Re: yet another non-entailment

From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Subject: Re: yet another non-entailment
Date: Tue, 27 Aug 2002 13:50:33 +0200

> 
> 
> Peter:
> > This message brings up an interesting point about entailments.
> 
> > Why should I, or Jeremy, or the working group, approve of or disapprove of
> > a particular entailment?
> 
> > My stance is clear.  I approve of exactly the entailments that follow from
> > the model theory.
> 
> My view, which I think is largely in line with this, is that the entailment
> tests and the model theory are two sides of the same coin.
> 
> Clearly if we approve tests which do not follow from the model theory then
> before we are finished we either need to revisit the tests or to revisit the
> model theory.

Agreed.

> I feel that the tests provide clarity by making it clear to the non-expert
> some of the consequences of the model theory. In RDF Core, where in truth,
> only Pat is expert in model theory, this has been a very helpful feature of
> entailment tests.

Agreed, except that there are more experts in the RDF model theory outside
of the RDF Core WG.

> Thus if an entailment followed from the model theory that the WG did not
> feel confortable with then this was an indication to Pat to go back and
> tweak the model theory; or alternatively to argue why some or other feature
> of the model theory was sufficiently desirable to merit a test that did not
> accord with the groups intuitions.

Agreed.

> Thus, at the moment there seems to be some discussion about the merits or
> not of Pat's articulation of the model theory versus Peter's. Both Peter and
> I have expressed this discussion by providing concrete examples of
> entailments that hold in Peter's world and don't hold in Pat's world. I
> think this is a helpful concrete way of articulating the differences. These
> are likely to be more accessible than discussion about domain of discourse
> and non well found set theory.

I believe that Pat's current changes (which I have not seen yet) will have
the effect of making Pat's model theory be an embedding of my model theory
in the RDFS model theory.  It should then be the case that the two model
theories will have the same consequences when restricted to OWL knowledge
bases.   To this end, the points that Ian and I made in the teleconference
last week were pointing out bugs in Pat's model theory.

> When we make decisions about the tests there is a need to make them
> coherently. The peicemeal nature of test case approval will result in some
> mistakes that will need to be revisited. But this seems to be adequately
> covered by the process. In the final stages of our work we can expect
> implementations that fully implement our specs including the model theory.
> If they do not pass all the tests then the tests themselves are faulty and
> will need to be corrected.

I am much less optimistic about this point.  I feel that it is necessary to
subject tests to close scrutiny before approving them, as it is all too
easy to approve a nonsensical test, and all too hard to remove one.

> > Jeremy's stance below appears to be that he will disapprove of entailments
> > on their own.
> 
> I expect the group has enough expertise to either avoid the pitfalls, or to
> climb out of any holes that we end up in. I see the tests as guiding the
> model theory just as much as the model theory guiding the tests. Both will
> need to be corrected.
>
> Jeremy

peter

Received on Tuesday, 27 August 2002 08:10:19 UTC