RE: Logic and Using The Semantic Web Toolbox

Pat,

you wrote:
> I was left with this impression after various conversations
> with Stefan Decker and Dan Connelly, from reading the W3C
> documents concerning RDF,  Tim Berners-Lee's writings, and
> from subsequent discussion concerning the contraints on DAML
> which arise from its having to be RDF-compliant, and from
> reading various discussions by various people on the RDF
> email archives.

Ah, but you should have read the RDF M+S spec :-)

Seriously, I believe that "semantic extensibility" means exactly that:
*adding* some semantics. It seems that if we invent more stuff, we can do
all kinds of things. I thought we were discussing what RDF was specificed to
do/be.

> Ah, I discern a ray of hope. Am I correct, then, in thinking
> that by asserting a reified description of a triple (that is,
> the four triples which describe it as being a triple plus
> giving its three components), one is NOT thereby also
> asserting the triple itself? That would greatly simplify the
> semantics of RDF, certainly.
Exactly! Asserting the reification is independent from asserting the triple
itself. And only those statements we have asserted we consider as "facts".

> But if this is correct, then what exactly IS asserted by the
> assertion of a reified description? If it is asserting that
> the described triple exists, then there is literally no point
> in saying that.

It is not asserting that the triple exists. But since we cannot assert
anything about any triple (existing or hypothetical) without "modeling" it
the way we now do by reification, it makes sense to me to have this
mechanism. We wanted to be able to say "Pat believes that the moon is made
of cheese" without saying that the moon is made of cheese.

> Perhaps an actual example of the intended use of reification
> would help make this clearer. You referred to some: can you
> cite or point to where they are described?
The applications we had in mind when we started work on RDF were:

1) PICS (the W3C content rating formalism) which requires one to distinguish
between statements about content, and statements about ratings (and rating
are statements about content).

2) Signatures: again, being able to state things about statements that are
being signed.

3) Trust models.

There is, however, one issue which worries me and would have to be
investigated (unfortunately I do not consider myself knowledgeable enough
when it comes to modal logic): In some ways, reification could allow us to
implement modal operators. Imagine the following example:

	A---x-->B---y-->C

i.e., two triples x(A, B) and y(B, C). These two triples (lets call them p
and q) have both been asserted, so we have p&q. Now, instead of p let's have
the reification of p (because we want to state some modality about p instead
of p itself). How does this modal operator distribute over conjunction? That
is, applying the modal operator to p would (maybe) require us to do the same
for q. Depends on your choice of the system of modal logic, right.

This may reveal my incredible naïvete about modal logic, but perhaps someone
more knowledgeable could clarify.

Regards,

	- Ora

--
Ora Lassila, mailto:ora.lassila@nokia.com, +1 (781) 993-4603
Research Fellow, Nokia Research Center / Boston

Received on Tuesday, 12 December 2000 14:10:58 UTC