RE: RDF Semantics: Interpretations and Modelling

>My humble opinion is that RDF (that is, the W3C recommendation) shouldn't
>define any syntax without a formal definition of semantics as well.

I agree, though political/legacy issues have obliged us to retain a 
few 'meaningless' constructions. Hey ho, one has to sup with the 
devil at times.

>If,
>e.g., bags are not to be included to the RDF Semantics, then the
>associated syntax should be informative only.

Please, keep the discussion on a level here. The RDF Semantics *does* 
provide a semantics for the RDF container vocabulary. What the 
semantics document says is that this vocabulary has no *other* 
semantic conditions over and above those implied by the basic RDF 
model theory in sections 1 and 3.1. But this is a perfectly sharp, 
formal semantic specification in itself.

>Let the blank nodes and
>properties (defined elsewhere, perhaps in a OWL kind of working group) to
>do the trick.

The container membership properties DO do the trick; that is the 
point. They need no other "extra" semantic conditions to be added to 
them. Contrast this for example with rdfs:SubClassOf, which needs 
quite elaborate 'extra' semantic conditions to ensure that it is 
transitive and properly related to rdf:type.

>The core should be as simple and clear as possible.

I think we all agree on that, but 'as possible' has different 
meanings for everyone :-).

>  (I guess it's too late
>to get rid of all the alternative ways of serialising RDF in XML :-) )
>
>.. *blink* ..
>
>Reading the bag discussion I begun to wonder if the (restricted) idea of
>"assertions" is appropriate in the context of RDF in the first place.
>
>In several contexts, RDF is described as an "assertional language"
>asserting "propositions". In practice, RDF Semantics begins with this very
>phrase. This is fine if we consider only simple entailments.
>
>However, given the possibility of making semantic extensions of RDF,
>emphasising the property "assertional" seems unnecessary (and perhaps
>misleading as well): introducing new semantic conditions and combining
>different vocabulary entailments (the useful scenario) probably leads into
>RDF graphs which could be described as hybrid networks (when associated
>the appropriate processing rules).

The reference you cite below says: "What is common to all semantic 
networks is a declarative graphic representation that can be used 
either to represent knowledge or to support automated systems for 
reasoning about knowledge."
which seems to me to be a restatement of the notion that these are 
assertional languages. The assertional nature of RDF was in any case 
fixed by our charter.

>
>To ground this discussion, a nice, easily accessible classification about
>various "kinds" of semantic networks can be found, e.g., at
>
>	[1] http://www.jfsowa.com/pubs/semnet.htm
>
>(Note: I'm NOT assuming the people reading my mail don't know the terms, I
>simply give the above address so that there would be something at hand to
>ground the terms to.)

I agree this is a very nice intro, but Sowa is rather all-inclusive 
here, eg he counts neural networks as a kind of semantic network. 
Most authors would restrict 'semantic network' to refer to graphical 
notations for assertional languages, such as Sowa's own CG notation, 
based on Pierce's pioneering work (which are most emphatically not 
'hybrid'.).

>Following the concepts in [1], I believe that most RDF applications would
>treat RDF graphs as hybrid networks for doing all sorts of things.

I would not expect this. Or at any rate, if it does happen, then it 
is using RDF in ways that are not supported by the RDF spec (which 
might well happen, of course.) . RDF isn't really meant for DOING 
things, its a notation for CONVEYING content. Think of it as a medium 
for exchanging machine-usable content on the Web. It really is not a 
programming language. Now, in practice, Im sure people will use it 
for all sorts of things; but we have to write the spec with an 
intended set of uses in mind.

>That's
>for RDF is for, a framework?

For *defining* resources, not for building arbitrary processes.

>
>If there's something that prevents, e.g., a group of mathematicians
>defining an RDF extension suitable for doing ordinary first-order logic
>(*),

Not only is there nothing to prevent this, a small group of people 
which includes myself and John Sowa are in the process of doing this 
right now for CL, a slightly extended version of full FOL. We hope to 
have a preliminary report ready by July 2003.

>that design choice of RDF should be spelled out with BIG letters in
>the cover page of RDF specs. As I see it, I can always choose not to use
>[or invent a pathological model for that part] a specific, too restricting
>vocabulary if I don't want to. Or is this prohibited somewhere in RDF
>specs?
>
>I understand that the nature of "assertional propositions" states that
>there's no superior authority who can say which assertions (and the
>associated semantic extensions) are correct and which to use, but to me,
>the same argument applies to all other languages as well (i.e.
>"first-order logic could thus be considered assertional since you can see
>it 'only' as a set of asserted conventions (**)" [but then, this point of
>view will probably get you in a trouble with your math teacher...]).

Never mind the math teacher, but the key point for us is that the 
language needs to be specified *by the spec* in order that we can 
have an agreed-on semantics as part of the interoperability 
requirements. There is more going on here than just agreeing on a 
syntax: to do that we would just need XML.

>
>--Ossi
>
>(*) Well, I did suggest that the RDF model theory should use only
>countable universes... :) ..so that the fuzziness related to "ordinary"
>model theories

Can you spell out what you mean by 'fuzziness' here?

>could be technically avoided (I dislike the attempt to
>formalise sets that are not well-defined in the procedural sense).
>
>(**) You could easily invent a RDF-based syntax, e.g., for any first order
>language (predicate and function symbols would benefit from a definition
>of lists...).

Indeed, and OWL (not full FOL but much closer than RDF is) is done 
this way.  But it is somewhat more tricky to arrange things so that 
the FOL being thus described is in fact an RDF *semantic* extension 
also.

Pat
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam