RE: Dark triples motivation

We had what I found to be a helpful discussion after the RDF Core telecon on
dark triples.

It made me reconsider what are the drivers for the problem.

Previously I posted the following Problem Statement [1], and had a positive
ack from Pat.

>Problem statement:
>
>Layering on top of RDF involves using graph syntax to encode OWL concepts.
>If such graph syntax is asserted then the RDF model theory assigns meaning
>to such syntax specifically denoting things in the domain of discourse that
>correspond to the syntactic expressions of OWL. Given this, an OWL model
>theory with appropriate entailments (in which the theory of classes in OWL
>is recognisable as a subtheory of some well known set theory) is difficult
>and problematic to construct.


I have a feeling that this does not do justice to the divide between two
different schools of thought, roughly of "computer scientists" vs
"mathematicians" (or perhaps "hackers" vs "theoreticians"). At some level we
seem to touch on one of those religious wars which DAML+OIL delicately
balances.

Using a langauge as its own metalanguage is known to be impossible. Hence,
the theoretically minded don't even try. Hackers see it as a challenge :).

RDF M&S follows the hackers' tradition. In as much as RDF metatheory can be
encoded in RDF, RDF M&S tries. In fact, it seems to go too far. e.g. the
notion of reification is a mess in M&S.

The model theory for RDF (edited by Pat) superbly captures RDF M&S intent,
by having properties that have special semantics according to the theory,
within the model, also behaving as normal properties. This is unusual from
the point of view of language design, particularly mathematical language
design. A type statement in a language is typically very different from a
user-defined property; and the ability to take a subproperty of the type
property (in RDFS) is surprising. Thus RDFS uses the technique of having
"things in the domain of discourse that correspond to the [special]
syntactic expressions".

My view is that the RDF Model Theory shows that, when done well, this
technique does allow a significant portion of the metalanguage to be encoded
in the language; which may make it easier to extend. This also satifies the
hacker community "eat your own dog food" mantra.

Essentially I am now reading the problem statement as the WebOnt WG being
unhappy with that game plan because:
- there is theoretical risk of inconsistency etc. (Illustrated by the
Patel-Schneider paradox)
- the existensability is not viewed as valuable (I would not be surprised if
some members see it as a liability)


I think the original paragraph [2] from our f2f which called out
"entailments" as an issue, was probably misleading. I do not believe that
identifying the desired entailments and a solution like [3] will meet the
needs of Pat and Peter.
I am currently seeing them as needing a clear separation between OWL syntax
and OWL semantics; which is fundamentally obstructed by the use of asserted
triples.

DAML+OIL appears to be at a very delicate compromise between these two
opposed positions. The axiomatics semantics is on my reading consistent with
encoding language significant syntactic structures in the semantics. The
model theory, in contrast, seems to go from the graph syntax to the model,
without going through the image of the graph syntax in the model. There is
no daml:Ontology for DAML+OIL (although the axiomatic semantics suggests
that there could be one).

Jeremy

[1] Jeremy's summary of Problem
http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0128.html

[2] Amsterdam paragraph
http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0164.html

[3] A non-dark triples "solution"
http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0155.html

Received on Monday, 22 April 2002 12:28:53 UTC