W3C home > Mailing lists > Public > public-owl-dev@w3.org > July to September 2012

Re: actual effects of stepping outside OWL DL?

From: Bijan Parsia <bparsia@cs.man.ac.uk>
Date: Fri, 21 Sep 2012 22:19:54 +0100
Cc: public-owl-dev@w3.org
Message-Id: <4B53819B-F2A4-4E3F-A4B5-8209BC6862CC@cs.man.ac.uk>
To: "Dan B." <danb@kempt.net>
On 21 Sep 2012, at 21:06, Dan B. wrote:

> What actually happens when you step outside OWL DL (in the context of
> using a DL reasoner on an OWL ontology)?

Depends on the reasoner and the mode it is engaged in. The strictest mode is to reject such ontologies.  In some modes, a reasoner will ignore some subset of the ontology and work with the DL subset. However, this is usually a nondeterministic process as there are typically several possible DL subsets of the super-DL ontology.

> Does doing so just mean that a DL reasoner can't make use of information
> in non-OWL-DL constructs?  (Can it still reason over the OWL-DL part of
> the ontology, or does DL reasoning break down?)

See above.

> Does doing so just mean that a DL reasoner might exceed the algorithmic
> complexity guarantees it would meet for DL constructs (depending on your
> particular ontology, non-DL features, and ontology and dataset size), or
> does it mean that it's very likely to exceed them in almost any case?

Well, if you throw in something they don't how to handle, then they just sorta don't work for them.

> Does doing so have other implications?
> (Where I'm coming from is wondering what the cost is of having non-DL
> portions of an ontology.  I'm thinking of the animal-book-subjects modeling
> example where one option is to use classes (for animals) as objects of a
> hasSubject property of books.

This probably fits into OWL 2 DL as a kind of punning.

> If a bigger ontology had a non-DL portion, would the DL portion still be
> processable directly, or would one have to strip out all the non-DL
> axioms/assertions before passing the ontology to a DL reasoner?)

That's the safer way to go since then you know which axioms are removed.

But for your particular example, punning should do the job.

Received on Friday, 21 September 2012 21:20:19 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:58:21 UTC