Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[protege-owl] Enumerated Classes and Special Relationships

Poovendran Moodley moodleyp at cs.ukzn.ac.za
Fri Jan 22 01:47:45 PST 2010


Dear Dr. Thomas,

On Fri, Jan 22, 2010 at 2:37 AM, Thomas Schneider <schneidt at cs.man.ac.uk>wrote:

> On 21 Jan 2010, at 16:31, Poovendran Moodley wrote:
>
>> On Thu, Jan 21, 2010 at 5:38 PM, Thomas Schneider <schneidt at cs.man.ac.uk>
>>> wrote:
>>> Hi Pooven,
>>>
>>> If you really want to say *all* individuals in A are in relation P with
>>> *all* individuals from B, then it's not enough to say this explicitly only
>>> for all individuals asserted to be instances of A or B. You'd want this for
>>> inferred individuals as well, I suppose.
>>>
>>
>> It's unlikely that an Individual will be inferred to belong to either
>> class A or B. These classes are, for all intents and purposes, primitive
>> classes, it's just that I've made them enumerated classes because I was
>> curious to see how my hierarchy would look; they don't form domains or
>> ranges for any of the object properties and are subclasses of the classes
>> that are specified in the domains and ranges.
>>
>
> But even if they are named, what if you later extend your ontology by
> definitions, say A = Person and owns some Animal, then you declare Bob to be
> of type Person and to own Fluffy, who is of type Bird, which is a subclass
> of Animal. You can then infer Bob to be of type A, so wouldn't you want to
> include all pairs of individuals containing Bob into the extension of
> property P?


Yes, you bring up an important point... we should always model our knowledge
in such a way so as to allow growth as knowledge is acquired. Thank you for
reminding me of this.


>  I'm sorry Dr. Schneider, I'm not sure why the explicit relationship won't
>> be enough? I would have expect that even inferred Individuals would have
>> inferred upon them, any additional relationships I define in the equivalence
>> class - I mean, since I'm using the value keyword and not some. Could you
>> perhaps explain this a bit more?
>>
>
> Well, if you say A subClassOf P only B, then the only thing you have said
> is that every individual of A has only P-values in B. But this doesn't say
> that every individual in A has *all* B-instances as P-values. In fact, it
> even allows for any A-instance to have no P-values at all. Turning the
> "subClassOf" into an equivalence or adding the same axiom with A and B
> exchanged doesn't solve this problem.
>
> Or were you thinking of another way to express this?


Well, I wasn't planning on using the universal restriction... it would
prevent me inferring the domains and ranges of my object properties and I
quite like such information to be found by the reasoner. I was thinking of
listing all of the P-values of B as equivalence classes in A like so if if A
is the set {a1, a2, a3} and B is the set {b1, b2} then I was considering
using the following equivalence class on A:
isAppliableTo value b1 and isAppliableTo value b2

Of course this means that inferred individuals of B would be ignored, as you
have suggested. So I've read the material you've suggested (so grateful for
the link!!) and it was just what I was looking for. The paper was a bit
difficult for me to understand but the presentation was clear. They suggest
using an intermediate individual to allow, what they call, a concept product
- aptly named.

So they suggest (on page 9 of the presentation) the use of roles *R1* and *
R2* such that for *A*: R1 value x, and for *B*: R2 value x. Then R can be
defined to be the superclass of R1 *◦* Inv(R2). I'm not sure what *◦* is
suppose to mean but I've interpreted it as *R* being equivalent to both R1
and Inv(R2) and, the subclass of these object properties. I also made
Inv(R2) transitive and the reasoner infers that every individual in A has
role R with every individual in class B.

While it does give me the results I was searching for, are there any
shortcomings or reservations that I should be aware of? I was thinking of a
transitive role and an intermediate individual but wasn't sure if it made
sense to model knowledge in that way... but it seems it's an effective work
around :)

Kind regards
Pooven*
*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.stanford.edu/pipermail/protege-owl/attachments/20100122/0caf56cc/attachment.html>


More information about the protege-owl mailing list