W3C home > Mailing lists > Public > www-webont-wg@w3.org > August 2002

Re: TEST: FunctionalProperty InverseFunctionalProperty 3.4 4.1

From: Ian Horrocks <horrocks@cs.man.ac.uk>
Date: Thu, 1 Aug 2002 16:28:53 +0100
Message-ID: <15689.21429.625470.903423@merlin.oaklands.net>
To: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Cc: <www-webont-wg@w3.org>

On July 26, Jeremy Carroll writes:
> Following my action from yesterday I have constructed test cases for these
> features and issues.
> The test case follows the pattern of my proposal:
> http://lists.w3.org/Archives/Public/www-webont-wg/2002Jul/0101.html
> as modified by DanC:
> http://lists.w3.org/Archives/Public/www-webont-wg/2002Jul/0133.html
> but I haven't yet done the Manifest files.

I don't particularly like the premises-conclusions style of test:

1. It is unnecessarily complicated - every test needs 2 files.

2. It makes it very difficult to automate tests. Even if there is only
one conclusion, then the standard way to perform such a test would be
to check that the combination of the premises ontology and the negated
conclusions ontology is unsatisfiable. Unfortunately, OWL does not
support negating ontologies, so we have to go outside the language in
order to perform the test. 

E.g., in FunctionalProperty test 001, the conclusion is
sameIndividualAs(object1, object2). The normal way to check this would
be to add the negated conclusion to the premises, but OWL does not
support adding complementOf(sameIndividualAs(object1, object2)) to an
ontology. Of course, given the semantics of OWL, I can figure out that
I can achieve what I want by adding differentIndividualFrom(object1,
object2) to the premises ontology and checking its consistency, but I
need some extra machinery to do this automatically.

If there is more than one conclusion in the conclusions ontology, then
negating the conclusions ontology gets even trickier.

Why not simply provide a single ontology file for each test, with the
test simply being that the ontology is consistent/inconsistent. (Note
that we should also provide test cases where it is tricky for systems
to determine that an ontology is consistent, e.g., cases where there
are models, but no finite models. It doesn't seem very natural to
express this in the premises+conclusions style.)

I am also concerned about tests in the style of FunctionalProperty
test 003. Here, we are asked to conclude that a property is
InverseFunctional given that its inverse is functional. This is a
higher order inference that is a consequence of the semantics of OWL,
but cannot be proved within OWL. I.e., OWL does not state anything

forall P,Q . FunctionalProperty(P) ^ inverse(P,Q) -> InverseFunctionalProperty(Q)

I would *NOT* expect an OWL reasoner to find this inference.

The test could of course check an OWl inference that depended on the
relevant semantics, e.g., that if both object1 and object2 are related
to object3 by a property whose inverse is a FunctionalProperty, then
sameIndividualAs(object1, object2).

Regards, Ian

> The test files are:
> (all these links are forbidden, zip attached)
> http://www.w3.org/2002/03owlt/I3.4/bad001.rdf
> http://www.w3.org/2002/03owlt/I4.1/bad001.rdf
> showing the deprecated names, and
> http://www.w3.org/2002/03owlt/FunctionalProperty/premises001.rdf
> http://www.w3.org/2002/03owlt/FunctionalProperty/conclusions001.rdf
> http://www.w3.org/2002/03owlt/FunctionalProperty/premises002.rdf
> http://www.w3.org/2002/03owlt/FunctionalProperty/conclusions002.rdf
> http://www.w3.org/2002/03owlt/FunctionalProperty/premises003.rdf
> http://www.w3.org/2002/03owlt/FunctionalProperty/conclusions003.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/premises001.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/conclusions001.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/premises002.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/conclusions002.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/premises003.rdf
> http://www.w3.org/2002/03owlt/InverseFunctionalProperty/conclusions003.rdf
> showing the basic meaning of the terms.
> I have not done any tests that involve datatype properties since I am not
> sure where we are at with those; however IMO these will be needed.
> I will try and do Manifest files later today, but it may slip until Monday.
> I will also run the files through the converter to get N-triple variants.
> Jeremy
Received on Thursday, 1 August 2002 11:32:05 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 23:04:33 UTC