suggestions for changes in RDF and RDF Schema

Over the past few days there has been a flury of comments about RDF and RDF
Schema, with Pat Hayes, Drew McDermott, and I pointing out problems that we
see with the current situation.

In response to our criticisms there has been some call for constructive
criticism.  Therefore, in the spirit of constructive criticism, I have
formulated a list of changes that I would like to see made to RDF and RDF
Schema.   This list does not contain complete arguments for these
changes---complete arguments would turn this from a short note into a long
journal article, and, moreover, many of the arguments have been at least
outlined recently in this group.

Not having participated in any W3C working group I have no idea how likely
it is that any or all of these changes will be considered or made.  Even if
these changes are unlikely to be made, however, they should serve to give a
good outline of what I see that is wrong with RDF and RDF Schema.  Make of
them what you will.

Peter F. Patel-Schneider




	A Personal View of Changes to Make in RDF and RDF Schema
	to support good representational practices

		Peter F. Patel-Schneider
		Bell Labs Research

Here are a list of changes that I would like to see made in RDF and RDF
Schema that, in my view, would make them much better representation
systems.  I am not stating that the result would be an ideal underpinning
for the semantic web in my view, just that the result of the changes would
be much better for this purpose than RDF and RDF Schema currently are.  I
am also not stating that if these are the only changes that I think need to
be made to RDF and RDF Schema to make them suitable, just that these are
the important ones that come to me just now.

1/ Stick to a single purpose

   RDF, in particular, tries to exist on too many levels.  It attempts to
   be a universal low-level mechanism for transferring information encoded
   in triples, but also includes some very high-level constructs, such as
   reification.  I would like both RDF and RDF Schema to try to do one
   thing, but do it well.

   I think that it would be much better to have RDF stick to representing
   information in the form of ground triples of URIs.  This would require
   removing all the talk about collections and reification from RDF.  RDF
   Schema could then be concerned with classes and properties, and, maybe
   collections, but without alternatives and distributive statements.
   There are also a number of smaller things that should be removed from
   RDF Schema.

2/ Don't grab all content

   RDF provides a meaning for everything that it sees, it turns everything
   that it sees are triples, and it gives a meaning to every triple.
   I think that RDF should be changed in one of two ways:

   1/ Let RDF give a meaning to everything that it sees, including RDF
      Schema triples, but make RDF not be representational, in the sense
      that the RDF model is not supposed to represent informaition, but is
      instead yet another data structuring convention.  Giving
      representational meaning to (some of) the triples would be the job of
      some other formalism.

   2/ Restrict RDF to giving meaning to triples whose predicate is defined
      as an RDF predicate.   RDF would not give meaning to other triples,
      which could then be used in extension formalisms.  Alternatively, RDF
      could allow other things besides triples to show up, and these
      non-triples could be used in extensions.

   Of these two, I favour the second, as there are already lots of data
   structuring mechanisms.  RDF should do something representational.

3/ Make RDF Schema monotonic

   RDF Schema has some non-monotonic features, including domain.  In this
   context the non-monotonicity comes from the fact that you can process
   part of an RDF Schema collection of documents and have an error (in this
   case a triple has an invalid resource as its subject) but later find out
   (by means of another domain statement) that the resource is actually in
   the domain of the property of the triple.

4/ Semantics
   
   Both RDF and RDF Schema need a good semantic foundation.  This can be
   provided in a number of ways, such as the KIF axiomatization of RDF and
   RDF Schema.  The reason that semantics is last, is that the semantics
   would be for the new RDF and RDF Schema, not the current ones.

Received on Tuesday, 3 April 2001 06:40:22 UTC