Re: SEM: Classes as instances

From: Jeremy Carroll <jjc@hpl.hp.com>
Subject: SEM: Classes as instances
Date: Fri, 6 Sep 2002 22:35:09 +0200

> I said I would try and share an idea.
> 
> What I think I have heard people say is:
> 
> RDFS semantics work OK because it doesn't have any of the complicated bits 
> like restrictions and cardinalities.
> 
> DL semantics work OK because they have a clear two level system, with classes, 
> properties, instances and data values as all conceptually distinct.
> 
> Put them together and you get a mess.
> 
> As far as I can tell, most of the "classes as instances" use cases are 
> addressed by RDFS.
> 
> Thus, could we conceptually do something like the following:
> 
> 1: apply RDFS completely (with cycles etc)
> 2: then take a DL view of the result
> 
> and not do both together, avoiding some of the pitfalls that Ian worries 
> about.

I don't understand what you are trying to do here.  RDFS is not something
that you apply, so I'm not sure that is going on in this section.  What you
might want to do is to

Given a collection of RDF triples:
1/ form the RDFS closure of these triples
2/ treat the result as a DL KB.

Unfortunately, the RDFS closure of a collection of triples that is not a DL
KB to begin with is unlikely to be a DL KB afterwards.  In fact, the RDFS
closure of any collection of triples is guaranteed to not be a DL KB,
because of rdfs:Class belonging to itself.

> I can see that at some level I am just proposing a scruffy hack, but it seems 
> we have two well defined and well understood processes that we are nervous 
> about combining together in an unconstrained way. So how about combining them 
> in the above constrained way?
> (I think I hear much of the group having digestive problems ....)
> 
> Jeremy


peter

Received on Sunday, 8 September 2002 20:37:29 UTC