Re: blog: semantic dissonance in uniprot

On Mar 25, 2009, at 2:55 PM, Bijan Parsia wrote:

> On 25 Mar 2009, at 18:41, Matthias Samwald wrote:
>
>>>>> Another predicate is needed that is less "rigourous" -
>>>>> owl:kindOfLike  :-)
>>>>
>>>> What do we gain from non-rigorous statements?
>>>
>>> "rigorous" is not the right term. "Less strong" is the point
>>
>> Bijan,
>>
>> Was the possibility of introducing an owl:kindOfLike - like  
>> property discussed in the OWL2 working group?
>
> No.
>
>> If yes, what came out of it? If not, why not?
>
> There was no proposal, no user base, and no implementations.
>
> OWL 2 came out of OWLED. At the first OWLED, the organizers  
> (including myself) tried to get people to figure out what they  
> needed and what could be done and do as much as we could that was  
> needed.
>
> For years, whenever I diss the mapping use of sameAs I try to get  
> people to discuss alternatives. It usually goes badly at that point :)

Same here.

>
> One obviously difficulty is that we Just Don't Know what's needed.  
> There's no real traditional semantics we can just borrow (the way we  
> do for most features in OWL; the way we used logical equality for  
> sameAs).
>
> Here's a simple issue: Should we have owl:kindOfLike only? How about  
> owl:almostCertainlyLikeInMostCircumstances?
>
> What behavior do you want from these? If it's always ad hoc (i.e.,  
> you are willing to write a program) then just coin some annotation  
> properties.
>
> It's notoriously difficult for users to just come up with a feature.  
> Generally, when I'm discussing extensions, I have to propose various  
> things and spend a lot of time explaining the trade offs to get to a  
> reasonable starting point.
>
> If HCLSIG, or some member, is interested in exploring better  
> options, I would be happy to contribute my technical expertise and  
> community connections.
>
> But I can't just pull something useful out of my ass and expect it  
> to come out all rainbows and violets. Believe me, I've *tried*. ;)

As have several of us. A lot of people with many varied backgrounds  
all agree that this needs to be done, but there doesnt seem to be any  
clear consensus on what it is that needs doing

My first question for suggestions like this ("kindofLIke") is, what  
properties would you expect it to have? First obvious questions:
is it

(1) symmetrical  (A KOL B  =>  B KOL A) ?  If not, can you give an  
example to show why not?

(2) transitive (A KOL B & B KOL C => A KOL C) ? Bear in mind that this  
can give rise to arbitrarily long chains of A KOL B KOL C KOL ... Z,  
and then A has to KOL Z. If not, can you give examples showing why  
not? If its sometimes transitive and sometimes not, can you  
characterize the differences between the cases?

(3) can any thing be KOL something, or does it only apply to some  
special kind of thingie (names, identities, whatever.) ? If so, can  
you say why it is restricted like this? (owl:sameAs is just good old  
mathematical/logical equals, which can be applied to absolutely  
anything, including classes and properties.)

(4) if A has some property or is in some class, and A KOL B, can one  
infer anything at all about the properties or class memberships of B?  
What?

In my experience so far, this is enough to reduce the discussion to  
incoherence, because no two people will agree on the answers to just  
these questions. But if KOL hasn't even got enough coherence to make  
questions like this answerable, then its not really worth trying to  
define it in any ontology language. Just make up some OWL property and  
use it, and give it no axioms. Or use rdf: seeAlso, which was put in  
the language pretty much for this reason, to be a blank carry-all for  
a loose association with no formal properties.

Pat

>
> Cheers,
> Bijan.
>
>
>

------------------------------------------------------------
IHMC                                     (850)434 8903 or (650)494 3973
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes

Received on Wednesday, 25 March 2009 23:51:29 UTC