RE: A Few Thoughts on an Ontology as a Self Organizing System

["Cox, Bruce" <Bruce.Cox@USPTO.GOV>]

Some equally fuzzy remarks:
1.  As I understand it, for structure to appear in a complex system, the
system components have to exhibit some kind of individual behavior, as
chemicals in a test tube where, based on complex, spontaneous chemical
interactions over time, the system can become more organized rather than
less; somewhat like the emergence of heavy elements (more complex nuclei
etc.) and solar systems during the evolution of the cosmos from
primordial gas.  I can see the analogy in the Web as a whole, where
humans are involved, or perhaps where servers are active without
premeditated coordination.  In a system that consists only of
vocabularies, where is the activity from which structure emerges?  The
addition, subtraction, and change that you describe is imposed from the
outside, not from within the elements; elements have only static
properties that don't change without external intervention.  I don't
suggest that life is required, but some sort of internal activation
(like electromotive forces or human passion) seems to be.
2.  Are such systems of vocabularies in a domain sufficiently large and
homogenous to exhibit the statistical properties of complex systems?  I
don't know.  On the Web there may be large enough populations of
homogenous objects in a domain to exhibit such behaviors, rather than
their vocabularies.  To continue a typical ontology example, there may
be enough different bottles of wine in the world, or even types of wine,
but I wonder if a collection of vocabularies would be sufficient.  If
you combined all the vocabularies from all the domains?  (But what would
be the point of doing that?)
3.  Those who maintain taxonomies can testify to the forces acting on
the domain that create the need for change in the taxonomy.  Ask
(Chemical | Physics | Biological) Abstracts how and why they update
their indexing vocabularies.  At some point in time, the number of
papers that mention a specific topic, and the demand for access to
papers on that topic, rise to the point that indexers take note, and
then revise the vocabulary to include terms for the topic.  This does
not indicate that some new topic has appeared, only that it has gotten
enough attention to justify the effort of revising the taxonomy.  The
point in time when the topic first appeared in papers is uncertain
because 1) the pre-revision taxonomy did not "know" it was there and 2)
it will have been discussed informally long before appearing in a
refereed paper.  A specific revision is not an emerging behavior of a
taxonomy, no matter how complex, but an outside influence, forcing
change on what normally has a great deal of inertia associated with it.
Stability (inertia) is necessary for the success of a taxonomy, but not
sufficient.
4.  If you want to model the evolution of an ontology, you may want to
start with the population in the domain who evolved its vocabulary in
the usual human manner.  A taxonomy is a subset of that general
vocabulary, where each term is assigned scoped and documented meanings
with the consensus of the population.  Are taxonomies a property that
emerges from the behavior of the experts in a domain?  Are ontologies
(assigning specific terms from a taxonomy to specific objects [is that
right?]) a kind of second-order property that emerges from the behavior
of a subset of experts in a domain?  If the population of a domain is
large enough, it may be that it would qualify as a complex system, but
is the subset of experts who make ontologies large enough?
5.  None of this solves the problem of the cost of maintaining an
ontology, which I presume is the motivation for automating it.  No
matter the answers to any of the above questions, none of them will
render the process automatic, as far as I can see.  Expanding a taxonomy
requires judgment based on a sense of the market and its needs.
Assigning terms to objects requires an understanding of the object and
the terms that transcends a coincidental lack of common character
strings.  It will require the attention of humans, the commitment of at
least one institution, and investment from at least the population of
the domain (using it and paying for it), to maintain an ontology.  Even
if complex systems analysis is applicable, I understand that it is a
statistical science, which does not in general apply to individual
events but to a population of such events, so I doubt it could be used
to automate the revision of an ontology.

Bruce B. Cox
SA4XMLT
USPTO/OCIO/AETS
703-306-2606


-----Original Message-----
From: Roger L. Costello [mailto:costello@mitre.org] 
Sent: Friday, September 26, 2003 4:57 PM
To: xml-dev@lists.xml.org
Subject: A Few Thoughts on an Ontology as a Self Organizing System

Hi Folks,

This is a continuation of the discussion that we had last week on
complex systems.  I have a couple of somewhat fuzzy ideas that I would
like to throw out.  My objective is to stimulate the flow of ideas, and
perhaps bring clarity to my ideas. 

Mike Champion made an interesting statement last week while discussing
semantics:

> But how about the messy real world most of us must operate in, where 
> there is an intent to deceive (spammers, virus writers, software 
> companies with patents on common sense, politicians starting wars [or 
> questioning the definition of "is"], ad nauseum)? How about in pop 
> culture contexts where meanings of words are changed literally for the

> fun of it?

That got me to thinking.  Suppose that we define the collection of all
XML tags that are used within a domain as a "system".  Let me refer to
each individual tag as a "part" of the system.  The system is
dynamically expanding and shrinking, i.e., parts (tags) are being
introduced/withdrawn all the time.  There are both fixed and changing
interactions in the system, i.e., some parent/child, sibling, semantic
relationships are fixed, others change.

What we have is a complex system.  I could continue on with this
description and talk about system properties, emergent properties,
attractors, etc.  However, since the topic is semantics, I would like to
focus on the use of ontologies in such a system.

Ontology languages such as RDF Schema and OWL provide the ability to
*statically* capture semantic relationships.  However, as Mike points
out, semantics is a continually evolving thing.  As a system evolves, so
must the ontology evolve.  In fact, an ontology must be part of the
system.

"The essence of self-organization is that system structure often appears
without explicit pressure or involvement from outside the system."[1] To
manage evolving semantics a system must self-organize as semantics
evolve.  In other words, an ontology must be a constantly evolving
entity.

How can we create an ontology that evolves?  Here is a thought: express
semantic relationships in an XSLT document!  An XSLT stylesheet has an
interesting property of being able to output a modified version of
itself, i.e., the output of the stylesheet is another, modified,
stylesheet.  The output stylesheet may contain template rules that have
been modified to reflect changing semantics, and additional template
rules that contain new semantic relationships.

Honestly, I am not sure how one would express semantic relationships in
a stylesheet. For example, how would you express that a SLR is a type of
Camera, or aperture is synonymous with f-stop?

Well, that's it.  As you can see my ideas are rather fuzzy, but perhaps
they will stimulate your thoughts.  /Roger

[1] Self-Organizing Systems FAQ for Usenet newsgroup
comp.theory.self-org-sys