[Top] [All Lists]

Re: [ontolog-forum] Heterarchy & Hierarchy, oh my my

To: "[ontolog-forum] " <ontolog-forum@xxxxxxxxxxxxxxxx>, Wacek Kusnierczyk <Waclaw.Marcin.Kusnierczyk@xxxxxxxxxxx>
Cc: bfo-discuss@xxxxxxxxxxxxxxxx, obo-relations@xxxxxxxxxxxxxxxxxxxxx
From: Alan Ruttenberg <alanruttenberg@xxxxxxxxx>
Date: Sat, 3 May 2008 01:17:51 -0400
Message-id: <6A4BE6C0-C1BC-423E-99E8-EF0993A156D1@xxxxxxxxx>
On May 2, 2008, at 3:30 PM, Wacek Kusnierczyk wrote:    (01)

> I was repeatedly complaining on the OBO-  and BFO-related lists  
> about the insistence, within that framework, on  single  
> inheritance.  Illustrative cases of the arbitrariness of choice
> which distinction is to be prioritized over others (i.e., placed  
> higher  in the taxonomic tree) can be found in virtually any OBO  
> ontology; for  example, PATO starts at quality, distinguishes  
> qualities of occurrents  from qualitites of continuants, and then  
> within each of these  distinguishes, independently, monadic and  
> relational qualities.  It  could well have been the other way round.
> I mention this because the answer to my complaints, if any, was   
> invariably that single inheritance a) increases efficiency of  
> reasoning,  b) is more natural and easier to use, and c) is good for  
> interoperability.
> To my simple mind, these claims do not correspond well to yours  
> above.   It would be interesting to see a discussion on this issue,  
> though the  argumentation provided by proponents of the other view  
> typically did not  go beyond rather vague handwaving.    (02)

I'm familiar with this policy, and balked at initially, but now  
consider it (slightly modified) to be quite useful.
I haven't heard the argument (a). The argument for (b) that I agree  
with is that it's easier to make mistakes when multiple superclasses  
are asserted. (It's really easy to make mistakes even when only  
*single* inheritance is used ;-). I'm not persuaded that naturalness  
has anything to do with it, though people I respect have different  
views (e.g. Werner :).    (03)

I've seen the disadvantages of multiple asserted inheritance when  
reviewing, e.g. the Cell ontology in OBO, where it was demonstrated  
(by the authors) that it was quite easy to find mistakes of the sort  
where all of the properties described in the definition of (multiple,  
and transitive) superclasses were not true of instances of the class  
in question. Regarding (c) what is perhaps being referred to is that  
if one practices "normalization" in the sense that Alan Rector  
proposes [1] then the component single inheritance ontologies from  
which more complex terms are constructed are more likely to be able to  
be reused by other projects. Certainly that's the intention of  
creating and using PATO. I've found the exercise of factoring  
definitions in this manner is often helpful and is conducive to  
helping the sorts of people I work with think carefully when  
constructing ontologies.    (04)

The current thinking about working on the OBO ontologies is, primarily  
for reason of easy maintainability under distributed development, that  
while a reasoner may conclude that there are multiple superclasses,  
only a single named superclass of a class is asserted. Class  
definitions are constructed by adding conditions that use a smallish  
set of properties that related to terms from other hierarchies/ 
ontologies. In addition, defined classes (those defined using  
necessary and sufficient conditions) can be added pretty much at will,  
and while a reasoner may infer multiple superclasses because of this,  
these are less of a problem because the reasoner manages their  
placement.    (05)

Perhaps such methodological approaches will be less important as time  
goes on and use of powerful reasoners become commonplace, in  
conjunction with more axioms about classes being stated in a way that  
is faithful to the intention of the writer and available for reasoning  
with. Such automated checking is invaluable and routinely finds  
inconsistencies in my and other's work. In the end I expect the number  
of errors which are found by reasoners (and then corrected) will far  
outweigh the number that are prevented by choosing one methodology  
over another. OTOH, not all kinds of errors can be found by reasoners,  
and any methodology that helps its practitioners think clearly is  
beneficial.    (06)

A couple of comments to close:    (07)

1) The specific are of PATO that Wacek mentions is *not* considered to  
be well done, and there are quite a few OBO people, myself included,  
who have it as a target for cleaning up. This is no secret, as we have  
discussed it on the relevant OBO lists, and on the PATO tracker.    (08)

2)  I'm not familiar with FCA - I suppose I should read up on it, but  
I have to say it makes me suspicious when I see claims about any  
methodology that "automatically generates a ... system that is  
*proved* to be consistent", if the methodology is one that is intended  
for people, rather than machines, to apply.    (09)

-Alan    (010)

http://www.cs.man.ac.uk/~rector/papers/rector-modularisation-kcap-2003-distrib.pdf    (011)

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    (012)

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