W3C home > Mailing lists > Public > www-webont-wg@w3.org > September 2003

RE: Dave's modified tests

From: Ian Horrocks <horrocks@cs.man.ac.uk>
Date: Sat, 20 Sep 2003 11:34:58 +0100
Message-ID: <16236.11602.889725.2009@merlin.horrocks.net>
To: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Cc: "Jim Hendler" <hendler@cs.umd.edu>, <www-webont-wg@w3.org>

On September 16, Jeremy Carroll writes:
> 
> > Agreed - no problem regarding automatic (or even systematic manual)
> > rewriting - this could easily be part of an implementation. There was,
> > however, some talk (from Sandro?) of proofs relying on human
> > intervention/guidance - this obviously can't count as a pass.
> >
> 
> I do not believe Dave's intervention is automatic, but I do believe it is
> systematic.
> i.e. if the conclusions of a positive or negative entailment includes blank
> nodes of (possibly inferrable class) type owl:Class or owl:Restriction, and
> such nodes root subgraphs that (other than missing type expressions) conform
> with the OWL DL syntax for *description* then, the tree rooted in such nodes
> can be inferred by the comprehension axioms. Since Dave has decided that he
> does not want to provide such inferences with his reasoner's native
> capability (which has a very general "what do you know?" interface) he adds
> these class expressions to the premises - which is sound because the class
> expressions are inferrable from the premises.
> (Hmm the human verification of that soundness is not completely trivial -
> the components used in the class expression must be inferrable).

I am not trying to be difficult, but if I understand correctly, the
systematic intervention amounts to something like: "at this point I
add information that the algorithm would have inferred if I had
implemented it and if it is correct". I don't find this tremendously
convincing.

Ian

> 
> I'm thinking of a reply along the lines of:
> 
> 
> [[
> Thanks for the offer of tests which do not exercise the comprehension
> axioms.
> There is not consensus in the group to include easier versions of tests that
> we already have; and stronger feeling against making any of our current
> tests easier
> in this way.
> 
> As you will have noticed that are some features of OWL which are
> inadequately tested,
> if you develop any tests for such features please let us know.
> 
> As far as reporting your test results go, some others implementors have used
> systematic
> manual intervention at some points in the test cycle.
> While our main report on the tests passed by various implementations will be
> restricted
> to entirely automated test runs, we will have a subsidiary report indicating
> the passing of tests which have been manually reformatted in a systematic
> way.
> Hence we encourage you to submit two tests reports, ones for the tests that
> you run out-of-the-box, and a second for tests which you have had to
> manually reformat.
> 
> ]]
> 
> I think that captures where we have got to ... although I am not sure
> whether Sandro is intending to report manually reformatted test passes at
> all.
> 
> Jeremy
> 
> 
Received on Saturday, 20 September 2003 06:37:03 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:58:02 GMT