Re: LANG: Nested Class definitions and the RDF striped syntax

From: "Jonathan Borden" <jonathan@openhealth.org>
Subject: Re: LANG: Nested Class definitions and the RDF striped syntax
Date: Fri, 22 Mar 2002 20:53:47 -0500

> Frank,
> 
> 
> > Jonathan Borden wrote:
> >
> > > It has been claimed that allowing nested class definitions in OWL would give
> > > difficulties for tools that wish to roundtrip ontologies. I don't think that
> > > that alone would be the case for the following reason:
> > >
> > > One of the properties of the RDF/XML syntax is that an OWL class definition
> > > with nested class definitions can be converted into a form consisting of
> > > multiple 'top level' classes. [follows example]
> >
> > It is exactly as you say.
> >
> > Indeed, tool builders for DAML+OIL
> > have reported problems with "round tripping": once a "frame-style"
> > DAML+OIL ontology has been parsed into RDF triples, it is very hard,
> > if not impossible, to reconstruct the original grouping of the
> > statements from those statements, causing he "frame-style" to be lost (even
> > though the formal meaning remains identical), just as happens in
> > your example.
> 
> As I have shown, for every class definition using nested syntax there exists
> an equivalent set of class definitoins that use 'top-level' syntax. We agree
> that the triples are in some way equivalent (modulo bNodes that are given
> explicit names in the top level syntax).

This is not sufficient to prevent round-tripping problems.  

What would be sufficient would be to also show that there is an inverse
transform from the ``top-level'' syntax to the ``nested'' syntax that
results in a knowledge base close to the original.  Otherwise it is
possible that a nested syntax knowledge base is transformed into a
different-looking knowledge base, and that the differences in appearance
are significant.

Actually, this would not be quite sufficient, as the desired syntax may
contain not only nested definitions but also other information.  To be even
more precise you would have to define equivalence classes of knowledge
bases and show that round-tripping ends up back in the same equivalence
class.

> > The point is very well documented in [1].
> >
> > An important feature of the OWL Light segment is that it's > *only*<
> > syntax is
> >   of slots grouped into frames, so it doesn't suffer from the problem of
> > flattening grouped definitions into multiple top-level statements.
> 
> You see, I am saying that the ability to flatten is a _feature_ not a
> problem. A tool which uses the 'frame-style' can regenerate the
> 'frame-style' by regenerating the RDF graph, repacing all bNodes with a
> name, perhaps a random name such as "genid()".

This is what you have not shown.

> What I am saying is that if the tool _only_ uses the top level syntax
> (equivalent) and always converts nested syntax into top level, then the
> created ontologies can be roundtripped. Certainly roundtrippind _nested_
> ontologies will be difficult, but the whole point of "OWL-Lite" is that the
> ontology editor will only use the top level syntax. Ok, so convert nested
> into top level (its all the same to RDF) and just use top level.

You haven't even shown this.

> Perhaps I am missing something? Of course the other option is to edit your
> ontologies as XML and parse as RDF when needed. This is how most XML
> systems do things, e.g. even though XML is parsed into an internal object
> model e.g. the DOM, _editors_ don't edit the DOM, they edit the XML. Indeed
> if I were writing an ontology editor, I would edit your abstract syntax
> directly and then generate XML/RDF from that. I will say more about this
> later.

What you are missing is that round-tripping has to end up with a knowledge
base that has the same extra-logical connotations (whatever they may be) as
the original knowledge base.  These extra-logical connotations may include
even such things as the order of definitions.

> Jonathan

Peter F. Patel-Schneider

Received on Friday, 22 March 2002 09:10:07 UTC