ISSUE 5.3 some of the HP discussions

I drafted this before the weekend traffic concerning changing the defn of 
owl:Thing and owl:Class in section 5.3.
These discussions preceded the reopening of the issue, and resulted in us 
deciding not to ask for the issue to be reopened.



Summary: HP is not intending to oppose if-then layering between OWL DL and
OWL Full, as reported by Peter; but would prefer a more thorough fix,
retaining the iff in the consensus document. We are unwilling to do much
work towards such a fix, nor would we want to see the WG delayed by months
in such a task.


1) The counterexamples were surprising, and if no substantive changes are
made, some changes will be needed so that the reader of the OWL documents
will be less surprised.


2) Peter says:
 >However, this does not violate any of the claims in S&AS, as the relevant
 >theorem is that if an entailment holds in OWL DL then it holds in OWL Full,
 > not vice versa.


Seems reasonable, weaker than we realized, but reasonable.

3) It violates the concensus statement:

      "KB large-OWL-entails C  iff  KB Fast-OWL-entails C"

but its not so clear that it fails to satisfy the original issue.
The original issue required layering on RDF/RDFS which OWL Full
achieves. OWL/DL is, in any case, not part of that original requirement
through not being an RDF/RDFS supserset.
Having entailments over OWL/DL be only a proper subset of the OWL Full
entailments doesn't feel like it makes matters that much worse.

4) Some editorial changes would be needed to S&AS is modified
to make this clear.

5) We would prefer a rabbit to be pulled out of a hat and the
claim from the consensus statement to be credible.
However, HP is not expecting to make a contribution to such
rabbit-pulling, nor are we expecting other WG members to do
such work. If it proves easier than we expect then we may change
our position.

6) We are concerned that OWL systems and APIs to OWL systems
will necessarily be more complicated, since reasoners will
need to declare themselves as subscribing to the OWL DL or
OWL Full semantics.

7) Point (6) is not a complete disaster since the APIs will
inevitably be quite complicated, given that different OWL Full
reasoners will have different capabilities

Thanks particularly to Dave Reynolds for his inputs.

Jeremy

Received on Tuesday, 1 July 2003 14:24:30 UTC