Re: Valid representations, canonical representations, and what the SW needs from the Web...

Paul Prescod writes:
> People are always talking about using web pages to represent concepts as 
> if we were going to use _pre-existing web pages_ to represent concepts 
> in the semantic web. In my opinion that's wrong because the semantics of 
> those pre-existing web pages will be way too vague. Clarifying semantics 
> is the whole point of the semantic web. Therefore what is much more 
> likely is that semantic web concepts will be represented by RDF 
> documents which are explicit about what they represent. It won't be that 
> you use GM's home page to represent GM and I use Ford's. Much more 
> likely, we will both point use a URI hosted by NYSE designed 
> _explicitly_ to represent Ford and GM as their stock symbols do.
> Similarly, I don't think that the page that represents "me" (at least 
> for FOAF purposes) is my HTML home page or SMTP address, but rather my 
> FOAF.rdf page.
> Yes, there exist URIs with ambiguous underlying semantics. That's what 
> the semantic web is supposed to solve. It can't presume the problem to 
> be already solved as its starting point.

I see what you're saying.  Yes, we can treat URIs as being
by-and-large imprecise identifiers.  (Yeah, each one points to exactly
one resource, but it's unlikely that anyone will have a good idea
exactly what that resource is.  So it appears imprecise.)  The
existing ones would be essentially useless for the semantic web, as
you say.

In this view, the use of URIs (esp http URIs) in RDF buys us almost
nothing but confusion.  At best, it makes looking up some
documentation for a term a little easier.  Really, though, UUIDs or
tag: strings would be vastly less confusing and work about as well.
Google or a few well-run directory services would provide
documentation links, and could actually lead to better updated &
maintained documentation after the term-coiner has lost interest (and
his domain name :-)

There's an alternative view, however.  Maybe http: URIs are already
rather precise.  Each one identifies exactly one web location [1].
They can't be used directly to identify the Sun, etc, only indirectly
via their content (or using fragment IDs, in one version); that
indirect location works just fine for RDF and the semantic web,
letting us bootrap off the existing billions of pages instead of
waiting for new ones.

In other words, there are some existing mappings from URI strings to
things of interest.  Some of these mappings are already out there and
working quite well.  Why not use them?

    -- sandro


Received on Monday, 3 February 2003 15:32:23 UTC