Monday, November 21, 2005
Center of Excellence Proposal
à
ONTAC stands for Ontology and Taxonomy Coordinating Working Group
It is a working group of
Semantic Interoperability Community of Practice (SICoP)
Communication to the working group
from Paul Prueitt
****
(not posted to general ONTAC discussion)
I am posting only to those who I feel might be interested in a technical discussion about foundational issues. Please advise me if you do not wish to be included.
Denise,
I was reviewing the messages on the ONTAC general discussion index and came across one by you:
http://colab.cim3.net/forum//ontac-forum/2005-10/msg00159.html
to quote some part of your message:
... it is not the exact relationships that are described here that
are as important as the idea that relationships can have meanings, and
that the
meanings can be defined and managed.
This is what we are beginning to do by
examining the relationships among concepts in large training sets which
are
domain specific.
There is one other problem area that we also need to tackle in this
community -
the difference between a classification (mathematical set theory) and a
classification of concepts (biological taxonomy). These are two different
behaviors, and I see them referenced too often as the same approach.
<end quote>
I found this insightful on the most critical foundational issues, in particular the nature of relationship. I have worked a long time to get a handle on the nature of relationship.
Subsumption is perhaps so overused as to make the development of a theory of relationship difficult. See the BOF (Basic Formal Ontology) relationships as an example of a reasonable theory of relationship.
http://www.uni-leipzig.de/~akumar/coling.pdf
Gerald Edelman’s (Nobel prize winning in immunology) concepts about degeneracy is one way to test the BOF theory of relationship. A review of this notion finds it related to structure-(substructure)/function mapping, and thus to the phenomenon of emergence. Thus, the concepts related to a processes beginning and end is bring into the picture. It is my opinion that these are hard issues only if one avoids addressing them.
Also see the key Topic Maps author, M Biezunski’s distinction between Semantic Web and other types of standard means to define relationship
http://www.gca.org/whats_xml/xml_files/issue31/swgca.htm
In my work I have always thought of the relationship as like a range element in a classical elementary function as in f(x) = 2x. The range element is the 2x when a domain element x is given. Domain element of 3 --> range element of 6. (3,6) = (x,y).
Associations in Protege editor have domains and ranges. The domain is the set of subjects to the association and the range is the set of predicates to the association. The association is a relationship also in the graph sense, < a, r, b > with the r being the connector. But in the graph sense the relationship is not really always a "function" with domain and range, as it is with Protégé. The graph sense is "weaker" and thus has more possible interpretations.
So the notion that a relationship is the range element of a larger construction is useful in allowing the relationship to change as the real world situation changes. To create something that computers can use that is mapped to this type of thing we need to know about regularity in the real world and to find design patters. (I would think).
This is not "the normal" way most consider ontological notation. It is, however, consistent with how the Soviet school (Dmitri Posplev and Victor Finn) considered what they called "syntagmatic units", ie something of the form < a, r, b >.
Categorical collapse around observed ontological structure
was considered to be how combinatorial issues DO NOT ARISE.
I know that this statement is often not regarded as understandable by many in the computer science community. I try to say this without insulting anyone and with great care. But it is important to realize that there is an opposition to doing something that is in fact what needs to occur. My feeling is that the opposition comes because the notion of relationship used most of the time in the semantic web community is one that does not support the highest quality of ontological structure. The hierarchy has to be set aside if this highest quality is to be achieved – in my opinion.
And I have written about how the Soviets saw this, and how Peirce saw this (and how others have seen this). There is an order to the world that is part of its reality (and thus any high fidelity ontological model must reflect the real order in the real world ).
A participant in the forum wrote:
All,
Regarding John Sowa's
suggestion of a collection of modules, I question if it is possible to achieve semantic
interoperability with this approach, since we will certainly have n-squared
modules?
Jim
and I cannot understand why this person feels this way, unless I understand that his strident objection is based on an inability to move to the suggestion with the limited subsumption tools. At this point I realize that he is making an important objection, but one that has to be understood.
Of course he must understand that there is no combinatorial problem with truly representing human knowledge. Or so I reflect.
It is my observation that there can be no understanding of human knowledge unless the regularity in nature is reflected in our representation of knowledge.
Suppose that we have a biological reactant that is changed by a catalyst. Its relationships to other reactants has undergone an transformation. Is there knowledge that can serve as a design pattern (as in Gangemi's work) so that an ontological model can be produced having knowledge of the reactant before during and after the transformation?
Of course there is. But it has to come from real biological scientists, not computer scientists. There may also be, as Gangemi suggests, domain dependant patterns. And if the domain becomes very detailed, we find that this is the cutting edge of science.
A similar "suppose" could be generated for business processes where whole greater that the sum of parts occurs.
This type of work involves both induction and abstraction work by someone who really understands something about the reality of the world being modeled.
The trusts of my objection regarding formalization and theories has to do with my sense that the notion of subsumption in general, and is-a and part-of in particular is to often depended on to claim that a specific ontology is "correct".