Krivov's question: "Why RDF?"

re: http://lists.w3.org/Archives/Public/public-webont-comments/2004Mar/0020

I provoked some discussion of this matter on the the DAML/EU Joint
Committee list recently.  (The question there being whether or how to
layer a rule language (eg SWRL) on RDF.)

   Peter F. Patel-Schneider:
      I argued long and loud in the W3C WebOnt working group about
      problems that using the RDF syntax caused.  This argument didn't go
      anywhere, so I gave in and created a partial solution for OWL.  

   Sandro Hawke: 
      Do you remember why the WG disagreed with you?

   Peter F. Patel-Schneider:
      Because all Semantic Web langauges have to be same-syntax
      extensions of RDF. 

   Frank van Harmelen: 
      Yes, I must support this. The *only* argument for many WebOnt
      members to accept/put up with the RDF syntax for OWL was
      political pressure (perceived or real) from W3C.

         -- http://www.daml.org/listarchive/joint-committee/1639.html

I wasn't in the WG for those discussion, and I suspect the history
isn't as important as the future.   For people developing a rule
language these are important issues for the future; for this working
group there may be some important explanations or lessons that
can be offered, perhaps in response to Prof. Krivov's question.

Here's a strawman answer:

   > I completely do not understand why RDF is necessary as an
   > intermediate layer between XML and OWL.

   The short answer is that it's not theoretically necessary, but
   after weighing the options and issues, the working group decided
   that OWL would be most useful for the web community if constructed
   in this way.  The abstract syntax for OWL, or the XML syntax you
   mention, would certainly work for expressing OWL ontologies, but in
   the end they would not support the evolution of the Semantic Web
   quite as well.

   The arguments against they layering are fairly well known, but in
   terms of theoretical challenges and in terms of parsing, as
   explored in "Parsing OWL in RDF/XML" [1].

   On the other hand:

      (1) Relying on RDF poses no additional burden, because the
      properties and classes offered by an OWL ontology are offered
      largely for use in RDF instance data.  This means that
      programmers, software systems, and users working with OWL can be
      expected to be already comfortable with RDF and convinced of its
      value.  OWL users are likely to see RDF as a natural part of
      their system (and visa versa), while XML itself may be
      irrelevant to them.  (This does not justify the OWL syntactic
      layering, but it does help explain why the presence of RDF isn't
      as expensive as it might first seem.)

      (2) RDF users may want to use small bits of OWL that fit
      naturally into their RDF.  Their entire "ontology" may consist
      of an owl:sameAs triple, or one owl:InverseFunctionalProperty
      statement.  Why should they have to switch to another language,
      when these simple bits (like all of RDFS) fit elegantly into
      RDF?  If they go on to define Restrictions and other complex
      forms, there is no sudden jump to a new language, just a gradual
      use of more difficult concepts and constructs.

      (3) RDF systems are expected to become very sophisticated in
      merging data from web data sources, with caching, provenance
      tracking, publish/subscribe features, trust reasoning, etc.
      If OWL ontologies are just more RDF data, they can more easily
      provide these services for the OWL data needed in reasoning
      about the RDF data.

   There may be more arguments, of course.  It's not clear whether
   these are the arguments that swayed the working group when it first
   approached this question.  The group was largely following along a
   path which included RDFS, DAML-ONT, and DAML+OIL, and it may also
   have been influenced by the vision of RDF as the data-bus of
   semantic web [1], but one need not subscribe to these notions to be
   convinced.

   The bottomw line is that we expect certain practical advantages to
   result from this layering approach, and we think they outweigh the
   difficulties in parsing OWL from RDF/XML.  We expect that in due
   course these difficulties will be well understood and few people
   will have to deal with them directly.

   [1] http://www.w3.org/2000/Talks/1206-xml2k-tbl/slide10-0.html

Or something like that.  :-)

    -- sandro

Received on Wednesday, 10 March 2004 17:26:36 UTC