RE: Documents, Cars, Hills, and Valleys

Bill de hÓra wrote,
> Miles Sabin wrote,
> > And tho' I'd agree that it's reasonable to assume that any use of
> > a   URI is _intended_ to identify a unique resource, in the
> > presence of   multiple uses and multiple intentions the net
> > result is that there is   no one resource which can be taken as
> > the referent ... IOW, the bare   URI is ambiguous.
>
> Yes. We'd need a theory of reference or pointing, or symbol
> grounding, no?

I sincerely hope not ... I don't think we want to be hanging around
for another couple of millenia ;-)

> Or we could just impose order and have a semantic web of Newspeak. 

I don't believe this is either achievable (or desirable) outside of
closed systems.

> Another option is to and allow code to reason statistically or 
> negotiate term grounding about whether a URI is being used to refer 
> to the same thing. Dan Brickley has talked about the social compact 
> aspects of URIs before, but the only way I can conceive of running 
> such common sense though a computer is statistically or 
> probabilistically.

In many, perhaps most, cases there's simply no problem, because 
implicit context is sufficent to resolve any practical ambiguity (eg. 
in the context of a GET, http://www.markbaker.ca/ will only ever refer 
to a document, rather than to Mark in the flesh). In others we have 
more work to do, but who says it has to be fully automated? Given a 
sufficiently rich variety of disambiguation mechanisms we could hope 
to rely on content producers and consumers to manually maintain 
metadata expressing their referential intentions.

I grant you that a semantic web composed of a shifting patchwork quilt 
of partially manually maintained metadata is perhaps less exhilarating
than the seamless fully automated vision. But in many ways it's more
consistent with the web as it actually exists, and it might actually 
be implementable.

Cheers,


Miles

Received on Wednesday, 24 April 2002 09:12:03 UTC