Sunday, March 19, 2006
The Second School of Semantic Science
ontologyMapping Glass Bead Games
Lars,
Recently I had the opportunity to do an analysis of translatability between the four types of information structures
which I label as:
1) RDBMS
2) object oriented (and UML) information models (example: SOA-IM)
3) RDF with description logic
4) n-ary ontology
As you know, this is a hard problem, in general.
It is my thinking that a topic map is best typed as an n-ary, ie having a most general graph like nature and a localization to a single node. Topic maps have a findability technology based on Xlink, I understand from reading. Description logic is not applied to topic maps, through rules may be. The findability technology for the more general n-ary has several different realizations, one by Richard Ballard (which I have not tested) and one based on a type of “keyless” hash table.
http://www.ontologystream.com/private/NIMAtechnical.htm
This is an entirely different approach that we, the second school, hope to make transparent, by illustrations to cell and gene expression science.
Perhaps the transparency is seen as a difference between a fixed n-ary ontology and a formative one. We feel that it is proper to say that this difference is most noticeable at choice points (where ambiguity/degeneracy is observed).
The n-ary is
1) merely a location in the graph (in the case of the fixed ontology)
2) is created through the entanglement (awareness) of templates and invariances) (in the case of the formative ontology)
If you have a comment about this labeling, I would be interested in making a change. It is noted that the ontology designer passes through choice points (without knowing this at each moment) when making decisions about the assertions made in the ontology.
I remember that (from my recall) the excitement about topic maps changed because a "XTM process model" could not be agreed on. Process is a type of entailment, so the absence of description logic for TM seems consistent, AT LEAST BASED ON THE CURRENT VIEWPOINTS.
I feel that there is an “unexpected” process theory and am proposing this as a formative category theory, with formalism and a possible OASIS specification.
Your work at:
http://www.ontopia.net/topicmaps/materials/tmrdfoildaml.html#N463
is useful. We are all making some progress, I see.
A number of issues have arisen that you likely have insight to.
Perhaps it is best to use the language in a recent post to the BioPAX forum
> But this link now shows they are just a design pattern of triples:
>
> http://biopaxwiki.org/cgi-bin/moin.cgi/AssociationPattern
>
> Can OWL use such structures?
Sure. It has limits, so
> some (over) complex constraints seem to need extensions.
Physical entity participant (PEP) is exactly a role-association
pattern.
It defines the role of a physical entity in the context of an
interaction. In many senses it is a blank node. However it gives us
nightmares when we try to express the actual semantics with
constraints.
One problem is the desire to associate every information related to the
role of a physical entity in the context of an interaction to the PEP.
In that sense PEP is a bit overloaded. We could have solved it by using
"multiple association nodes", one for every constraint. But
it creates
parallel links and decreases the readability of the ontology.
The role-association is not a magical thing, I am not even sure if it
classifies as a pattern. Every
good OO designer uses it, and there are
papers out there to use AOP to make it even easier. When you are using
it in a OO language, you typically want to check extra constraints as
invariants. In the case of OWL,
one simply has no means to define
constraints over n-tuple associations , so one ends up not constraining
it. One obvious solution is to
use rules, (please correct me if I
am wrong) there is not any standard rule system yet.
(I note that several of the participants are very committed to the reductionist process (as a matter of “science”), so that natural complexity (as defined by Robert Rosen) is ignored. We are fortunate that most of the BioPAX group who know first hand that living systems are not simple and thus not (at least shown to be) computational.)
In the OASIS BCM specification rules are used, but there is also the notion of a choice point. The BCM choice point is still overly complicated, but in essence is a “let the human(s) decide” halting conditions for web services.
The nary can be (and perhaps should) be associated with a formative relationship between "players", to use the language of
http://biopaxwiki.org/cgi-bin/moin.cgi/AssociationPattern
In vivo (in the reality of social discourse or cell expression), the “association” is created as players (reactants) express through roles that are defined by the “emerging association”. In the formative process the players and the roles are selected in a way that is properly termed “complex” and it is this complexity that the second school asserts is not reducible to simple rules. Once the expression has occurred, there is (a possibility of) a reduction to a simple model that reflects the global functional responses of the expression as well as the combined constraints of the physical entity participants. It is these functional responses that end up making the expression full constrained.
In an analogy, it is the business person reflecting on the BCM blueprints and internal experience that ends up making the decision as to which collaboration contract to agree to.
What seems natural then is to treat the formation of the “specific” physical entity as a reification process, where instances are contributing to the formation and/or reinforcement of natural category.
A formative ontology architecture is to be proposed as an OASIS specification and is to be applied to the formation of the BCM choice points. But the theory is general and so should the specification.
We are looking for comment.