Re: Anonymous resource names -versus- variables

Hi Sergey,

On Thu, 4 May 2000, Sergey Melnik wrote:

> Guha wrote:
> > 
> > Yes, Anonymity is not a property of the object, but a property
> > of the description of the object. The situation is similar to
> > indexicals. I can use "he" to refer to Dan, but that does not mean
> > Dan is of type "he" or "indexical".
> > 
> > guha
> 
> Guha, Dan,
> 
> although you are making a valid point, it seems that we are approaching
> the edge of metaphysics and axiomatizability of RDF. I tend to view RDF
> as a logical assembler language (which you can still associate semantics
> with), rather than the expression of the semantics itself.

If you mean that there is plenty still to build on top, and plenty to
clarify (esp. around URIs), I'm in agreement.

> 
> You are arguing that Anonymity is not a property of the object. Is being
> an instance of a class really a property of the object? Furthermore, if
> you are concerned about the "correctness" of RDF, have a look at
> rdf:Seq. Can we seriously claim that, for example, an instance of
> rdf:Seq can be a creator of a resource?

I agree that RDF has very sensibly backed away from making any strong
claims about the "intrinsic"ness or otherwise of properties, eg. we are
nicely agnostic about classic philosophical rathole of whether 
"colour" etc is genuinely a property of an object or contextual. 2000+
years is a long time to wait for rough consensus let alone running
code... ;-)

All that accepted, I still thing you might be mistaken, and for very
practical reasons. (Actually I think we're both proposing cludges that skirt around
some Web architectural design issues that need clarification...)

eg:
A "smart" RDF store might be able to do joins / aggregation using
principles other than sameness of URI. For example, two anonymous mentions
of the resource whose util:personalMailbox is mailto:connolly@w3.org,
combined with knowledge that util:personalMailbox is an individuating
property of a resource, can allow us to do the join. 

Whatever deep philosophical approach we take to question of whether
properties are really "of" a resource, it strikes me that in the case
above, we're certainly talking about the same resource. A 3rd mention of
that resource which _did_ provide a URI is still talking about the same
resource. Now if we ask whether the resource is of rdf:type Anonymous,
that's not a sensible question. It depends on too much context.
However if we ask whether some software agent or RDF database util:stores
some web:uri for that resource, that seems to me to be a more sensible
question. So I'm not averse to expressing these things in terms of RDF
properties instead of as a URI scheme hack, I just don't think the model
you propose is intuitive, since the meaning of being of rdf:type Anonymous
is highly contextual, and you're not exposing that context in the model.

> To address typing and order, "semantic transparency" wrt rdf:type,
> rdf:Seq and ordinals is required. Those can be viewed as building blocks
> of a low-level (structural) language.
> 
> BTW, as simple as RDF looks like, it can still be split into two layers:
> 
> 1) Object identity + binary relationships (no predefined vocabulary!)
> 2) Basic typing (rdf:type), order (rdf:Seq, rdf:_1, etc), n-ary
> relationships (currently missing)

Also datatyping and xml:lang -- properties of literals. 

But yes, there is a layering there. Issue is the complex relationship
between object identity and the data structures (URIs) that we use to
exchange information about object identity. This is just more (meta)data
but slippery to represent since we take object identity as a primitive
building block for representing data.

> Both approaches to anonymity (var:... and rdf:type AnonymousResource)
> have differences from the operational (rather than theoretical)
> viewpoint, too. Given a generic RDF repository it may be significantly
> simpler to ask for (X, rdf:type, rdf:AnonymousResource) than to look at
> the syntactic representation of IDs which may require building an index
> over strings.

Earlier, I was about to make the opposing claim about your scheme, but
changed my mind. Any implementation can make some special case treatement
of this issue. I was tempted to claim that the storage requirements for
using var: URIs would be less than that of storing rdf:Type Anonymous
triples, but realised that was pretty bogus, since these could be
represented behind the interface by the same machinery. An RDF database
could expose either or both approach and could in practice use either (or
some alternate strategy) for actually storing this info.

cheers,

Dan

--
mailto:danbri@w3.org

Received on Thursday, 4 May 2000 15:56:05 UTC