RE: Proposal to resolve ISSUE-86

The problem, as presented in ISSUE-86 is the representation of inverse
object properties in *facts*, i.e., ObjectPropertyAssertions.  I find it
easy to get confused when we add to the discussion the use of object
property expressions in restrictions, which does not have a problem in
RDF.  I've limited my comments below to focus on object property
assertions.


Reusing the example fact from Boris's email [1], 

(1) ObjectPropertyAssertion( InverseObjectProperty( P ) i1 i2 )


Proposal #1) Disallow this by modifying the definition of
ObjectPropertyAssertion.  The following axiom (2) is logically
equivalent.

(2) ObjectPropertyAssertion( P i2 i1 )


I believe this restriction matches OWL 1.0.  It was said on the telecon
that this does not meet the requirements of some stakeholders, the DIG
community in particular was mentioned.  This brings us to


Proposal #2) Allow these facts in the functional syntax but invert the
assertion prior to the mapping to RDF.

This allows (1) in the functional syntax but serializing it in RDF as
(2).  Thus, it is not round-trippable through RDF.

To which I'm sure some will reply that round-tripping through RDF is
necessary.  So,


Proposal #3) Allow these facts in the functional syntax but invert them 
and add an annotation prior to the mapping to RDF.

So (1) would be mapped to the following (3) before going to RDF.

(3) ObjectPropertyAssertion (
 Annotation( owl11:invertedObjectPropertyAssertion "true"^^xsd:boolean )
 P i2 i1
 )

The annotation facilitates round tripping. I don't think that the
details of the annotation are important, only its presence, and consider
the example above replaceable.  I find this solution attractive because
it limits the solution to the RDF to/from FS mapping, which reflects
where the problem is.  The triple level details of this solution are
dependent on the resolution of issues related to the mapping of
annotations on axioms.


I prefer any of the proposals outlined above to those proposed which
require minting canonical inverse URIs for every property.

Further, I believe that the approaches here will produce a more elegant
behavior in OWL 1.0 and RDFS tools that are not explicitly updated for
OWL 1.1.


Regards,
-- 
Mike Smith

Clark & Parsia

[1] http://www.w3.org/mid/001801c88f7b$10383e90$4012a8c0@wolf

Received on Wednesday, 26 March 2008 21:11:02 UTC