ontolog-forum
[Top] [All Lists]

Re: [ontolog-forum] misunderstandings about OntoClean

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Chris Welty <cawelty@xxxxxxxxx>
Date: Fri, 14 Sep 2007 09:09:49 -0400
Message-id: <46EA881D.60906@xxxxxxxxx>

Hi Claudio,    (01)

There is a wiki page on ontoclean at ontoworld.org. Probably I should put a 
copy 
on ontolog's wiki as well. I only maintain it in brief spurts, however.    (02)

Claudio Cardone wrote:
> Studying the OntoClean methodology, I think I have found some matter of 
> misunderstanding. I try to explain my point of view,  please give me 
> some feedbacks, because I'm a novice here:
> 
> In the OntoClean papers (e.g. see 
> 
>http://ontolog.cim3.net/file/resource/presentation/OntoClean--ChrisWelty_20041118/guarinowelty_final_v4.pdf
> 
> 
><http://ontolog.cim3.net/file/resource/presentation/OntoClean--ChrisWelty_20041118/guarinowelty_final_v4.pdf>)
> 
> rigid, non rigid and anti-rigid metaproperties are assigned to each 
> class (property) of a given taxonomy, isolating it from the others and 
> studying its meaning.
> It seems to me that I can define a non rigid or an anti-rigid class only 
> if I have a rigid class of reference that has some identity criteria.    (03)

Its a subtle issue, so excuse the convoluted reply.  Rigidity as defined has 
little to do with identity, but the rigidity judgement does need to be made 
with 
some notion of identity in mind.    (04)

You have to ask yourself whether an instance of the class in question can 
possibly change its membership; are there circumstances in which an entity 
exhibits the property and circumstances in which *the same* entity does not.  
As 
I believe you are suggesting below, in order to answer this often you must 
determine whether the property contributes to identity or not.    (05)

We studied this in one of the first OntoClean papers ("A Formal Ontology of 
Properties"), and we determined that a property with +O and -R makes no sense, 
which again points to the relationship between rigidity and identity, but 
actually stemmed from the definition of +O, which is basically a rigid form of 
identity.    (06)

Later we relaxed this a bit as we tried to account for different kinds of 
identity, in particular something we tried to call "local identity".  There are 
identity criteria (IC) in some systems that admit to being local to some 
particular not-necessarily-rigid class.  For example, a "student id" may be 
used 
as an IC for student in a system that allows a student to become an alumnus, 
and 
the "student id" may not be an IC for the alumnus class.  The point is that it 
is possible for a class to introduce its own sort of IC (which is what +O is 
supposed to mean) without being rigid.    (07)

The point of the +I and +O properties are that every *instance* must have 
membership in a class that is +O, not that every property subclass one.  There 
are properties (especially roles) that can hold of many different types of 
entities.    (08)

> In 
> fact, i can't say that "student" is an anti-rigid property if i don't 
> have in my mind the concept of human as a rigid class with some identity 
> criteria. To say that some instance of "student" can change and become 
> something else without disappearing i need to know that the instance 
> holds its identity because it belongs to another rigid property (that 
> necessarily subsumes "student"). So it can be misunderstanding to assign 
> the non-rigidity or anti-rigidity to one property if first we haven't 
> found a +R and +O (or +I) class that includes all the instances of that 
> property.     (09)

In a sense yes, but be careful.  The papers outline a clear methodology to 
establish the "backbone" (the taxonomy of +R+I classes), but you can be too 
inflexible about this.  In the examples, the "Red" class is -I-R, really we 
probably meant that property in the sense that only material objects can have a 
color, so one may argue that Red should be a subclass of something with 
identity.  But we did that intentionally, to make sure that its clear that the 
existence of -I classes does not mean there are -I instances!  Every instance 
must instantiate a +O class, the -I on a property just allows a certain 
sloppiness about what classes of entities the property can be applied to.    (010)

So, in that example, *anything* can be an instance of Red, but all instances of 
Red are also instances of some other class with a rigid IC; this is what the 
-I-R classes are - properties of anything. We called them attributions.    (011)

Some people might argue against attributions in ontologies.  From the 
perspective of metaphysics it is probably a fair point, as metaphysics is 
supposed to account for *everything*.  But from the perspective of systems 
design, where we are not supposed to account for everything, it does introduce 
problems for reusability.  Imagine this case: you have an ontology for domain 
X, 
and the property Red in it, and you pretty much want to say anything can be 
red, 
but you insist there be no attributions, so you make the class Red a subclass 
of 
some arbitrary top class, call it X-thing.  Then someone else decides they want 
to reuse your ontology in domain Y, for which they have a top class Y-thing, 
and 
they make X-thing and Y-thing subclasses of a new class XY-thing.  In this new 
merged ontology, none of the Y-things can be red, unless they do some 
complicated mapping of Y-things into X-things.    (012)

This happens *all the time* on the semantic web and, again, stems from the fact 
that we rarely, in designing ontologies for IT systems, consider ourselves to 
be 
building a theory of the whole universe.    (013)

> Instead, it seems that I can always say if a class is rigid, 
> because I "state" its rigidity in my given world, choosing a point of 
> view, except if I have already chosen another point of view: for 
> example, I can state that a student is a rigid class, except if I had 
> already defined "human" as a non rigid class, according to the 
> ontoclean's rules.    (014)

Yes, of course - OntoClean is not an ontology, it helps you to communicate the 
point of view your ontology takes. *This* is the most common misunderstanding 
of 
OntoClean.  There is no class that OntoClean dictates is rigid, or anti-rigid, 
etc.  The decisions you make wrt these metaproperties communicate more 
accurately what your point of view, or rather the point of view of your system, 
is.  There is a little more of this discussion on rigidity in the Welty & 
Andersen paper, published in JAO.    (015)

> In the OntoClean papers the above mentioned difference between rigidity 
> and non/anti rigidity assignment is not underlined, so may be that this 
> has contributed for some misunderstanding in the use of the methodology.
> What do you think about?    (016)

There are a lot of OntoClean papers......and no real "tutorial".  That 
contributes,too.    (017)

-Chris    (018)


> 
> Thank you
> 
> Claudio
> 
> 
>  
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
> Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
> Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
> Shared Files: http://ontolog.cim3.net/file/
> Community Wiki: http://ontolog.cim3.net/wiki/ 
> To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
>  
>     (019)

-- 
Dr. Christopher A. Welty                    IBM Watson Research Center
+1.914.784.7055                             19 Skyline Dr.
cawelty@xxxxxxxxx                           Hawthorne, NY 10532
http://www.research.ibm.com/people/w/welty    (020)

_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (021)

<Prev in Thread] Current Thread [Next in Thread>