Re: DTTF: high-level summary (attempt)

From: Massimo Marchiori <massimo@w3.org>
Subject: DTTF: high-level summary (attempt)
Date: Thu, 2 May 2002 12:11:51 -0400

> Ok, I don't know to what thread to reply, and how to avoid sending 4
> different ones... 
> Well, let's just try to draw a line, and see where we are.
> 
> Is is true there's some weak consensus that, bad or good, we can try to
> leave RDFCore out, and work out the solution in WOWG? Or, is there still
> somebody saying he can't live with this? This is useful to do a first
> little step: let the semweb cg know that this dependency on RDFCore is
> discharged. 

Well this would be fine by me, but I believe that it would go outside what
I believe to be the currently-acceptable parameters of the WOWG situation.
Again, I am perfectly happy to change these parameters, but I am unwilling
to put (more) work into a solution that is not compatible with the current
parameters.

> Secondarily, the base two points. I think there are two main issues
> boiling up in all the threads.
> a) One is Peter's point, about paradoxes and expressive power.
> b) One is Pat & Peter point, about semantic extensions.
> 
> They are somehow related (b somehow subsumes a), yet can be usefully
> distinguished. 
> a) essentially says, look, what we have is very powerful (essentially, graph
> construction), and we risk that the resulting language is not well defined
> (which implies, yes, a disaster).
> b) essentially says, there are big troubles in general, to do a semantic 
> extension of RDF.
> 
> Now, I think all the DTTFers are right (errr, is this a paradox? ;) when they
> claim respectively that a) and b) can't be solved with RDF as is (RDF: bad),
> and that a) and b) can in fact be solved.
> They are all right, because each starts from different assumptions, making
> their reasoning perfectly valid, and resulting in us clashing on the
> conclusions, rather then on the premises.

> Now, on a) and b), the personal viewpoint on "how it can be done" without 
> changing RDF (so, the context).
> On a):
> + Yes, this is a big problem. but a way out is to define the semantics of OWL
> using graph rewriting (cf.
> http://lists.w3.org/Archives/Public/www-rdf-comments/2002AprJun/0018.html 
> for a subset of a possible definition for RDF semantics). This can be seen as 
> "considering the right syntax" (so, similar to Peter's syntactical
> constraints, only that the constraints are not on the graph, but on the
> rules used to extract the semantics).
> This works because it doesn't follow the "everything denotes", which is
> the understood assumption that, I think, is based on the
> RDF-cant-work-here argument (which is correct, under this assumption!).

I am not convinced that this approach work actually work.  You would need
quite a number of rules.  For example, you would need rules that could
determine membership in classes.  What sort of a rules would you have to
have to make the determination that john is in the intersection of student
and employee from the premise that john is in the intersection of employee
and student?  I believe that the rules that you end up with would be
analogues of inference rules, and that they would end up being
inconsistent, because they would have to generate and consider the
paradoxical situations.

> Onto on b):
> And, yes, this is related to the semantical extension problem. I think
> the understood assumption of the RDF-cant-work-argument has been that you
> have to model the semantics of an extension *using the same domain* (so,
> essentially like a semantical embedding).  And under this assumption,
> yes, this is correct, RDF can't probably be used here (unless changing it
> with dark triples or similar).  The way out here is to drop the
> assumption to use a semantical embedding: the domains of interpretations
> can be different. Doing so, we earn in flexibility of extension, and we
> lose something in semantic interoperability.

Yes, something has to be done, of course, but what?

> I can provide extensive details on both a) and b) on demand, but before
> doing so (i.e., debate on the correctness of a specific assumptions =>
> conclusions approach), it's way more important to notice the assumptions
> we come from. Under some assumption (that were natural and almost
> understood for me, but are likely not at all for others) there is a way
> out.  Under others assumptions (reasonable as well, just different),
> there's likely no way out other than doing some heavy changes to RDF
> (personal view, but I do agree with Peter and Pat, once realized the
> context/assumptions they are reasoning in).

Well, yes, there are lots of modifications to the RDF world view that make
the layering problems go away.  They have been already laid out, and boil
down to doing one or more of 
- go beyond RDF syntax (so that not all syntactic constructs have to denote) 
- don't use RDF semantics (so that not all syntactic constructs have to denote) 

> Is this a fair high-level summary?

I did find it a bit confusing at first, but I think that now understand it.

> -M

peter

Received on Thursday, 2 May 2002 14:41:07 UTC