getting the ball rolling (or at least providing a target to shoot at)

Here are some thoughts on the meaning of URI references that I recently put
together.

Enjoy, and happy hunting.

peter



		Semantic Web Meaning of URI References

		Peter F. Patel-Schneider
		Bell Labs Research


Suppose you have a Semantic Web document that contains a URI
reference.  What is the meaning of this URI reference?   

As far as RDF itself is concerned, the meaning of the URI reference is
defined by the RDF model theory.  In the RDF model theory the meaning of
the URI reference in an interpretation is some element of the domain of
discourse.  If the domain element is a property, then there is additional
meaning provided, namely the extension of that property.  (RDFS extends
this meaning for classes, in some sense, by using part of the extension of
the meaning of rdf:type to define the class extension of the domain
element.)  All of this is, of course, constrained by the rest of the
information in an RDF graph in which the URI reference is contained.

This is the entire meaning of the URI reference in RDF.  No
information from anywhere else is used.  Note that the RDF meaning is
thus relative to an RDF graph.  This is really no different from the
situation before the advent of the RDF model theory, as the RDF meaning was
then the RDF data model.


Can one do better than this?  In particular, is there a way to provide
information about what might be called the intended meaning of a URI
reference?

OWL goes further than RDF here in two ways.  First, OWL is a more
powerful formalism than RDF and thus when working in OWL there are
more kinds of constraints that can be placed in documents.  Second,
OWL has an imports statement that, in essence, allows multiple
documents to be combined.  However, the meaning of a URI reference in
OWL is still relative, as it depends on which OWL document or
documents are under consideration.  So OWL does not provide a way of
getting at *the* intended meaning of a URI reference.

One way of providing intended meanings for URI references in the
Semantic Web would be to require that part of the formal meaning of a
URI reference be the constraints embodied in the document that is
accessable by looking up the URI reference (without any fragment) in
the World Wide Web.


Is this useful?  I claim that it is not, and for three reasons. 

First, the document retrieved could be in some language that is not
understandable by a particular tool.  For example, an RDF tool might
obtain an OWL document or even a natural language document when
looking up a URI reference.  It is not possible for the RDF tool to
understand all of (or even any of) the meaning of the document.  So
this view of the meaning of URI references needs to be tempered by
requiring only appropriate documents be incorporated into the meaning
of a URI reference.  This means that the meaning of a URI reference
depends on which Semantic Web language is being used.

Second, there might be several documents available, due to content
negotiation, and these documents might (and probably will) incorporate
different meanings for the URI reference.  This is actually a good
thing, as it allows several versions of the meaning of a URI reference
to be available.  Each Semantic Web language could provide different
rules for which document to use.  This again means that the meaning of
a URI reference depends on which Semantic Web language is being used.

Third, and most important, requiring that the use of a URI reference
pulls in a particular document, even if this document varies depending
on which Semantic Web language is begin used, would eliminate many
possible uses of the Semantic Web.  In essence, it means that sharing
URI references requires sharing world views.  Two agents in the Semantic
Web may disagree about the precise meaning of a URI reference but
still want to communicate.  Under this view of the meaning of a URI
reference they cannot.


So how to proceed?

Perhaps only a part of the meaning of the document above could be the
meaning of a URI reference.  However, this weakening is not viable
because it is just not possible to determine what part of the document
is relevant unless the document has several completely disconnected
groups of information, which is very unlikely, or the language allows
the indentification of groups of information in a document, which
neither RDF nor OWL do.  Even if this was possible, it would still
fall prey to the third problem above.

So it is not useful for there to be a single meaning for a URI
reference.  Does this mean that there is no way of obtaining consensus
on a preferred meaning?  Of course not!  The above method of obtaining
a document provides a fine way of providing the intended meaning of a
URI reference.  All that is required is that applications are free to
ignore that meaning if they so desire.

Under this view of Semantic Web meaning an application is (relatively) free
to retrieve the above document and use its meaning when it utilizes a URI
reference.  However, not all applications need do this.  If the meaning of
the above document is required for some purpose then the importation of
this document can be made explicit in some body of information.

This solution appears to me to have precisely the right
characteristics, as long as documents are reasonably written.  The
mere use of URI reference does not commit an application to the
preferred meaning of the URI reference, but the importation of a
document commits an application to the entirety of information in that
document (and also all the documents that it imports).

Some have argued that requiring that the meaning of an entire document
be used here is too broad.  The solution to this problem is easy -
just use smaller documents, tied together in whatever way is
appropriate.

There have also been arguments that this sort of importation can
easily result in pulling in the entirety of the Semantic Web.  Yes,
this is possible if all documents in the Semantic Web are so linked.
However, requiring that the mere use of a URI reference commits an
application to the intended meaning of that URI reference is very
likely to pull in much more than explicit importation.  Care is indeed
needed when writing documents to not pull in too much, but any other
method requires both more support, such as methods for grouping
information in documents, and even more care in writing documents.

Received on Friday, 19 September 2003 13:37:03 UTC