UPDATE: document on layering OWL on top of RDF

Here is the action item document on layering OWL on top of RDF.

peter





		Layering OWL on Top of RDF


	Peter F. Patel-Schneider
	Dieter Fensel
	Ziv Hellman
	Michael Smith



At the the Web Ontology Working Group face-to-face meeting on 14 - 15
January 2002 several ways of layering OWL on top of RDF were presented.
These options for layering OWL atop of RDF are reviewed here, with special
emphasis on the consequences of selecting each one alongside a list of
benefits and drawbacks.

The benefits and drawbacks are determined with respect to a series of
parameters that include the following:

* will RDF parsers correctly comprehend every OWL document
* will OWL parsers correctly comprehend every RDF document
* will RDF syntax be interpreted equally by both RDF and OWL
* will every RDF conclusion be a valid OWL conclusion
* will every OWL conclusion be a valid RDF conclusion
* will OWL reasoning be 'smooth' -- i.e. free of contradictions and not
  overly difficult

This document is a bare-bones document, really suitable only for use within
the working group or for interested parties to see what is going on.  It
does not contain any introductory material.  A different document is being
prepared that contains introductory material, etc., etc.  This document
should be ready by mid-February although portions of it are already being
created.

Comments are very welcome, either to the working group or directly back to
us.


The four layering options that were presented at the face-to-face were:

1/ OWL has the same syntax as RDF and extends the semantics of RDF.
2/ OWL has syntactic features that go beyond RDF, and extends the semantics
   of RDF. 
3/ A syntactic embedding of OWL into RDF.
4/ OWL and RDF are different, both syntactically and semantically, but
     have an interesting shared core, both syntactically and semantically,
     where OWL extends RDFS.


1/ Same-Syntax Extension:

Having the same syntax as RDF and extending its semantics is the most
appealing option, at least at first glance, but as pointed out below, this
option leads to semantic paradoxes in OWL.  In this relationship, which is
the same as the one between RDF and RDFS: 
a/ all syntactically valid RDF would also be syntactically valid OWL,
b/ all syntactically valid OWL would also be syntactically valid RDF,
c/ any RDFS conclusion from an input would also be an OWL conclusion from
   that input, and 
d/ an OWL reasoner could draw more conclusions than an RDFS reasoner on an
   input. 

Benefits:

1/ The same parser can be used for both RDF(S) and OWL.
2/ An RDFS reasoner can be considered as a sound but incomplete OWL reasoner.

Drawbacks:

*/ The OWL semantics fails due to the presence of semantic paradoxes.

Discussion:

This option results in the following restriction being in all interpretations.
   _:1 a owl:Restriction .
   _:1 owl:onProperty rdf:type .
   _:1 owl:maxCardinalityQ 0 .
   _:1 owl:hasClassQ _:2 .
   _:2 oneOf _:3 .
   _:3 owl:first _:1 .
   _:3 owl:rest owl:nil .
This restriction has as members all resources that are not one of its
members, i.e., it is the set of all resources that do not belong to it.
Because this restriction is in all interpretations, no consistent extension  
can be given to the rdf:type property in any interpretation, so the
semantics of OWL under this option are ill-formed.


2/ Extend RDFS

Having OWL be an extension of RDFS is another option.  In this
relationship, which is the same as the one between propositional logic and
modal logics:
a/ all syntactically valid RDF would also be syntactically valid OWL,
b/ some syntactically valid OWL would not be syntactically valid RDF,
c/ any RDFS conclusion from an RDF input would also be an OWL conclusion
   from that input, and
d/ an OWL reasoner could draw more conclusions than an RDFS reasoner on an
   RDF input.

Benefits:
2/ An RDFS reasoner can be considered as a sound but incomplete OWL reasoner.

Drawbacks:

1/ New parsers would have to be built for OWL.
3/ Reasoning could be very difficult for OWL.

Discussion:

The syntactic extensions in OWL would be mostly for restrictions.  For
example, a maxCardinalityQ restriction might look something like
   `minCardinality friend 5 Doctor` 
   (resources that have at least five friends who are doctors)
and would be used like
   hypocondriac subClassOf `minCardinality friend 5 Doctor` 
   (hypocondriacs have at least five friends who are doctors).

Reasoning in OWL could be very difficult indeed.  OWL constructs like
transitive properties could end up being conditional, as in 

Jake a [a owl:Restriction; 
        owl:onProperty foo;
	owl:toClass owl:TransitiveProperty;
	owl:hasClass [ owl:oneOf ( bar baz ) ] ] .

One result of this statement is that either bar or baz is transitive.



3/ Embed OWL in RDF

Embedding OWL in RDF, in the sense of using RDF's syntax for OWL, is a
third option.  In this relationship, which is the same as the one between
RDF and XML:
a/ not all syntactically valid RDF would be syntactically valid OWL,
b/ all syntactically valid OWL would also be syntactically valid RDF,
c/ an OWL reasoner need not draw all the conclusions that an RDFS reasoner
   would on the same input, and
d/ an OWL reasoner could draw more conclusions than an RDFS reasoner would
   on the same input.

Benefits:

Drawbacks:

1/ New parsers would have to be built for OWL.
2/ An RDFS reasoner could not be considered as an incomplete OWL reasoner.
4/ Syntactically valid OWL would have a different meaning in RDFS.

Discussion:




4/ Differing from RDF

Making OWL different from RDF, abeit with a common core, results in the
following situation:
a/ not all syntactically valid RDF would be syntactically valid OWL,
b/ not all syntactically valid OWL would be syntactically valid RDF,
c/ any RDFS conclusion from an input that is syntactically valid for both
   RDFS and OWL would also be an OWL conclusion from that input, and
d/ an OWL reasoner could draw more conclusions than an RDFS reasoner on an
   input that is syntactically valid for both.

Benefits:

3/ The OWL reasoner would not have to worry about unusual interactions
    between problematic RDFS constructs and OWL constructs.

Drawbacks:

1/ New parsers would have to be built for OWL.
2/ An RDFS reasoner could not be considered as an incomplete OWL reasoner.

Discussion:

One interesting common core would be triples whose subject and object are
neither classes nor properties, and rdf:type triples whose subject is not a
class.




Summary

The following table summarizes the status of the four options above with
respect to the benefits and drawbacks at the beginning of this note.

Benefit/Drawback			Option
					1	2	3	4

Every RDF doc accepted	 		Yes	Yes	no	no
 	by OWL parser

Every OWL doc accepted			Yes	no	Yes	no
	by RDF parser

RDFS syntax interpreted			Yes	Yes	no	no
	in OWL by RDFS standards

Every RDFS conclusion			Yes	Yes	no	no*
	is also valid OWL conclusion

Every OWL conclusion			no	no	no	no	
	is also valid RDFS conclusion

OWL problems				para-	harder	Yes	Yes
					doxes	inf.
		
			* Yes for common syntax

Received on Thursday, 31 January 2002 08:45:51 UTC