Re: LANG: A proposal for the layering problem

Dan Connolly wrote:
> 
<snip>
> 
> To me, RDF Schema is a vocabulary of terms (e.g. subClassOf) to be used
> within RDF, i.e. within the Resource Description Framework,
> for describing resources such as properties and classes.
> 
> WebOnt should be another vocabulary of terms (e.g. disjointWith)
> that can be used in the same description framework.

We are in complete agreement on this point. As I said earlier:

> > All I am proposing is that we provide definitions of the
> > namespace in a language other than RDF triples.

The main difference is that I think RDF triples are insufficient for
defining the meanings of the terms we want to use in our RDF documents.
Even with my suggestion, we could still mix RSS and Dublin core
properties and classes with those from some new WebOnt ontology in order
to state facts in some RDF document. RDF Schema-only or WebOnt-only
agents would only have partial understanding of this document (because
they couldn't understand the term definitions from the other), but
nevertheless, both agents would still get the complete set of ground
facts. Furthermore, agents that have access to translation capabilities
or could grok both will have full understanding of our document.

<snip>
> 
> Second, all this translation is
> a huge drag on the network effect
> of the Semantic Web.

Certainly, too much translation can be problematic. If we do our job
right, then people will begin to recognize that WebOnt is the best
approach for defining terms on the Web, and hopefully RDF Schema-only
resources will become scarcer and scarcer. If this happens, then these
translations will be few and far between.

> > This language can
> > coexist with RDF Schema and translation services can be written to allow
> > maximal understanding between RDFS and WebOnt agents.
> 
> Yes, that's entirely possible. But it's not very interesting.
> 
> I'm pretty much obliged to help this WG thru the process
> of building a W3C Recommendation based on whatever
> technical consensus it comes up with. But I can
> tell you right now that if it's not integrated
> into the Resource Description Framework, my
> heart won't be in it: it'll simply be
> yet another legacy data format that I need
> to write grokOntology.xsl converters for.

I never suggested that we shouldn't integrate with RDF, it's RDF Schema
that my beef is with. I think using RDF triples to define RDF terms is
like finding a hammer and thinking everything is a nail. Still if
someone can layer WebOnt on RDF Schema by using a syntax that doesn't
have any ugly worts and a semantics that really works, then I will be
perfectly happy. If, however, this can't be done, then I ask that we
seriously consider alternatives, such as a non-RDF triples (but still
XML) syntax for defining the ontologies of terms we use in our RDF
documents.

BTW, I really don't want to turn effort this into something that you
find boring. It might help me better understand your point of view if
you could provide some use cases for why our language should use
RDF-triples to define terms. What is it that you want to be able to do?

> I could get into the development of an RDF2...
> i.e. something where stating n-ary facts
> isn't so painful... something with universal
> as well as existential quantification...
> that sort of thing.

Certainly, an RDF2 might help. Unfortunately, I don't see that happening
in our schedule. 

> But yet another XML syntax that can't be
> automatcially merged/conjoined with data
> written in other vocabularies? Yawn. Sigh. Trundle trundle...

Once again, my suggestion is that data stays as RDF. Thus it can still
be combined easily in the ways you mention. However, the ontologies
themselves would not be RDF. Sure, this means you can't combine
ontologies by simple concatenation, but that would be a dangerous thing
to do anyway. Ontology merging is a complicated process, just ask
Deborah McGuinness who does a lot of research in this area. Just to give
you one quick example, two ontologies might assign different meanings to
the same term; in order to merge these ontologies you will have to
rename one of the terms.

Received on Friday, 29 March 2002 10:24:53 UTC