RE: DDTF/layering: weak class theory seems good enough (5.3, 5.10)

At 8:29 AM -0400 5/29/02, Peter F. Patel-Schneider wrote:
>From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
>Subject: RE: DDTF/layering: weak class theory seems good enough (5.3, 5.10)
>Date: Wed, 29 May 2002 13:06:07 +0100
>
>[snip]
>  > In OWL1 we can choose to:
>>  A: not get all the set theoretic entailments we would like, but get the
>>  layering right
>>  B: get the set theoretic entailments we would like, but screw the
>>  layering
>>  C: take enough time to make a better job on the layering and get all the
>>  set theoretic entailments
>>  D: take even longer and get both the layering and the set theoretic
>>  entailments right.
>>
>>  I think both the last two (C and D) can be postponed until OWL2 (after
>>  this working group).
>>
>>  I personally prefer prioritizing the layering (A), since set theory is
>>  easier.
>>
>>  Jeremy
>
>The question is what is the correct layering.  If you mean a same-syntax
>extension, then, again, that is impossible for the uppper layers without
>going to a very non-standard logic.  If, however, you allow syntax
>extensions, and I see absolutely no reason not to allow syntax extensions,
>then the situation changes completely.  (Of course, you are still left with
>dealing with the odd parts of RDF.)
>
>peter

Peter - I must admit that I am confused, and I'd like to be straight 
on this.  As I understand the world, the current DAML+OIL semantics 
is an option, but only if we are willing to live without entailments 
that many of us think should be there.  I base this on the wording in 
Issue 5.10 which you authored, to wit:

>DAML+OIL semantics (both the model theory and the axiomatization) 
>are too weak. For example, it does not allow the inference of 
>membership in any restrictions that are not present in the knowledge 
>base, even though many of these are desirable consequences. For 
>example, if John is an instance of both Person and Employee, 
>DAML+OIL does not sanction the conclusion that John is an instance 
>of an intersection of Person and Employee.

I took this to mean that one option we have is to live with these 
weak semantics by  saying we give up on desirable consequences (that 
is - if we rule this issue as POSTPONED we would tacitly be accepting 
this solution).

This is also how I read your earlier document, the layering proposal, 
and is what you presented at the first f2f -- this is all summed up 
in note you produced at the last f2f which stated:

>  >If the WebOnt language is to
>>  1/ use RDF syntax, and
>>  2/ be an extension of RDF, and
>>  3/ to be able to entail appropriate class memberships,
>>then the WebOnt language needs to have a facility for having RDF triples
>>that do not denote in the RDF model theory (a.k.a. dark triples) and URI
>>references that do not denote in the RDF model theory (e.g., URI references
>>that only exist in dark triples).
>

which I take to mean that we could have 1 and/or 2, but not 3.

In the message above, you claim we cannot do it by accepting these 
weak semantics, which seems to contradict my previous understanding.

PLEASE NOTE - I AM NOT EXPRESSING AN OPINION ON THIS - I just want to 
make sure that I (and the WG) have a handle on this --

  -JH


-- 
Professor James Hendler				  hendler@cs.umd.edu
Director, Semantic Web and Agent Technologies	  301-405-2696
Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
Univ of Maryland, College Park, MD 20742	  240-731-3822 (Cell)
http://www.cs.umd.edu/users/hendler

Received on Wednesday, 29 May 2002 12:30:35 UTC