes, I very much agree from my own reading that TBox and ABox splits can be mysterious. That is one reason we attempted to set out the “work” table above. I also think now that we are seeing traction with linked data, it is useful to dust off the DL discussions from a few years back in light of widescale, distributed datasets being published under numerous ontologies. In part, that was part of my intent.

I disagree totally with owl:sameAs. True, as I understand it, ABoxes are about instances, but more importantly their role has to do with assertions. owl:sameAs carries some really heavy iff entailments. Why do we need that when all we are asserting is a supposed identity, when that has not been tested? I would much prefer a predicate such as assertAsSame or some such, that when tested at the TBox level might then be expressed as sameAs, but not until.

In fact, I am wondering why any linked instance data needs to use any OWL at all.

It may be that you are concerned about the use of owl:sameAs with resources that may be classes. This immediately puts the ontology/RDF dataset into OWL Full. Is this a problem?

Well, we needed to go to OWL Full for UMBEL for that very reason, but, no, in this case it is not a cause of my points. I will say I’m quite pleased that we can avoid this problem with the OWL 2 proposals.

Thanks, Mike

How to differentiate T and A box can be a bit mysterious, but it seems to me that owl:sameAs talks about instances, so it is Abox. The fact that it is transitive does not impact this determination. A custom property (such as partOf, for example) can also be transitive.

It may be that you are concerned about the use of owl:sameAs with resources that may be classes. This immediately puts the ontology/RDF dataset into OWL Full. Is this a problem?

Not sure that this is true. There is nothing in OWL DL that would prevent one from describing properties of members of class Human, then describing different properties for members of class HairlessBiped, and then say that the classes are equivalent. This again seems to be all about context. For example, in our NASA work we may have an ontology that describes Devices from a functional perspective and another one that describes them from the structural perspective. As a result, there could be two Device classes (different name spaces and qnames) with different descriptions, but the same members. They would not violate any of the OWL DL principles.