W3C home > Mailing lists > Public > public-semweb-lifesci@w3.org > May 2007

Re: Advancing translational research with the Semantic Web

From: Pat Hayes <phayes@ihmc.us>
Date: Mon, 21 May 2007 21:01:42 -0700
Message-Id: <p06230941c27814213794@[192.168.1.4]>
To: Chris Mungall <cjm@fruitfly.org>
Cc: Phillip Lord <phillip.lord@newcastle.ac.uk>, public-semweb-lifesci@w3.org

>On May 21, 2007, at 11:24 AM, Pat Hayes wrote:
>
>>
>>>  >>>>> "CM" == Chris Mungall <cjm@fruitfly.org> writes:
>>>
>>>
>>>   >> Out of curiosity, can you describe how different or similar this
>>>   >> is to the result that you can achieve in the N-ary relation
>>>   >> design pattern for OWL?
>>>   >>
>>>   >> Obviously, building things into the DL is nice, but it's not
>>>   >> currently representable in OWL, so would require tooling support,
>>>   >> while the OWL N-ary relation pattern doesn't.
>>>
>>>   CM> I'm afraid I'm unclear how to state the OWL n-ary relation
>>>   CM> pattern (http://www.w3.org/TR/swbp-n-aryRelations) where I
>>>   CM> really need it. In all the examples given, the "lifted"[*] n-ary
>>>   CM> relation was never truly a relation in the first place and
>>>   CM> always better modeled as a class. It's kind of cheating.
>>>
>>>Well, it is kind of cheating, yes, although if it works...
>>
>>No, really, its not cheating. This reduction of n-ary relations to 
>>binary+unary relations is quite general and quite sound, and has 
>>been known and thoroughly understood for over a century. It can 
>>always be done, and it often makes perfectly good intuitive sense. 
>>The 'thing' that the arguments all relate to is something like the 
>>event, or fact, or situation, or state of affairs that holds, etc.. 
>>(choose your favorite terminology)
>
>Sorry, I shouldn't have used a pejorative term like cheating, I was 
>just pointing out that if I am presented with a template for a 
>solution for problem X and that solution is to transform to 
>representation Y, this doesn't help if my problem was actually X2 
>and I would have used Y for X in the first place.
>
>Or let me be more concrete: using an ontology like BFO (perhaps 
>DOLCE too) as a guiding framework

Well, that is already a poor decision, IMO. But let us not start a 
quarrel about foundations.

>, one would use owl classes and not owl object properties for 
>diagnoses, observations, purchases and flight segments (the examples 
>given in the n-ary note).

There is something very wrong with this idea. Take diagnoses. A 
diagnosis is, presumably, a diagnosis OF something. That is, it is - 
or at any rate intimately involves - at least a binary relationship. 
If you model these things purely as classes, you cannot describe a 
binary relationship (such as the between a diagnosis of a disease and 
the disease that it is a diagnosis of.) You must involve properties 
in the mix somewhere.

>  Applying the same pattern to relations that should be modeled as 
>relation leads to more design decisions and more complex 
>representations.

I am at a loss to understand why you feel that anything 'should' be 
modelled in one way or another.

>>>   CM> What if my n-ary relation is transitive or if the 3rd argument
>>>   CM> is a temporal interval over which the relation holds?
>>>
>>>The former is hard because it's not clear what do you with n-ary
>>>relationships. I think that this is true for any
>>>representation. Fundamentally, if you say "a is part of b" and I say
>>>"b is part of c", then is "a part of c" and according to whom?
>>
>>Right, it simply isn't clear what 'transitive' means for relations 
>>other than binary. Try writing it out as an axiom in logic to see 
>>all the different possibilities.
>
>This would have to be done on a case by case basis. I believe 
>someone is working on this for the time-indexed temporal relations 
>in the OBO relation ontology

It is really a pity that people are not doing this in an adequately 
expressive logic. It is straightforward, for example, to describe 
arbitrary patterns of transitiveness, functionality, etc., in Common 
Logic. There is no need to make catalogs of 'cases'. BTW, Bill 
Andersen at Ontoworks has a 'foundational ontology' which defines 
things such as N-ary relations which are functional in the M-th 
argument place, and the like. I prefer to call this a "structural 
ontology", myself. It is really concerned with giving names to a 
number of useful inferential patterns. Cyc also has a large number of 
such structural definitions in its publicly accessible core.

<snip>
>>
>>My advice to new users is to forget completely about N-ary 
>>relations. Tell yourself that there are no relations in nature 
>>above binary. If you think you need one, re-think what you are 
>>trying to say so that it all fits into the binary case. Chances are 
>>this will be fairly easy to do, and as a side-effect, you will 
>>probably then be much more clear on what exactly it is that you 
>>want to say about transitivity, functionality, domains and so forth.
>
>I'm afraid that one of the core ontologies we are using disagrees with this:
>http://genomebiology.com/2005/6/5/R46

As far as I can see from a quick scan through the paper, this is 
purely concerned with binary relations except for one special case of 
trinary relations where the third argument is times; that is, binary 
fluents. This one exception is well-known and conceptually clear if 
used consistently (as it is here). One can think of this special 
temporal third argument as a kind of temporal subscript to a binary 
relation: the special nature of this case is reflected in the special 
terminology of continuant/occurrent used to handle temporal 
individuation. Apart from this, all relations seem to be binary: and 
as I say, this is 'really' binary.

To encode the temporal structure in OWL takes a little care, but this 
particular case has been worked out already, and can be handled 
uniformly in all the cases where it arises. It needs to be handled 
specially in any case if OWL reasoning is intended to mirror the 
special temporal inferences which arise from the continuant/occurrent 
framework. In particular, it would be a mistake to treat a binary 
fluent as though it were merely a 3-place relation.

Pat

PS. There is one very unfortunate terminology clash in the 
above-cited paper. It has been conventional throughout KR work for 
about the last 15 years to use "is-a" for the individual-class 
relation of membership and "subclass" for the class-class relation. 
OBO regrettably uses "is-a" for the latter, which is almost certain 
to lead to enormous confusion. It is probably too late to do anything 
about this now, but it is certainly a most unfortunate jumbling of 
established usage.
-- 
---------------------------------------------------------------------
IHMC		(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.	(850)202 4416   office
Pensacola			(850)202 4440   fax
FL 32502			(850)291 0667    cell
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes
Received on Tuesday, 22 May 2007 04:01:57 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:20:26 UTC