Re: DTTF: another summary

From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Subject: DTTF: another summary
Date: Fri, 3 May 2002 10:45:34 +0100

> I believe we have three live proposals:
> 
> - comprehensive entailments: Jeremy
>   [1]
> - dark owl: Peter
>   [2], [3]
> - dark lists: Pat
>   [4]
> 
> We are agreed that the first does not use dark triples
> and that the second and third do.
> 
> Pat and Peter have asserted that comprehensive entailments
> involves significantly more risk and research than the
> other two proposals. I have not conceded this.
> 
> I believe that DanC has understood my proposal.
>
> There has been no evidence that anyone has understood either
> of the other proposals. I have claimed to have understood
> both, but seem to have been mistaken vis-a-vis Pat's.
> I have also proved unable to articulate Peter's.

Well, in general, your proposal boils down to having comprehension axioms
for only the right classes.  This much I understand.  I even understand
which classes you view as the right classes.  However, your comprehension
axioms produce an ill-defined class from the empty knowledge base.
Further, I disagree with your definition of the right classes, and I do not
know if there is any way of coming up with a set of acceptable classes that
do not include any of the problematic ones.

> In particular there does not appear
> to be any evidence that Pat and Peter agree on anything other
> than that the right proposal involves dark triples. Since the
> "right proposal" is an unbound variable in that agreement it
> seems to me to be high risk to build on that agreement.

I don't think that darkening only lists is sufficient.  However, darkening
all owl constructs is definitely sufficient, and is known to work.  (It
results in something very close to a standard description logic.)

Think of the analogous situation with respect to comprehension.  No
comprehension (similar to all dark) is known not to be appropriate.
Complete comprehension (similar to no dark) is known to produce paradoxes.
This is much worse than the case with dark triples.

[...]

> On the relationship between the proposals and the problems.
> 
> Comprehension + Paradox
> =======================
> 
> The comprehension rules are shared between the
> comprehensive entailments and the dark lists proposal [6].
> 
> These rules are the way both these proposals
> address the comprehension issue without falling into
> the paradox problem.
> 
> In particular, my attempt to articulate to Pat how
> I understood Peter's proposal as addressing Peter's
> paradox [7] fizzled out.
> 
> I don't feel we have seen from Peter an account of how
> either issue is addressed.

In a standard description logic there is no need for any comprehension
principles, because descriptions are not elements of the domain (i.e., they
are dark).  In a standard description logic there are also no paradoxes,
again because descriptions are not elements of the domain.  

> Layering
> ========
> 
> Peter and Mike see the impossibility of layering as
> self-evident.
> 
> I have disputed this, seeing an ill-formed list as a semantic
> rather than a syntactic problem.

The problem is not just with ill-formed lists.  Consider the log: extension
of n3.  The paradoxes here do not involve lists, they involve
self-reference and falsehood.

> I believe that a clean monotonic layering of OWL on top of RDFS
> is highly desirable, and that we need to have good evidence of
> its impossibility before not attempting it.
> 
> A monotonic layering would be one in which
> 
> if
> 
>   a rdfs-entails b
> 
> then
> 
>   a owl-entails b
> 
> and both rdfs-entails and owl-entails are monotonic.

Well there are many layerings that satisfy this requirement.  The problems
occur when extra conditions are added, such as all owl KBs must also be
rdfs KBs.  Considering semantic layering without specifying the other
conditions is not useful.

[...]

> I wonder whether Pat and Peter see strict layering as
> advocating that the object denoted by a URI in the RDFS layer
> is the same as that in the OWL layer. I see that as an
> irrelevance. Layering expressed using Mike semantic property
> 1 does not reuse an RDFS interpretation as an OWL interpretation.

Identity is not really the issue here.  The issue is equivalence.  (In
other words, if it looks like a duck, walks like a duck, and quacks like a
duck, it may as well be a duck.)  If an OWL interpretation has to work like
an RDFS interpretation then it may as well be an extension of an RDFS
interpretation.   (Again, this is modulo syntactic conditions, as allowing
syntax extensions makes life much easier.)

> *****************************
> 
> An important aspect of the difference between the
> proposals is syntax vs semantics.
> 
> The dark stuff is seen as syntactic, the not dark stuff
> is seen as semantic. Hence whenever we are talking about X
> there is often miscommunication because the advocates of
> dark X think we are discussing the syntax and its semantics,
> whereas the opponenets assume we are talking at only a semantic
> level. Quite who is on which side varies with X.

Well, my view is that non-dark stuff is the stuff that has to act exactly
like RDF(S) stuff, in all its glory.  Dark stuff is free to diverge from
the very restrictive RDF(S) vision of how the world works.

Received on Friday, 3 May 2002 10:33:58 UTC