SUO: Re: Architecture of an intelligent ontology development algorithm
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
JA = Jon Awbrey
RC = Richard Cooper
RC: Jon, I still like the advice you gave a few iterations ago,
so I snipped it out of some recent emails and reapplied it
as below:
JA: I would have thought that a fairer summary of what's been said here
all summer long on many different threads is that there is really no
such thing as a hypothesis-free algorithm for discovery -- actually,
that's more like a summary of what's been discovered about discovery
over the last few thousand summers, but who's counting? So I think
that a better question might be something along the following lines:
JA: How are concept-driven (analytic, axiomatic, rationalist, top-down) methods
and data-driven (synthetic, contingent, empiricist, bottom-up) procedures
best to be integrated in human inquiry, or in the reconstitutions thereof,
given that the distinction between analytic and synthetic is more relational,
interpretive, or "situated" than it is absolute, invariant, or "essential"?
RC: This succinctly states the core idea of how an automated or even
computer-aided tool could proceed to generate useful ontologies.
RC: Hypothesis formation is based on learned conceptualizations
from past experience. I'm still reading your refernce to
"anticipatory" systems:
http://www.anticipation.info/texte/rosen/anticipatorysystems_rosen.pdf
Did I give you that? I meant to give you Mihai Nadin's home page:
http://www.code.uni-wuppertal.de/welcome.shtml
RC: but I like the general tone of a faster-than-real-time model M guiding a
real time situated system S in modifying its behavior. However, the linear
system analogy is a bit antiquated, and the term "hypothesis formation" (or
"abduction" as you Peirceans prefer) isn't conceptually compatible with
a linear system model.
RC: I like the GA and GP models better. Using symbolic expressions
rather than linear projections seems more intuitive and natural
for the areas I like to concentrate on. Forming new expressions
from combinations of older expressions makes it possible to trace
the formation of hypotheses to experiences, much like requirements
can be made traceable to designs using a graph model. So a GP model
is probably most natural.
Many people like these things, but I tend to shy away from flip-a-coin methods
for much the same reason that I don't buy theorem provers that say "Yup, it's true
-- but I can't tell you why". It's the understanding that I'm after, not the simple
reassurance.
RC: Then there's the issue of where data comes from to drive the formation of hypotheses.
For an automated system to reach "good" hypotheses, it should have "good" data --
data from naturally occuring intelligences. So I subscribe to the idea of
data mining of English text that has been well written, well annotated,
and reviewed by various interested parties.
Sigh, if only we could get Nature to write better.
RC: It seems to me that the nominal task of the SUO group could be best
addressed in producing such a tool (automated, or computer-aided, ...)
to browse English sentences, data models, UML designs, CGs, or other data
we would consider useful for the purpose. At least that approach avoids the
legislative extremes, because we can argue about what data to use, what methods
of representation, and how the system should function instead of what the ontology
itself is.
RC: We seem to be running dry on ontologies themselves.
But you're asking for oceans just to prime the pump ...
and then all you'd get is salt water anyway ...
Jon Awbrey
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o