Re: How to layer the semantic web properly?

This is a rather late reply (by webont standards at least) but there has
been considerable discussion of this and I would like to put forward (yet
again) some of my current views.

peter


From: Dieter Fensel <dieter@cs.vu.nl>
Subject: How to layer the semantic web properly?
Date: Fri, 04 Jan 2002 02:41:26 +0100

> How to layer the semantic web properly?
> 
> We had recently a large amount of discussions on the right
> layering of OWL and RDF. This is not really surprisingly
> because it touches a key issue of the semantic web. How
> to understand the layering of the various elements that
> Tim put in his visionary slide on top of each other (see
> [1]). I would like to provide the following intuitions:
> 
> 1. This layering cannot be interpreted in model-theoretic
> terms. That is, we cannot expect that each set of inferences
> at a higher level is a super set of all inferences at a
> lower level. Each level will have its specific style on
> how to draw inferences. 
[...]
> Instead of viewing this as a bug we should view it as a
> nice feature that provides the world with different reasoning
> styles.

It may indeed turn out that the semantic web formalisms end up in this
situation.  If so, however, I strongly feel that the semantic web will
*not* be layered and all concerned should be extremely careful to *not*
promote the view that it is.  To me, layering *strongly* implies that the
upper layers respect the views of the lower layers, which is not the case
if a higher layer violates the meaning of a lower layer.

> 2. I would strongly recommend to keep the layers syntactically
> as close as possible. Even if a Pat Hayes aware RDF agent draws
> some strange conclusions in the average we will win a lot
> if an RDF agent can process the syntax of OWL statements
> and if it draws conclusions which are reasonable for its
> level and even right in the average for an OWL agent.

I would strongly recommend that the different formalism *not* be kept
syntactially identical, except in that they all use XML.  Different
constructs have differing syntactic requirements, and, in particular,
triples are just about the worst way to provide syntax for most constructs.

> 3. It is highly desirable that OWL has a sound model theory. This
> can easily be achieved. Only if one uses RDF as a syntax AND
> as a semantics (i.e., if somebody makes the mistake to define
> the OWL semantics as an extension of the Pat Hayes RDF semantics)
> then he runs into problems. My personal conclusion: Lets
> define a simple semantics of OWL without reference to the
> model theories of other working groups. Or spoken as a
> sociologist: I see some changes that we agree on syntax between
> different working groups but we will run in endless war or
> non-compatibility if we try to agree on the basis of model theory
> between different working groups.

My view here is that OWL semantics should be close to RDF semantics even if
it is not an extension of RDF semantics.  My view is that the semantic web
is about semantics---if the semantics do not line up then there is no
unified semantic web.  (This is not to say, yet, that the semantics of the
various formalisms need to be completely compatible.)

> 4. The question whether to base OWL syntactically on XML or
> RDF is a question of high strategic impact. Such a
> question can neither be decided on the background of problems
> in expressing DAML+OIL lists in RDF nor based on problems
> in monotonically extending Pat Hayes RDF model theory. There
> need to be much more serious arguments to justify such a schism
> of the semantic web.

On the contrary.  It is precisely these technical arguments that determine
the syntax of OWL.  Political arguments may help determine the syntax of
OWL, but they are necessarily subservient to the technical arguments.  That
is, if there are no technical reasons for requiring one syntax over
another, political arguments can be used.  However, political arguments
should *never* be used to select a technically untenable solution, nor should
they be used to select a technically inferior solution.  (Of course,
political arguments are used to select technically inferior solutions all
the time, but this is a bug, not a feature.)

> Dieter

peter

Received on Friday, 18 January 2002 12:45:12 UTC