W3C home > Mailing lists > Public > www-webont-wg@w3.org > June 2002

RE: layering (5.3, 5.10): Sardinia compromise?

From: Massimo Marchiori <massimo@w3.org>
Date: Thu, 20 Jun 2002 19:46:53 +0200
To: "Jonathan Borden" <jonathan@openhealth.org>, <www-webont-wg@w3.org>

Apologies for the late reply, it's a hectic period of the year...

> -----Original Message-----
> From: Jonathan Borden [mailto:jonathan@openhealth.org]
> Sent: Monday, June 17, 2002 4:17 AM
> To: Massimo Marchiori; www-webont-wg@w3.org
> Subject: Re: layering (5.3, 5.10): Sardinia compromise?
> Massimo Marchiori wrote:
> >
> > I missed the aforementioned conversation, alas, but notice
> > that we just need to read better the wowg's archives: the
> different-domain
> > proposal has been my point since the very beginning
> > (cf.
> http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0037.html ,
> > also
> > reported at a teleconf...).
> Might you respond to
> http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0045.htm
> l ? (just
> so we can try to sort out if we are all on the same page :-)
> ...

Yes, will try to do so at the end of the message.

> > Even more, for me the "same-domain" had *NEVER BEEN AN OPTION*
> (i.e., not
> > even *thought* about imposing  a same-domain layering chain),
> and took me
> a
> > while
> > to digest what Peter & Pat actually meant by their "layering is
> impossible"
> > argument (again, cf the above email...).
> >
> By "different domain" are you suggesting that OWL might provide
> an entirely
> different semantics for a graph of e.g. N-triples, than RDF, i.e. that
> triples become the _syntax_ of OWL which provides its own model
> theory (i.e.
> semantics) ?

Precisely so. When studying years ago the possible kind of semantical
extensions (in the field of RDF-logic), I've met a variety of options,
and explored the most liberal ones too. And the most liberal one
(well, one of the most liberals) is just to use different domain
of interpretations, and establish some very weak semantic link.
The reason is, essentially, if X is an extension of RDF then:
X-aware-processors will treat X-extensions as they like, and
rdf-aware processors will treat X-extensions just as plain RDF.
This is the bottom-line. Then, the more semantic links you put
between the domain of interpretation of X and the RDF one
(or seen differently, between their two logics), the better,
but that's just a plus.
That's why I've never digested the "extending RDF is impossible"
leit-motive, because it just starts from very tight assumptions.

Now, quickly back on
"RDF defines a large set of RDF graphs; for OWL we syntactically define a
subset W (or alternatively its complement). OWL then only applies to graphs
in W."
is just a particular case of saying you're considering two different
domains of interpretation. Like said above, an OWL-processor is free to
treat its extensions however it likes.
The objection in
to my email, just stays the same in this context: how do you define
the subset of graphs OWL applies to? In other words, how to define
"A possible characterisation of  W is that a graph is in W if there is an
unproblematic mapping of it into some DL syntax" (using 0101's words)?
But in any case, the objection in 0045 is partly off-focus: the rules
just shape the syntax of the language, and map it onto some domain
(if not pure graph rewriting, use some more powerful constraint mechanism).
If some construct is problematic wrt the semantic link to RDF, shape the
differently and forbid those instances of the syntax where you would get
problems from (see eg
for an example of such constraints).
Now, I don't view this as a problem at all: RDF can be seen just as
a syntax, and it's very reasonable that in order to define some
meaningful language you have to constraint such syntax (think eg
at programs written in C and their corresponding parsers).
There might be many rules, but that's just a natural consequence
of defining a language.

HOWEVER, there is a point (aka "0045's arguments aren't just in sky"),
and it's that all such different-domain reasonings were developed in
various flavours of RDF-logic on top of
RDF, *not* of RDFS. Therefore, it might well be the case (didn't
do a formal analysis for this), that in fact by having the
restriction that OWL must inherit all the RDFS inferences, you
get into trouble (because essentially, now the problem would
be a "same-domain" extension of RDFS, so to say...). But if
it formally turned out this to be the case, I would say the best
way to handle this would be to reshape (/restrict) RDFS accordingly,
so not to have problems (essentially, build an "RDFS+OWL" on RDF,
and as RDF has very limited inferences, there would be enough room
for such a language without any problem).

So, in a nutshell: if in this OWL extension chain, we have a problem,
it might just be RDFS and not RDF. Therefore, it would make sense
to reshape RDFS, rather than touching RDF with dark triples.

Hope I'm not being too dense here... detailed in-tech clarifications
(on the RDF-logic side) available on request, time allowing to write
them down in a decently readable form (sigh and ri-sigh...:( ).

I'll come back on the specific 101's approach at a later stage, but anyway
I see it much better than dark-triples, as it embraces the different-domains
rather than the dark-triples one. Just, beware of the "OWL extends RDFS"
that might be the real source of all the trouble...

-M (likely really dense in this email, after re-reading, sorry).
Received on Thursday, 20 June 2002 13:47:26 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:57:50 GMT