Re: Where are the semantics in the semantic Web?

Quoting Harry Halpin <hhalpin@ibiblio.org>:

> To use XML Schema to define abstract concepts would force everyone
> sharing that concept, if it was defined *solely* in terms of XML Schema,
> to use the same set of XML serializations.

RDF is not that different.  While it's true that RDF/XML allows
certain variations, and that there are other syntaxes, in practice
you are still pretty narrowly constrained; and the downside is
that the variations allowed in RDF/XML make it difficult to
process it with many XML tools.

In the end, the different concrete representations of some RDF
turn into much the same triples.

> What is much more interesting and much more feasible for data
> integration and discovery (and the SemWeb in general) is to let people
> format their data in XML however they want (and there's lots of
> differing schemas out there...) and then provide a method for
> "connecting" them if they refer to the same ontological abstraction.
> This I believe what the RDF bindings for WSDL, WSDL-S, OWL-S, etc are
> providing, and this is a *good* thing despite the differing details.
> Think programming classs definition that lets one "hook" together all
> your various Java classes, Python, and SQL stuff and say "this class in
> Java, this class in Python, and this database table all refer to the
> same person".

I think a "hook-together" mechanism is a great idea, but it
doesn't have to be RDF and OWL that does it.  In my opinion,
after having used them for the purpose, OWL and RDF are not
a good fit for the data structures used in programming languages.

> And, as it must be pointed out again and again, XML does not provide a
> formal semantics. It's a serialization format like ASCII - you don't
> "reason" or "prove" things with just ASCII, do you?  Think of XML as a
> representation format for data, and think of the SemWeb as a
> representation format with a formal semantics for *what the data refers
> to* .

I am familiar with that point and don't find it convincing.
XML is not just like ASCII, for one thing; ASCII is at a lower
level.  While it's true that XML lacks a formal semantics,
in a sense that's just the lack of some agreed conventions
for treating XML Schemas (or something else) as class definitions.

Also, RDF has semantics in only a very limited sense.  It's
what's needed "to provide a technical way to determine when
inference processes are valid, i.e. when they preserve truth".
What URIs refer to can vary with the model / interpretation.

The RDF model theory says what it takes to be considered a model.
It doesn't give you any model.  It's like having the definition
of satisfaction for first-order logic plus the axioms for groups,
but without (yet) any groups.  If you see http://...#dog, it probably
means dog, but the model theory doesn't tell you that.  It has to
be specified separately what that URI means, and often that will
be done by natural-language text just as it is for Dublin Core
"creator".  (In some cases - perhaps that dog has an OWL class
definition - you get further before you have to resort of natural
language, but the problem still occurs lower down.)

There's also a question of how much reasoning is desirable.
In order to check whether I can do the equivalent of a
procedure call (which is what at least many invocations of
web services are), should I have to use a description logic
theorem-prover, or should I just have to do the sort of
reasoning done in OO programming languages?

-- Jeff

Received on Friday, 25 November 2005 17:54:12 UTC