Re: modelling issue?

Dear Paola:

I think this discussion can be kept pretty short: In OWL (and RDFS), 
domain and range specifications work different from what you would 
expect in a database world.

- In the DB world, modelers are used to employ domain and range 
specifications to constrain the class to which a property/relationship 
type CAN be attached (domain) or what class of individuals can be used 
as a value (range). Any mismatch can be spotted.

- In OWL, they are used to infer additional class membership.

If you say in OWL that the domain of foo:hasEyeColor is foo:bird and 
someone states that

foo:JohnLennon foo:hasEyeColor "Blue"

than a reasoner would infer that JohnLennon is a bird.

So instead of preventing unintended uses of your conceptual element, you 
can make things worse by adding unintended inferences. Instead of 
getting a warning that foo:hasEyeColor should not be attached to objects 
other than birds, you will get a side-effect that obscures the problem.

A reasoner cannot spot that there is a range or domain violation, unless 
you insert disjointness axioms.

If you add disjointness axioms between disjoint classes (e.g. humans and 
birds), then a domain or range violation can be spotted, though 
indirectly, because if "human" and "bird" are said to be mutually 
exclusive class memberships and John Lennon is said to be an instance of 
both classes, there must be a contradiction. But you still don't know 
whether the disjointness axiom or the property usage is incorrect.

For more background information, see

de Bruijn, J., Lara, R., Polleres, A., Fensel, D.: OWL DL vs. OWL 
Flight: Conceptual Modeling and Reasoning for the Semantic Web. 14th 
International World Wide Web Conference (WWW2005), Chiba, Japan (2005) 
623-632

Best
Martin


Paola Di Maio wrote:
> Thanks Dan and Lin
>
>   
>> Domain and range are characteristics of properties (aka
>> relationships, attributes), rather than of the things those properties
>> relate to. And yes, they're not mandatory, you can declare and use a
>> property without saying anything
>> about it's domain or range.
>>
>>     
>
>
> Okay, in another, offlist reply, among other comments I receve the following
> statement
>
> Most books I have read state that you should be cautious about making domain
>   
>> and range statements about properties.  The results, the books claim, are
>> difficult to predict in large ontologies.
>>
>>     
>
> I seem to get two contradictory statements there (and it would not be the
> first time, argh - I get different people telling me different things about
> this), and not sure how to interpret them in relation to each other
> sorry if it's me actually missing out something
>
>
>   
>>  RDF expresses everything in terms of triples. Sometimes the triple is
>> a relationship between a thing and another thing (which is often but
>> not always named with a URI). And sometimes the triple is a
>> relationship between a thing and a string (possibly language-tagged or
>> data-typed). RDF actually calls the link in both cases a "property",
>> but sometimes for communicating in other contexts you'll here
>> "relation" or "relationship" (typically when the link isn't to a
>> literal string value). And sometimes you will hear "attribute",
>> usually when the value is a string literal.
>>
>>     
>
> okay, in the modelling world,  that would be regarded as something to be
> worried about,
> we call these 'conflicts'
> when one thing can be interpreted in more than one way, the system is likely
> to be brittle
>
>
>
>   
>> We had an army of logicians and mathematicians go over the 2004 specs
>> in great detail. They may have missed something, but I suspect our
>> main difficulties are elsewhere...
>>
>>     
>
> well, maybe  a systems engineering perspective would provide additional
> views, or a system of systems engineering (as in network of networks),
>
>
>
>
>   
>>> Are the above points addressed in some RDF tutorial
>>>       
>> Did you see http://www.w3.org/TR/rdf-primer/ ?
>>
>> yes s its a great reference document, but it s very long to read it word by
>>     
> word, and there is no indication that it contains
> the answers I am looking for
>
> Lin,
> Jim H book is something that I want to get my hands on too, surely, still
> questions may have to be answered to issues
> that are not directly or explicitly addressed in the literature, I havent
> been able to set aside time to read it, but I ll try to accsss those chaps
> straight away
>
> P
>
>   

-- 
--------------------------------------------------------------
martin hepp
e-business & web science research group
universitaet der bundeswehr muenchen

e-mail:  mhepp@computer.org
phone:   +49-(0)89-6004-4217
fax:     +49-(0)89-6004-4620
www:     http://www.unibw.de/ebusiness/ (group)
         http://www.heppnetz.de/ (personal)
skype:   mfhepp 
twitter: mfhepp

Check out GoodRelations for E-Commerce on the Web of Linked Data!
=================================================================

Webcast:
http://www.heppnetz.de/projects/goodrelations/webcast/

Recipe for Yahoo SearchMonkey:
http://tr.im/rAbN

Talk at the Semantic Technology Conference 2009: 
"Semantic Web-based E-Commerce: The GoodRelations Ontology"
http://tinyurl.com/semtech-hepp

Talk at 

Overview article on Semantic Universe:
http://tinyurl.com/goodrelations-universe

Project page:
http://purl.org/goodrelations/

Resources for developers:
http://www.ebusiness-unibw.org/wiki/GoodRelations

Tutorial materials:
CEC'09 2009 Tutorial: The Web of Data for E-Commerce: A Hands-on Introduction to the GoodRelations Ontology, RDFa, and Yahoo! SearchMonkey 
http://tr.im/grcec09