Identifying Things (on the semantic web)

The problem: How do you talk about something when you don't already
share with your audience a common identifier for it?

Proposed answer: Find a document which 
  - can be identified to your audience
  - contains some text which, in the document, identifies
    the thing in a way your audience understands.

Then identify the thing with the pair (document identifier, text of
thing-identifier).   And if we use URI-references as document
identifiers, we have a pretty good system.    (URIs would work, but
it's easier to find a unique identifier in big documents if we allow
fragments.) 

Examples: I can be identified quite clearly as: "The thing called
'Sandro Hawke' on the web at http://www.w3.org/People/Sandro/".  
And a "URI-reference" can be identified as "The thing called
'URI-reference' in http://www.ietf.org/rfc/rfc2396.txt". 

I think people have been trying to approximate this by using
identifiers like
   http://www.ietf.org/rfc/rfc2396.txt#URI-reference
but that's a poor approximation because that construct already has a
different meaning.   The two meanings can perhaps be made to converge
by constructing documents with the right kind of A NAME= or DIV ID=
elements, but I think there would still be some semantic overloading
happening.

If you really need a single URI to identify your concept, then
construct one which combines the two elements, like
  http://www.w3.org/2001/01/19/idpair/URI-reference/http%3A%2F%2Fwww.ietf.org%2Frfc%2Frfc2396.txt

In either case, an audience member (eg a machine) can/should only
infer that two identifiers are for the same thing if the two strings
in the pair (or the one string if its combined) are identical.

A good document for these purposes would, of course, not change over
time in way which might alter the meaning of the thing you're using it
to identify.

  -- sandro
     (writing my first message to public w3c list...)

Received on Friday, 19 January 2001 16:27:50 UTC