Re: TODO++: daml-ont TallThing

On November 3, David Martin writes:
> Brian --
> 
> Ian Horrocks wrote:
> > 
> > On November 1, McBride, Brian writes:
> > > > > There are a couple of minor nits in the example ontology:
> > > > >
> > > > > 1) Father is desribed as having range Man, but presumably
> > > > >    inherits domain Animal from Parent.  Mother is similar.
> > >
> > > A mother of a fox is not human.  I'd expect the domain and range
> > > to match i.e. domain of mother is person if range is woman, or
> > > range of mother is female animal if domain is animal.
> > >
> > > A real nit - sorry.
> > >
> > > Brian
> > 
> > This is another example of a point I made in an earlier discussion on
> > rdf-interest, namely that domain and range restrictions are very
> > strong assertions, and that a value restriction on the domain class is
> > usually more appropriate.
> > 
> > For example, when the domain and range of a property P are restricted
> > to classes D and R respectively, the intended meaning often is (or
> > should be) that IF (i,j) is an instance of P AND i is an instance of D
> > THEN j must be an instance of R. 
> 
> Being new to this list, this is the first I heard of such an
> interpretation of domain and range restrictions, but it strikes me as
> thoroughly counter intuitive.  Aren't the domain and range restrictions
> expressed independently, and thus conceptually separate?  It seems to me
> there is nothing about these restrictions that indicates they are tied
> in the inferential way that you describe.  That's my main concern --
> that a user of DAML-0 would be very unlikely to arrive at your suggested
> interpretation based on intuition.
> 
> Could you please say more about why you think this should be the
> intended meaning?

Of course domain and range are independent. What I mean is that often
when both are stated there is some association between them in the
mind of the modeller. In this case the association should be expressed
by using a value restriction on the concept. Moreover, when a range
restriction is stated there is often a hidden idea of a corresponding
domain in the mind of the modeller - as in the father example above,
where there seems to be an implicit idea that the domain should be
Person. If this is expressed by changing the range restriction to a
value restriction on Person then the problem with "fox" is resolved.

Ian

Received on Friday, 3 November 2000 16:50:06 UTC