Re: resources and URIs

> Is it the webcam itself, or the view from the window, or (suppose the 
> webcam is in Peoria) Peoria?
...
> It emits representations OF a view in Peoria, true; and the owner of the
> webcam might indeed declare that his site is about Peoria; but that is a
> semantic claim, not a network claim.

To continue your webcam example: embellishing the representations with
appropriate metadata about the representations *themselves* seems a lot
more constructive than trying to assert properties about resources.

For example, the HTML may include machine readable metadata stating the
author, a brief description of the page, appropriate keywords, a link to
the next page in some sequence defined by the author, and so on.

The webcam images referred to from the HTML may include metadata stating
the time they were taken, the equipment used to create them, a textual
description of their content and so on. Alternatively, the image metadata
may be located in the HTML in the form of attributes on the link.

All of this metadata is describing representations, either those just
retrieved or those that will be retrieved in the future. None of it is
talking about resources, and URIs are only used at all to refer to the
representations that will be returned by dereferencing them using HTTP.

Doesn't this seem more useful than trying to decide whether the URI
identifies/denotes a document, a webcam, or Peoria?

<hypothesis>

The current design of the proposed semantic web relies on URI triples
rather than hypertext markup, which I believe discards most of what made
the web work.

If the design of the semantic web was adjusted to encourage the use and
interpretation of textual markup with less reliance on overstressed and
underspecified URIs the result would be a clearer and more powerful
architecture that did not clash so awkwardly with the existing web.

Some principles along those lines would be:

 * Only use fragment identifiers to refer to fragments of a representation
retrieved using the URI. This is what they were created for, after all.  
Using them to refer to non-existent parts of non-existent documents is not
doing the web any favours.

 * Move more complicated linking schemes such as XPointer into the linking
markup, to solve a host of issues by simplifying URIs, simplifying
encoding/escaping and allow more interesting linking patterns to emerge.

 * Stop trying to pin down arbitrary concepts using a unique URI. It is
not necessary for there to be a canonical URI to identify "the Porsche
911". It is sufficient to be able to say "the car in *this picture*" or
the car "described in *this advertisement*". Question: does tel:555-1234
identify a company, a department, a telephone handset, the person who
answers it, the employee role of the person who answers it, the person who
is *supposed* to answer it, or none of the above? Answer: it doesn't
matter! Because any reasoning agent can say "the company whose number
is..." or "the employee who answers when you call..." or any other
necessary clarification. If telephone numbers do not identify a unique
concept, neither do URIs.

 * Take advantage of markup as a primary descriptive tool, not just a
literal string hanging awkwardly off the end of an RDF predicate! But
perhaps other people can carry this aspect of the discussion.

</hypothesis>

Best regards,

Michael

-- 
YesLogic Prince prints XML!
http://yeslogic.com

Received on Thursday, 24 July 2003 00:03:09 UTC