Re: SUO: Architecture of an intelligent ontology development algorithm
Tom and Adam,
What Adam wrote gets to the heart of some issues that have
been debated on this list since it began nearly 3 years ago:
AP> .... The combination that SUMO has taken of top
> down (informed by past research in AI, philosophy and logic) and bottom
> up (driven by development of numerous domain ontologies, and the WordNet
> mapping project) is typical of any large, quality software engineering
> project. The "right" approach, if there is one, is certainly a balance
> of these extremes.
Some parts of SUMO, Cyc, and Dolce have been influenced by
applications. But the framework of upper-level categories
of each of them has been derived from the top-down. No
methodology such as Formal Concept Analysis (which is included
as one component of the IF Framework) has ever been applied
to derive the upper levels from the lower levels. The global
structure of each of them has been legislated from the top down.
AP> I also would echo your comments on evolution. An infinite (or even
> large) set of ontologies to choose from is not a standard ontology.
Correct. What we need is modular standard that relates every possible
ontology (including Cyc, Sumo, Dolce, and many others) to one another
and to the modules from which they were derived. That is the project
we approved in May.
AP> This retreads old ground, but it's important for any newcomers to this
> list to understand that while the bulk of discussion here is on topics
> unrelated to the work of creating an upper ontology, plenty of folks are
> quietly building, distributing and using formal ontologies that one day
> are likely to become de facto standards, even if largely ignored by this
> group.
Adam, what conceivable evidence do you have for the claim that
multiple independent efforts can ever or have ever or will ever
lead to a plurality of "de facto" standards?
TJ>> As to an upper ontology, I do not think that waiting for natural
>> selection,
>> operating on a pool of candidates into which some process (analogous to
>> genetic mutation) introduces novel candidates or candidate-components, is
>> likely to succeed in any time frame meaningful to us.
I agree with Tom. Adam is the one who is waiting for natural
selection to operate on that plurality.
TJ>> I think we need an
>> intentional, purposive design, tested for adequacy at each stage of its
>> evolution, against a wide range of low level ontologies taken from a
>> diverse
>> set of concerns represented by working databases in various areas of
>> academic research and business function.
What you are suggesting is similar to a motion, which I made
in April and which was approved in May. At the end of this note
is a version of that motion followed by another discussion with
Adam about ways of going about implementing that motion.
Adam, however, did not want to be "forced" [Adam's word] to
collaborate with the other groups mentioned in the combined
motion. He submitted a separate motion for an independent
project on Sumo. Adam's motion was defeated, and he withdrew
to work on his own approach with his own group.
John Sowa
_______________________________________________________
Source: http://suo.ieee.org/email/msg09372.html
Motion for a joint SUO project
Should the IEEE P1600.1 Standard Upper Ontology
Working Group commence work on a project to develop
a standard based on three starting candidates,
IFF, OpenCyc, and SUMO, and continuing as follows:
(1) The development process shall include a
collaboration of members of all three groups
and other SUO participants to determine how each
of the three starting candidates can complement
and support the contributions of the others.
(2) The results shall include a library of modules
derived from OpenCyc, SUMO, and/or other sources.
Each module shall consist of closely related
axioms and definitions for some aspect of a
standard upper ontology.
(3) The standard shall include the specification
of a methodology for testing the modules for
consistency, relating them to one another in
a generalization/specialization hierarchy,
new module that is larger and more specialized
than the modules from which it was derived.
_______________________________________________________
Following is a later note, in which I replied to
Adam's desire to "divide" the groups because he did
not want to be "forced" to collaborate:
Source: http://suo.ieee.org/email/msg09370.html
Adam,
I'm glad that you like the proposed three stages.
For reference, I'm repeating them at the end
of this note.
> This sounds reasonable. But I don't think we
> need to go to the trouble of making a motion on it.
> It's enough for now for us to agree it's a good
> idea. The only motion we need to make is to divide
> the groups for now. Can we proceed with that?
The groups are already divided. We are now at
stage 1. In an earlier note, you said that SUMO
was quite stable and usable as is. Therefore,
there is no need for a motion to do anything
more with SUMO by itself.
The same is true of OpenCyc. It is also quite
stable, and nothing more needs to be done with
it by itself. If the Cyc group as a whole
wants to extend OpenCyc, they can do so by
just contributing more of the Cyc knowledge base.
And the same is true of IFF. As Robert has
said, the theory is quite complete right now.
The main task now is to demonstrate that SUMO
and OpenCyc are both useful and complementary
contributions toward an SUO standard and that
tools and methodology based on IFF can be
usefully applied to organize and relate
both of them and any additional content
anyone may contribute.
Therefore, there is no point in a motion to
divide anything. And if we all agree that
it is a good idea for the SUO to work on stage 2
in order to produce the results of stage 3, then
lets formulate a motion to do so.
John
________________________________________________
Stage one. Three independent projects: SUMO,
OpenCyc, and IFF.
Stage two. Stock taking, analysis, and testing:
a) Study the commonalities and overlaps
between SUMO and OpenCyc and the
possible contribution of further
content from other sources.
b) Analysis of possible inconsistencies
in the various modules that make up
SUMO, OpenCyc, and other sources.
c) Development and testing of the IFF
theory and methodology on the modules
of SUMO, OpenCyc, and other sources.
Further refinement and clarification
of IFF and how it can assist in the
analysis, testing, and combination
of modules from various sources,
especially SUMO and IFF.
Stage three. Results:
a) A library of modules derived from SUMO,
OpenCyc, and/or other sources;
b) Organization of the library into a
generalization hierarchy (lattice),
certification of the consistency of each
individual module, statements of known
inconsistencies between modules, and
a history of testing of each module.
c) A methodology based on IFF and/or other
theoretical and computational techniques
that enables different modules to be
ordered within the generalization hierarchy,
combined with one another, tested for
consistency of each possible combination,
and accommodated to new modules derived
from independently developed sources.
This proposal allows changes in any or all of the
three starting projects. It doesn't commit to
a total adoption of everything from any one of them,
and it assumes that each contribution from any one
will have to survive the analysis and testing done
in stage 2. It also allows contributions from other
sources (including further work done by the same
people who initiated the first three projects).