OWL 2 RL Implementation report

** OWL 2 Implementation report

Dave Reynolds (HP Laboratories)
Chris Dollin (HP Laboratories)
Stuart Taylor (University of Aberdeen)
Jeff Pan (University of Aberdeen)

* System

Jena

http://jena.sourceforge.net/
http://jena.sourceforge.net/inference/OWL2RLExpt.html

This is an experimental implementation of OWL 2 RL based on 
instantiating a set of forward chaining JenaRules from a premise ontology.

* Profile

OWL 2 RL only.

We selected this as suited to a rule based implementation, similar in 
style to our existing implementations of fragments of OWL 1.

* Semantics

RDF semantics only.

Jena is RDF based and many of its users work at the RDF level with some 
use of RDFS, OWL and custom inference. The RDF semantics is the better 
match to Jena.

* Conformance

We believe the implementation to be conformant to the RL profile of the 
OWL 2 CR document, for a restricted set of datatypes. This 
implementation provides inference services, but at this time does not 
provide any convenience API for OWL 2 constructs beyond existing OWL 1.

We have tested on the RL Profile test cases (no fails) and RDF RL test 
cases (all pass).

We have been conservative and for Consistency tests we return "unknown" 
  (and thus report IncompleteRun) if no inconsistency is found. 
Similarly we report "unknown" for NegativeEntailment tests outside the 
bounds of Theorem PR1 when the entailment is not found.

We were surprised to find that so many of the RL Profile Test Cases are 
outside of the scope of the rule set published in the profiles document.

In terms missing features we do not currently support the following 
datatypes:

  rdf:PlainLiteral
  owl:real
  owl:rational
  xsd:dateTimeStamp

We do not currently intend to support rdf:PlainLiteral or owl:rational, 
we do expect to support xsd:dateTimeStamp at some point.

* At risk features

We support rdf:XMLLiteral and believe it should NOT be removed.

We do not support owl:rational and would be happy to see it removed.

* Ready to proceed to Recommendation?

We see no fundamental problems with proceeding to Rec.

We did find the conformance position around OWL RL confusing at first. 
The impression gained from publicity around OWL RL is that it is 
possible to implement by just implementing the published first order 
rule set. Whereas the situation is more complex given that the rule set 
is not complete for OWL RL. At a minimum implementations also have to 
check entailment queries against PR1 and return "unknown" instead of 
true/false for cases out of scope.  However, this confusion may be more 
a reflection of the outreach around OWL RL rather than the specs 
themselves.

Dave
-- 
Hewlett-Packard Limited
Registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England

Received on Wednesday, 5 August 2009 12:46:08 UTC