RE: OWL2 monotonicity (was How do you deprecate URIs? )

> From: Bijan Parsia [mailto:bparsia@cs.man.ac.uk]
>
> [ . . . ]
> > if the graph has entailments E without applying the OWL 2
> > semantics, then all of those entailments E should still be valid
> > when OWL 2 semantics are applied, i.e., the OWL 2 entailments
> > monotonically extend the existing entailments.
>
> This isn't possible, I think. I mean, let's say you define the
> "class" owl:TransitiveProperty to be equal to owl:Nothing (this is a
> monotonic extension of RDF). P type owl:TransitiveProperty is now
> inconsistent under that monotonic extension (thus has all
> entailments) but not in OWL.

I don't follow your example.  What has been added to cause entailments to be invalidated?  And why is it necessary that they be invalidated?

> . . . Monotonicity usually refers to what happens if you *add*
> axioms, not if you change the semantics.

I mean monotonicity in this sense:
http://en.wikipedia.org/wiki/Monotonicity_of_entailment
though I guess I phrased it poorly.

The point is that if I have a graph such as:

  :a P1 :x .
  :b owl:sameAs :a .

which entails:

   :b P1 :x .

and then I add an assertion such as:

  P1 rdf:type T .

then

   :b P1 :x .

should still be entailed (monotonicity of entailment), whether T happens to be owl2:AnnotationProperty or anything else.

> [ . . . ]
> > I don't mean to irk anyone, but I do mean to be clear that I think
> > semantic monotonicity is architecturally *critical*, so
> > that RDF tools
>
> Tools can regardless, obivously.
>
> > and reasoning can be safely applied to *all* RDF graphs, regardless
> > of whether or not a graph uses OWL 2.
>
> It's unclear that this is really a problem in practice.

Well, it may not be a problem for you, but we've just been discussing an example (annotation properties, and deprecating URIs) where it clearly *is* a problem for others.  And even if the problem didn't show up much right now, because we don't have much RDF data to combine, I don't think we should ignore the obvious consequence when more RDF data is available and combined.

> Frankly,
> there are all sorts of minor strangenesses with the alignment between
> all theses languages.

Okay, there may well be other flaws, but that isn't any justification for also allowing what I would see as a major problem (non-monotonicity).

>
> >   This is particularly important when RDF graphs are combined from
> > many sources, some of which may use OWL 2 and others not.
>
> Given the weakness of RDF semantics and the sort of entailments you
> may or may not have, I'm pretty skeptical that this is a serious
> problem.
>
> IOW, architectural nicety needs to be weighed against lots of other
> consideration, including functionality.

Well, you may view this as an architectural nicety, but I want systems to work, with minimal effort, when RDF data is combined.  And if OWL2 is not monotonic in this way, then they *won't* work without being forced to implement OWL 2 semantics.

I can understand that you and others in the OWL 2 working group may not see that as a problem.  After all, you're immersed in OWL 2.  But it is a much larger pill than I want RDF applications to have to swallow in order to make use of combined RDF data.  I want the RDF entry barrier to be as low as possible, and this would raise it substantially.

> BTW, just so we're on the same page, the phrase "architecturally
> critical" just tune me out. Please, give specific technical points
> instead.

Huh?  That's exactly what I *did* in the rest of the sentence:

> > I think semantic monotonicity is architecturally *critical*,
> > so that RDF tools and reasoning can be safely applied to
> > *all* RDF graphs, regardless of whether or not a graph uses OWL 2.




David Booth, Ph.D.
HP Software
+1 617 629 8881 office  |  dbooth@hp.com
http://www.hp.com/go/software

Statements made herein represent the views of the author and do not necessarily represent the official views of HP unless explicitly so stated.

Received on Thursday, 17 July 2008 18:38:26 UTC