Re:
PC> ... the funding agencies have a very narrow focus,
> at least with respect to ontologies. (01)
In an offline note, one member reminds me that the Cyc experience could
discourage funders.
I have some thoughts about that. (02)
It seems to me that the experience of Cyc could provide some useful
or at least interesting lessons on what can and can't be done with the
logical representation of knowledge, but all we have is anecdotes. I
wish someone with a good knowledge of the history of Cyc would write an
objective narrative of what went on there, what was actually tried, and
how the trials succeeded or failed. I am particularly disappointed
that they never seemed to put enough emphasis on the connection between
the ontology and language, though they have done and continue to do
some computational linguistics. To me there appears to be there is a
large amount of valid and useful information in Cyc, and the
documentation of each element, though often sketchy and sometimes
absent, is still overall the best I have seen in explaining intended
meanings of the concepts they represent. (03)
At this point, my interpretation is that, though (1) it is true that
having a rich store of existing background knowledge is essential to
enable the interpretation and use of new knowledge; and (2) the general
method Cyc uses to represent knowledge is valid and useful (perhaps a
little too baroque and impenetrable in parts?);
That (3) we are discovering that being able to use represented
knowledge effectively is probably 10 or more times harder than
representing the knowledge. (04)
In this interpretation, the fact that Cyc hasn't been able to impress
many people may be due to either:
(3.1) Cyc just doesn't have the resources (person-time) to exploit the
knowledge they have encoded, or
(3.2) Cyc doesn't really know how to exploit the knowledge they have
encoded, or
(3.3) both.
It doesn't necessarily mean that the knowledge they have encoded or
the representation they used is invalid or useless. There is still
parts of the story yet to be lived, not just told, since Cyc became
more public and the Cyc foundation was formed. (05)
Meanwhile, I find the Cyc ontology (and SUMO) a useful source of
represented knowledge that I intend to use so that I don't have to redo
the parts they already did that I find I agree with, which, in the Cyc
taxonomy chain, is over 90% (except for their execrable conflation of
physical objects with attributes and substances, which I have had to
disentangle). Also, John Sowa made a point in a previous post that we
should keep the representation as close as possible to linguistic
usage. I fully agree with that, both to make the ontology easier to
understand, and also to make it easier to create the links to
language-understanding programs. (06)
So should that make the agencies reluctant to fund other approaches to
knowledge representation? (07)
If they think that there is no hurry, and they feel a need to hoard
precious resources, perhaps the answer would be "yes" -- but . . .
As I mentioned earlier, in the brief interval between the time I got
the note that inspired this one, and the time I sent this, the country
as a whole has lost a lot more than the $20 million it would cost to
fund a plausible effort to build an ontology that had enough associated
demo applications to make it easier to use than Cyc, and had a much
wider following in the ontology community solely by virtue of a wider
range of participants with significant funded time to devote to the
project. So I think that a sense of urgency is justified to prompt
additional efforts, not just to duplicate what Cyc did, but to exploit
what they created as well as what others have created in the interim.
Now, PatH made the point that DARPA doesn't care a fig about the
strength of our economy, and even though the strength of our economy is
what makes DARPA possible, I can't argue with that. I just think that
if DARPA doesn't, someone should, but I am afraid it won't happen if
the funders are given the impression that there is no point in trying.
Can we agree on some plausible approach? (08)
My own inclinations are very experiment-oriented. As soon as we are
into building programs we become engineers, and as engineers we test
our hypotheses by asking whether or not our artifacts actually do what
we want them to do. I think that we have reached a stage in the
development of ontology science and technology that we can pursue
experiments at a finer-grained level than just "my program gets it 52%
right and yours only gets it 49% right, so I got the superior approach"
kind of bake-off comparisons, which provide precious little information
on how individual systems might be modified to perform the way we
really want them to. (09)
The approach of trying to find reusable ontology modules that can be
combined, which, as I interpret him, is what PatH suggests as one way
to go via the Web, is, I believe, legitimate, but to advance quickly
needs more coordination than can be achieved by opportunistic
aggregates of user groups that may form from time to time. The modules
can only perform within a much larger program unit, and the
evolutionary benefits of comparing the utility of one module against
another one with the same function is mostly lost if the larger parts
of the systems being compared are very diverse. We can achieve a
higher degree of precision in evaluating the performance of individual
ontology modules if we had one - or a few - reference full ontology
systems that can take these test modules and make a comparison of a
full system with only one module at a time being different. That is
one of the functions that I hope a foundation ontology as a "conceptual
defining vocabulary" can perform, to provide independently developed
ontology modules a "home away from home" where they can be tested in a
systematic way, and provided to the larger community as aggregates of
modules whose combined performance is known, at least in some test
application. The foundation ontology should be as modular as possible
to permit testing of alternative functional units, to take advantage of
the evolutionary power of many contributors. But to do that, there
would also have to be some test programs, which in turn should be
modular for evolutionary capability. (010)
The first program I would want to build associated with the modular
foundation ontology is the language understanding and generation
application that only has a vocabulary somewhere in the range between
the Longman linguistic defining vocabulary (2000 words) and the
"receptive vocabulary" of a 5-year old (about 5000 words). The purpose
of such a language unit would be solely (initially) to allow users to
inquire about the content of the ontology in simple ordinary English
(perhaps only a little more expressive than CLCE), and to add new
knowledge and define new concepts (to be compatible with the ontology)
using the defining vocabulary. But there should also be other diverse
test programs that will allow different modules to be compared for
different purposes. (011)
Regardless of the different approaches that we want to take, I think it
would be useful to coalesce an ontology community with more structure
than just the Ontolog forum - Perhaps a society? So that we can
sponsor at least one on-line journal. PatH suggests a Wiki. I think
that a new Wiki devoted specifically to topics of ontology would be a
good thing - at a minimum to provide summaries of the threads that
occur in this forum. But I think that a more permanent refereed
journal, with articles in standard doc or pdf format, modifiable only
by the authors, would also be a good way to allow members to present
their better-organized thoughts and results, freely available, stable,
and quotable. To make it stable, I think that the costs of maintenance
of the site would have to be deferred by society membership fees, as I
don't think we can take advantage of our host's generosity
indefinitely. Any thoughts? (012)
Pat (013)
Patrick Cassidy
CNTR-MITRE
260 Industrial Way West
Eatontown NJ 07724
Eatontown: 732-578-6340
Cell: 908-565-4053
pcassidy@xxxxxxxxx (014)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (015)
|