Re: SEM: comprehensive entailments without dark triples

On Mon, 2002-04-22 at 12:13, Jonathan Borden wrote:
> Dan Connolly wrote:
> 
> > On Mon, 2002-04-22 at 10:41, Peter F. Patel-Schneider wrote:
> > > The problem with these comprehension principles is that they do not
> > > generate non-tree structures.  Therefore classes that are not in the
> form
> > > of trees (yes, this is rather vague, but my previous message gives one
> > > example) will not be consequences.
> >
> > I'm having a hard time seeing this as a problem.
> 
> I have a relatively easy time seeing why this would be a problem. (gasp,
> perhaps this DL stuff is starting to slowly sink in...) This sort of issue
> might arise frequently in genetics (for example) and I was really hoping
> that the language would be able to make these sorts of inferences -- e.g.
> going from facts about individuals (i.e. patients) to classes of diseases.

You can make all sorts of inferences going from facts about individuals
to classes of diseases without cyclical structures like this ever
coming up, no?

i.e. it seems to meet your needs if the following holds:

      John rdf:type Person .
      Bill rdf:type Person .
      John child Bill .
 entailing
      John rdf:type _:1 .
      _:1 rdf:type daml:Restriction .
      _:1 rdf:onProperty child .
      _:1 rdf:hasClass Person .

even though the circular one below doesn't work.

> > This is the example from your message of
> > 22 Apr 2002 11:29:50 -0400; that's the one you refer to, yes?
> >
> >      John rdf:type Person .
> >      Bill rdf:type Person .
> >      John child Bill .
> > entailing
> >      John rdf:type _:1 .
> >      _:1 rdf:type daml:Restriction .
> >      _:1 rdf:onProperty child .
> >      _:1 rdf:hasClass :_1 .
> >
> >
> > I can't understand it well enough to see why I would want
> > it to be the case.
> >
> > By way of trying to relate this to some use case
> > or requirement, here's an attempt at a sort of
> > natural-language translation:
> >
> >   If John has a child, then John is in the class
> >   of things that have children that are in this
> >   class I'm talking about.
> >
> > It just looks like gobbledygook.
> >
> > It seems to me we can meet our requirements and
> > users' expectations without this entialment holding.
> 
> Peter et al. were asked to give a concrete example illustrating the issue.

What issue? Who asked him?

In another thread, I asked him for an example of an actual
semantic layering problem scenario (i.e. something related
to real-world use cases/requirements) where dark triples would
help, along with an explanation of how to use dark triples
to solve the problem.

In this thread, Jeremy has claimed/suggested we can
meet our requirements without dark triples; Peter brought
up his example from that other thread as a counter-argument
to Jeremy.

> He has done that.

He has not, to my satisfaction, connected the theoretic
problem to any real-world needs.

So the cyclic structure above isn't entailed, under
Jeremy's axioms. So what? What do I, as a real-world
user, lose?


> I imagine that this is just one of a likely infinite set of examples that he
> could generate if he has the inclination.

Examples of what?

> It might be a really long process
> going through these, and if the set is really infinite, even after any
> arbitrarily large amount of time and effort, we still won't be sure we've
> excluded any problems.

It might be long, but it shouldn't be infinite: presumably,
each example is, well, exemplary of a whole class of entailments;
and as Peter introduces and explains each one, the rest of us
learn about the space of these entailments.


> > Am I missing something?
> >
> 
> Are you sure you aren't?

No; that's why I asked the question.

-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/

Received on Monday, 22 April 2002 14:32:09 UTC