Re: comparing DAML-ONT and OIL (was Re: semantics of daml)

From: Dan Connolly <connolly@w3.org>
Subject: Re: comparing DAML-ONT and OIL (was Re: semantics of daml)
Date: Sat, 14 Oct 2000 22:59:58 -0500

> "Peter F. Patel-Schneider" wrote:
> >
> > Peter Patel-Schneider
> > 
> >                                 OIL-Standard            DAML 1.2
> > 
> > Semantics                       denotational            English
> >                                 complete, unambiguous   partial, ambiguous
> 
> I expect that eventually we can change that DAML entry
> here to "provided by OIL", but I haven't finished
> studying the OIL denotational semantics.
>
> TODO: see if OIL denotational semantics agree with
> my understanding of DAML.

OK, this would be a good thing, but then why not have presented DAML with
such a semantics from the start?  

> >   Meaning of a collection       always conjunctive      usually conjuctive
> >   of statements                                         sometimes disjunctive
> >                                                                 (domain from RDF)
> 
> I consider RDF to be always conjunctive; the suggestion
> (in the RDF spec) that domain is disjunctive is
> disputed; one of the RDF schema spec editors
> agrees that it's bogus:

If DAML-ONT is going to use stuff from RDF that is suspect in some way,
then there should be comments in the DAML-ONT document to the effect that
either
1/ DAML-ONT will not use RDF in a way that will trigger the suspect
behaviour
or
2/ DAML-ONT depends on a change to RDF
By the way, the most-recent RDF specification has no hint that domain will
be changed.

> >   Equivalence                   yes (defined classes)   maybe
> 
> I hope you consider this one "yes" by now.

No I certainly do not.  There have been comments that eequivalentTo should
be read as an aliasing mechanism. 

> >   Individuals                   yes                     yes
> 
> I'm not sure I understand what Individuals are; cf
> "universe of discourse" above.

Individuals are, well, individuals.  Things like Joe and Mary and a rock.
Some systems only allow classes.  There are lots of related issues, such as
whether a class is also an individual, and whether there are things like
metaclasses.  

> > Reasoning
> > 
> >   Specification                 complete                incomplete
> > 
> >   Completion Possible           no                      no
> >   Least Partial Model           no                      no
> > 
> >   Difficulty                    EXPtime complete (?)    unknown, at least NP hard
> 
> How do you conclude that DAML reasoning is at least NP hard
> without a specification for it?

OK. OK.  There might be a reading for the DAML-ONT constructors that makes
reasoning easy.  However, the only readings that I could construct do make
some reasoning hard in DAML-ONT.  I will produce (shortly) a longer message
making this point slightly more formally.

> Could you point me toward a specification for Least Partial
> Model? I learned about models a few months ago, and I
> was able to record my understanding in larch[fs]
> so I'm fairly confident I understand it, but
> I didn't run across the term "Least Partial Model".

Least Partial Model may not be a common term, but it does get used in some
places.  Perhaps the easiest way to show what I mean is to look at a
database, but without the closed world assumption.  Then there are many
possible models for a database, but there is a canonical, or least,
partial model, namely the positive portion of the normal (closed world)
model for the database.  On the other hand, most disjunctions don't have
this sort of model.  For example ``it is Tuesday or it is raining'' has two
possible small models, one for each of the disjuncts and no least partial
model.  The best work on this that I have seen was some stuff that Hector
Levesque did about 10-15 years ago.

Peter Patel-Schneider