Re: REQDOC: reification

From: Pat Hayes <phayes@ai.uwf.edu>
Subject: Re: REQDOC: reification
Date: Fri, 15 Feb 2002 01:11:08 -0500

> >One of the requirements in the requirements document is the ability to
> >associate properties with statements.  This would require, at least,
> >some version of reification of statements, and to work right would require
> >a way of referring to statings.
> >
> >RDF has made a total hash of reification.
> 
> I don't agree. RDF certainly screwed up in some ways. It is 
> incomplete, and the M&S is confused and ambivalent between divergent 
> readings, but the general idea is sound, if rather simple, and the 
> use cases that we have been able to discover all pretty much converge 
> on one of the two plausible readings of the M&S wording, so the WG 
> will probably give a reasonably clear ruling on this soon. So its not 
> a total hash. It is pretty useless, in my view, but some people can 
> use it coherently and seem happy with it.

Well, there is a document (RDF M&S) that says one thing (every statement
has exactly one reification), is often read as another (there can be
any number of reifications for a statement), and is used in a third way (a
reified statement corresponds to a stating).  What else can this be but a
hash?

> >The RDF Core WG is trying to fix
> >this a bit, but it is probably out of their scope to make any significant
> >improvment (as opposed to significant fix).
> >
> >I think that adding reification to OWL would be, at best, a significant
> >research project and, at worst, a black hole.
> 
> I think that what might be called simple object reification - the 
> ability to describe syntactic object tokens, including those of the 
> language itself, and some kind of external referential linking 
> mechanism analogous to ostensive pointing to an expression token - is 
> quite do-able, raises no deep black-hole issues (notice I carefully 
> did not say 'tr*th pre*d*c*te' ) and provides about 90% of the 
> practical functionality needed by enthusiasts of reification. So I 
> think that something useful can be done pretty easily. It will be 
> moderately trivial from a FOM perspective, for which we should all 
> breath sighs of relief.

Well, yes, if all you want is some syntax (rdf:Statement, rdf:subject,
rdf:predicate, rdf:object) that has no other meaning, then go ahead.  I
would say this gets about 0% of the practical functionality of reification
with about 0% of the effort.  If users of RDF want to use the non-existent
functionality of this syntax to hang any outside-of-RDF reification onto
then they can.   But why then make all this part of RDF?  You aren't
getting anything for it from RDF (except, maybe, a gentleman's agreement to
use the same URIs).

> >In any case, I don't see the motivation for the ability to associate
> >properties with statements (in the OWL logic, at least) from the shared
> >ontologies goal.  Surely it is possible to share ontologies without
> >associating properties with statements.
> 
> The real use case is associating them with statings, ie *tokens* of 
> expressions in ontology documents. And that is needed in for example 
> date-stamping, tracking provenances, that kind of thing.

Yes, sure, but how does RDF reification help in this?  Futher, this is not
a ``sharing ontologies'' motivation any longer.

> Pat Hayes
[...]

Peter F. Patel-Schneider
Bell Labs Research

Received on Friday, 15 February 2002 14:55:47 UTC