Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

SUO: Re: SUO Comment #2




Mike Uschold wrote:
> *     This standard will enable the development of a large (20,000+)
> > general-purpose standard ontology of common concepts to be
>> developed, which will provide the basis for [defining all of the
> > concepts in] middle-level domain ontologies and  lower-level
> > application ontologies.
>
>'all' may be rather a tall order. It is certainly a worthy long term
>aim, but
>it may not be attainable in practice.  There may always be interesting
>concepts that don't seem to fit anywhere.  At best, we might
>asymptotically
>approach this goal.

I agree with Mike. I vote for deleting the words "defining all the 
concept in" from the above statement.

Further comments:

1. Meaning of "defining".

The verb "define" occurs often in this discussion, but it would be 
useful to make clear that, strictly speaking, most concepts just 
CAN'T be defined, in the sense that there is no way to come up with 
necessary and sufficient conditions for them. For sure, the majority 
of concepts within a large standard ontology will be primitive, in 
this sense. What we call "definitions" will be in practice 
CHARACTERIZATIONS of a concept in terms of other concepts. This is 
the case of many dictionary "definitions", for instance.

2. Meaning of "upper"

At 11:11 AM -0400 30/8/2000, Jon Awbrey wrote:
>Douglas McDavid/Boulder/IBM wrote:
> > At some point we need to understand whether
> > upper means generic, or whether upper means abstract, or whether upper
>> means ubiquitous, or whether upper means meta, or just exactly what
> > does "upper" mean?
>
>Seriously folks, let me try to formalize the alternative possibilities
>that Doug McDavid mentions.  I will start out in the following way,
>just as an experiment with notation, nothing more:
>
>1.  "upper" -<- "abstract"
>2.  "upper" -<- "generic"
>3.  "upper" -<- "meta"
>4.  "upper" -<- "ubiquitous"
>5.  "upper" -<- ???

In my opinion, the meaning of "upper" we need for the SUO is the 
union of "meta" and "generic". That is, we need to include in the SUO 
all the meta stuff Douglas McDavid mentioned, plus a reasonable 
number of "generic" (or "general") concepts (a rough estimate of a 
couple of thousand is OK with me, as we discussed in the previous 
thread). With "generic" I mean here "common (and relevant) to 
multiple domains/applications". Of course this is a very generic 
definition of "generic"....

3. What kind of classes in the SUO?

At 6:21 AM -0600 30/8/2000, Douglas McDavid/Boulder/IBM wrote:
>Can we accept classes that would
>contain _identifiable_ things, or do we have a set of _properties_ that can
>be applied in a standardized manner to things that are only identifiable
>via more domain-specific ontological constructs?
>
>To me, the most useful piece of work we could provide would be a principled
>articulation of categories that are immediately applicable to identifiable
>things in the world.  By principled, I mean that we would argue about
>whether each category in the SUO should be completely disjoint from all
>others, and whether the set of categories should be all-encompassing of all
>possible domains of concern to users of the SUO.

This observation seems to be very much related to the recent work 
Chris Welty and I have done, presented at the AAAI tutorial on 
conceptual modeling and ontological analysis. Among other things, we 
present a formal way to distinguish between properties according to 
whether or not their instances are "identifiable things" (using the 
above expression). We further discuss the notion of a "backbone 
taxonomy" that only contains properties of this kind, plus their 
generalizations. This gives a criterion for deciding the properties 
to be included in the SUO, which is pretty much in line with 
McDavid's suggestion above. Disjointness constraints between these 
properties are the result of ontological meta-properties that reflect 
their formal behavior with respect to identity, unity, essence. See 
http://www.cs.vassar.edu/faculty/welty/aaai-2000/ for further info.

4. Lattices and constraints

At 7:51 PM -0400 31/8/2000, Patrick Cassidy wrote:
>But the experiment that urgently needs trying is to create as best 
>we can an inventory of basic concepts, and then work to define the 
>more complex concepts in those terms, putting our definitions of 
>complex concepts in a common repository for use by all. Only then 
>will we be able to learn whether those hundred thousand-plus complex 
>concepts will be definable as combinations of the basic concepts.

At 10:53 AM -0400 1/9/2000, John F. Sowa wrote:
>The long and honorable tradition for representing such
>combinations is called the theory of lattices.  Leibniz used
>similar combinatorial methods for his "Universal Characteristic"
>which was a 17th century attempt to develop a "standard upper
>ontology."
>[...] The techniques apply to categories at every level of the
>ontology from top to bottom.

Just a word of caution: theory of lattices is of course very 
powerful, but before exploring the combinatorial results of a set of 
basic distinctions we must first check for possible mutual 
dependencies. In many cases, basic primitives turn out to be NOT 
mutually independent, so that some of their combinations are 
naturally excluded. For instance, this is the case of the 
distinctions appearing in Sowa's "crystal", discussed in the ontology 
chapter of his book: according to my understanding, the distinction 
"continuant/occurrent" is not independent from the distinction 
"physical/abstract", since continuants and occurrents are both 
physical (i.e., located in space/time). Therefore, there are no 
entities that are both occurrents and abstract (scripts, for 
instance, carry information about occurrents, but they are not 
occurrents). [NOTE: this is just an example, whose detailed 
discussion would require a different thread. The point is that 
applying theory of lattices alone may lead to fascinating symmetric 
structures that don't reflect actual ontological constraints].

-- Nicola


  ---------------------------------

Nicola Guarino
National Research Council	phone: +39 O49 8295751
LADSEB-CNR		fax:   +39 O49 8295763
Corso Stati Uniti, 4		email: Nicola.Guarino@ladseb.pd.cnr.it
I-35127 Padova
Italy

http://www.ladseb.pd.cnr.it/infor/ontology/ontology.html
(***updated 28/8/2000***)