RE: inference

> From: Bernardo.Cuenca@uv.es [mailto:Bernardo.Cuenca@uv.es] 
> But as far as i know from the Formal 
> Semantics specification of RDF, RDF doesn't define any "inference or 
> reasoning engine". RDF defines "entailment" which appears to be a way 
> of RDF-graph matching, but doesn't provide any actual inference 
> capability.

Implementations of RDF must pass the various entailment tests, i.e.
given the presence of some triples, they must infer others.  So the
specification says what must happen without constraining the
implementors in the ways in which they can do that.

>  2) I read in a paper about DAML+OIL that OIL is based on a reasoning 
> engine called FAcT, based on descriptive logic. Nevertheless 
> i haven't 
> read a word about any reasoning engine for inference in the OWL 
> specifications.

Depends which OWL.

OWL-Lite has an effective decision procedure; Racer
(http://www.fh-wedel.de/~mo/racer/) is the engine that comes closest to
implementing Lite, although FaCT
(http://www.cs.man.ac.uk/~horrocks/FaCT) and Cerebra
(http://www.networkinference.com/) have partial implementations.  I
believe work is in progress on all of these to increase their
compatibility with OWL-Lite.

OWL-DL has no known *effective* decision procedure, as many of the
optimisations that are applicable in Lite can't be used in DL.

OWL-Full has no complete decision procedure, as it is undecidable.

> The specifications appear to be rather oriented to 
> teach how to write ontologies.

The Guide is.  Take a look at Semantics and Abstract Syntax if you want
a more formal description.

> Moreover, OWL makes the open world 
> assumption and this may yield to indecidability...

Only in OWL-Full.  Lite and DL are known to be decidable.

> 3) As far as i know there is no reference to a reasoning 
> engine in RDF-Schema

Can't help you on that one - I'm a DL-head :-).

		- Peter

Received on Sunday, 23 February 2003 06:38:20 UTC