W3C home > Mailing lists > Public > public-owl-wg@w3.org > October 2007

Re: comments on RDF mapping

From: Boris Motik <boris.motik@comlab.ox.ac.uk>
Date: Wed, 24 Oct 2007 16:26:14 +0100
To: <public-owl-wg@w3.org>
Message-ID: <000201c81652$372fada0$2711a8c0@wolf>

Hello,

I would just like to briefly supplement Ian's explanation of this issue. Having the fully typed syntax is not only an aid to the
parser; rather, it is sometimes strictly necessary in order to be able to transform a set of RDF triples into an ontology in the
structural specification. I'll try to explain this on an example.

Imagine that your ontology O contains all the following triples:

(1)  <X rdf:type owl:ObjectProperty>
(2)  <X rdf:type owl:DataProperty>

(3)  <Y rdf:type owl:ObjectProperty>
(4)  <Y rdf:type owl:DataProperty>

(5)  <X rdfs:subPropertyOf Y>

Note that triples (1) and (2), and (3) and (4) are allowed because you can have punning in OWL 1.1; hence, you can use the same name
as both object and a data property.

If you now try to produce an axiom that corresponds to triple (5), you have a problem: is this axiom representing inclusions between
the object property X and the object property Y, or between the data property X and the data property Y?

The problem arises because you've "overtyped" both X and Y. As similar problem, however, can occur if you "undertype" X and Y.
Consider an ontology O' containing only the triple (5). The translation of O' into the structural format is now not possible because
you don't know how to interpret X and Y.

Similar problems arise in the restrictions: when you see, say, an existential restriction, you don't know whether you should
transform it into a restriction on the object or the data property.



These problems become exacerbated if you use OWL imports. Assume for the moment that the ontology O contains only triple (5), and
that it imports the ontology O'' that contains triples (1) and (3). In this case, everything is fine; however, to be able to
transform (5) into an axiom, you can't look just at the triples in O, but you need to look at all the imported triples. But what if
O'' is in some other format (i.e., not in OWL RDF)? Implementing such a parser is a nightmare.


All of these problems become simpler if you can infer the type of objects from each triple. Hence, we modified the translation to
RDF to fix this.


I'd just like to finish this e-mail with an observation that I've seen quite a few ontologies that were broken due to the problems
outlined in this e-mail. Hence, addressing these issues should be a good idea.

Sincerely yours,

	Boris


> On 23 Oct 2007, at 11:49, Ian Horrocks wrote:
> 
> The motivation for properties such as subObjectPropertyOf is not  
> related to punning. The idea is to facilitate parsing/species  
> validation by enforcing strong typing. For example, it is illegal in  
> OWL DL for a datatype property to be a subProperty of an object  
> property, but when parsing a <P subPropertyOf S> triple, the types of  
> P and S may not be known. As a result, not only will a decision on  
> the species of the ontology need to be postponed, but it becomes  
> dependent on a (relatively) complex and non-local condition, i.e.,  
> that P and S are either both datatype properties or both object  
> properties.
>
> Ian
>
>
> On 23 Oct 2007, at 11:22, Jeremy Carroll wrote:
>
> >
> >
> > I think Jim's example about subObjectPropertyOf is compelling.
> >
> > In essence, I don't think we can publish a meaningful and helpful  
> > RDF Mapping document until we have decided whether or not we accept  
> > the 'punning' design in the member submission.
> >
> > I think this is one of the features of OWL 1.1 that causes the  
> > greatest unease with the HP developers. As I understand the design,  
> > language terms like subObjectPropertyOf are largely motivated by  
> > the punning design.
> >
> > A further possible motivation is that in OWL 1.0, at I think mainly  
> > my request, one design choice is that the triples version of OWL DL  
> > is strongly typed, in the sense that (nearly) every URI and blank  
> > node is required to have an rdf:type triple. Many of the required  
> > type declarations are unnecessary, and it may be a better design to  
> > allow unnecessary ones to be omitted. However, I think that the  
> > explosion of terms in the member of submission is unfortuante, and  
> > should be avoided.
> >
> > Jeremy
Received on Wednesday, 24 October 2007 15:27:07 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:13:26 GMT