Re: Determining the type of an IRI for OWL 2 Full represented in structural specification

From: Jonathan Rees <jar@creativecommons.org>
Subject: Determining the type of an IRI for OWL 2 Full represented in structural specification
Date: Mon, 23 Mar 2009 12:22:27 -0400

> 
> On Mar 19, 2009, at 8:09 AM, Ian Horrocks wrote:
> 
>> Dear Frank,
>>
>> Thank you for your comment
>>     <http://lists.w3.org/Archives/Public/public-owl-comments/2009Jan/0035.html>
>> on the OWL 2 Web Ontology Language last call drafts.
>> ...
>> The Structural Specification and Functional-Style Syntax document was
>> always intended as a specification of the features provided by OWL 2 as
>> a whole [...]
> 
> This statement came as a big surprise to me. I had thought the
> structural spec was only of interest for OWL 2 DL.

It is a big surprise to hear anyone in the OWL WG say this.  

It has been the intention of the OWL 1.1 developers all along, and the
OWL WG has done nothing to gainsay it, that SS&FS was a specification of
how the constructs in OWL 2 are supposed to work throughout OWL 2.

Take, for example, qualified cardinality restrictions.  SS&FS specifies
the syntax of qualified cardinality restrictions, namely that they are
like unqualified cardinality restrictions but add a qualifying
class/datatype.  How can there be any doubt that this is how qualified
cardinality restrictions work in all of OWL 2?  Is OWL 2 Full allowed
to, for example, have a property instead of the qualifying
class/datatype?  No!  

This even extends to the semantics of OWL 2.  The direct semantics
provides not just the meaning of qualified cardinality restrictions in a
DL setting, but it also provides the general meaning for qualified
cardinality restrictions in all of OWL 2.  Of course, the direct
semantics does not provide the overall setting for the RDF-based
semantics of OWL 2 nor does it provide the formal development of the
RDF-based semantics for qualified cardinality restrictions - that is the
job of the RDF-based semantics document - but the semantics for
qualified cardinality restrictions in the direct semantics document is
the controlling semantics for qualified cardinality restrictions
throughout all of OWL 2.

There are indeed places where this breaks down somewhat on the semantic
side, but the OWL WG has been quite careful in determining just where
they are.  (Unfortunately, the OWL WG was *not* very careful in
providing user-facing documentation of this, and other aspects of OWL 2,
thus the second last call.)

> When converting OWL 2 Full to the structural specification, how does one
> determine the type of each IRI occurrence? OWL 2 Full does not require
> declarations, so one cannot just consult declarations. Is the intent
> that an IRI be assigned a type at each occurrence in any manner that
> permits the construction of correct UML instances? Or is it that only
> those OWL 2 Full ontologies that provide declarations are convertible to
> the structural specification?

There are some RDF graphs that are not easily convertible to the
functional syntax / structural specification.  As you say, RDF graphs
that do not provide complete typing information are somewhat
problematic.  However, there is a way to think of these RDF graphs -
just make all properties object properties.  Why does this work -
because it is possible to think of all properties as object properties
(and all classes and datatypes as classes).  Why does this not work
completely - because direct semantics makes individuals and data
values disjoint.  One could if one wanted, eliminate the RDF-based
semantics for all RDF graphs of this kind and come up with a slight
variation of the direct semantics for these RDF graphs.

So, then, why have the RDF-based semantics at all?  Well, for starters,
the mapping from RDF graphs back to the functional syntax/structural
specification can be ambiguous, especially when the RDF graph uses the
syntactic vocabulary of OWL 2 (e.g., owl:onProperty) in ways so that the
OWL 2 constructs cannot be reliably determined and/or separated from
statements about the domain.

> More generally, what is the attitude toward OWL 2 Full ontologies that
> have no SS representation at all?  Or does the structural specification
> cover OWL 2 Full (i.e. permit representation of arbitrary RDF graphs) in
> some way?

The structural specification / functional syntax does not usefully cover
*all* RDF graphs, even ambiguously.  For example, it is not very useful
to reverse-transform RDF graphs that contain partial OWL 2 constructs
(e.g., qualified cardinality restrictions missing their cardinality).

> I do not understand the desire to represent OWL 2 Full in the structural
> spec; I don't see how this is at all useful. It seems much simpler to me
> to say that the structural spec is only intended for use with OWL 2
> DL. Then there is no need to answer annoying questions about
> declarations and dialect coverage.

I don't buy this at all.  OWL 2 DL has conditions to permit effective
reasoning.  There is no problem, however, in using the structural
specification (and direct semantics) to provide meaning for RDF graphs
that are reverse-translatable but that do not meet these conditions.
Similarly, OWL 2 DL has conditions to allow unambiguous
reverse-translation.  There is no problem, however, in internally using
OWL 2 ontologies that are not reverse translatable.  Further, there is
real utility in taking RDF graphs that are not formally
reverse-translatable and extending the reverse translation to them
(perhaps resulting in a reverse translation that is not monotonic),
which may result in an ontology in the structural specification that is
not an OWL 2 DL ontology but that nevertheless can be given meaning by
the direct semantics and that can be reasoned with using existing
reasoners.  In fact, Alan Ruttenberg has been agitating for the OWL WG
to do this.  (Why hasn't it happened then?  Partly because no one has
had the time to do so and partly because there are many choices that can
be made in such an extended mapping.)

It seems much better to say that the structural specification /
functional syntax covers all of the features of OWL 2.

> My apologies if this has been discussed (surely it has; but I am about
> 3800 messages behind in my reading). If so kindly direct me to the
> correct archived email thread.

Much of the underpinnings of the relationship between OWL and RDF was
discussed in the original WebOnt Working Group and has been carried
forward into OWL 2.  The OWL WG has discussed the relationship between
OWL 2 DL and OWL 2 Full several times.

> Thanks
> Jonathan

Peter F. Patel-Schneider
Bell Labs Research

Received on Tuesday, 24 March 2009 10:31:58 UTC