Layering on the Semantic Web

In various emails to the Working Group, and also in various off-line conversations, a number of people have become very worried about the layering of OWL on top of RDF and RDF Schema. 

The problems can be summarised as follows:
- RDF is not well suited as syntax carrier:
  - the problems that Peter has identified with "additional" tuples 
    (which specify only syntax) ending up in the RDF model and breaking the inference
  - problems with scoping
  - its inability to constrain the syntax adequately
    (e.g., the problem with strange combinations of restrictions
     which is responsible for much of the ugly DAML+OIL syntax).
  - general unreadability by human processors
- RDF Schema has some unconventional features in its meta-model. 
  These were left out of DAML+OIL to keep the language simple, 
  well understood on formal grounds and non-controversial, 
  and now threaten to creep back in again. 

Our conclusion is that these problems are much harder to solve than originally anticipated.
Therefore, we propose to take another route for specifying the syntax of OWL:

1. the syntax of class and property definitions in OWL (the ontology) is specified in XML
2. this is done in such a way that RDF can be used to specify instances of the ontology
3. we try to do 1. in such a way that significant parts of RDF Schema 
   end up as a sublanguage of OWL 

The advantage of using XML are both technical and (not unimportant) also political:
- XML is well suited for specifying syntax (in fact, that is its main goal in life)
- it comes with a host of additional technology and standards that can then be exploited
  for OWL (XLink, XPointer, XPath, XQuery, XSLT, etc)
  (we can think of useful applications for all of these)
- it will make our work immediately relevant to all of the XLM community. 
  They share many of our goals, but there is a constant danger that they 
  will use different (XML-based) technology, instead of RDF based technology.

RDF is still being used for what its good at: specifying ground data-structures (the "Abox" of old). 

Whether 3 is achievable or not must be investigated.
If it is achievable, RDF Schema will not be a sublanguage of OWL, but parts of it will be. 

We suggest that this proposal is discussed at the face-to-face meeting. We will try to have a concrete proposal for (1) on the table before then. It will be helpful if we can get reactions on this proposal well before then (ie. now:-). 

Ian Horrocks,
Peter Patel-Schneider,
Frank van Harmelen.

Received on Thursday, 3 January 2002 10:46:56 UTC