RE: Brainstorming Idea: Set up an online testbed
John,
I agree. The methodology we use is quite consistent with the general
software development cycle. It's just an expansion of that to include
ontology development.
The focus is on bottom-up analysis of any data sources whose semantics
you must capture (and that's hard because they typically have no easily
gleaned semantics: you must analyze schemas, typically physical
schemas, data dictionaries, and lean on the limited end user/developer
time you may have access to), combined with top-down analysis of
potentialities that today's isolated database users (e.g., database
integration as the focus; but this notion applies to other kinds of
applications, i.e., decision support, intelligence analysis, risk
assessment, etc.) cannot hope to achieve, and thereby develop use
cases, scenarios, competency questions.
This is indeed task-oriented.
However, the more times you go through the above process, and the more
times you link two communities with different ontologies, the more you
understand that commensurability/interoperability can best be solved by
common ontologies, i.e., embedding those domain ontologies in a common
ontology. If you don't, you end up creating integrative ontologies from
the bottom up. Multiply such efforts a thousand-fold and you end up
with incompatible integrative ontologies all over the place, a
willy-nilly progression upward toward middle and upper ontologies that,
had they been provided earlier in the process, would have shortened
your ontology integration/interoperability effort considerably, by at
least an order or two of magnitude. Unfortunately, once these bottom-up
conceptual stovepipes have been allowed to evolve on their own for any
extended length of time, it is very hard to rationalize them into
something that is coherent, consistent, and avoids huge duplication.
This is a hard-learned pragmatic approach. And I think it applies to or
is consistent with science and engineering more generally: you have to
always work both bottom-up and top-down simultaneously. Pure induction
from the bottom does not work well, though we might all prefer that.
You need simultaneous top-down theory. And it's actually more
complicated than that, since you need partial top-down to meet partial
bottom-up, again and again.
Thanks,
Leo
_____________________________________________
Dr. Leo Obrst The MITRE Corporation, Information Semantics
lobrst@mitre.org Information Discovery & Understanding, Command and
Control Center
Voice: 703-983-6770 7515 Colshire Drive, M/S H305
Fax: 703-983-1379 McLean, VA 22102-7508, USA
-----Original Message-----
From: John F. Sowa [mailto:sowa@bestweb.net]
Sent: Thursday, December 06, 2007 12:29 AM
To: Obrst, Leo J.
Cc: standard-upper-ontology@LISTSERV.IEEE.ORG
Subject: Re: Brainstorming Idea: Set up an online testbed
Leo,
I strongly support work on aligning ontologies and on developing
new ontological resources. But there are several very important
points about the way current systems (and people) interoperate.
And no one has ever suggested any reason why future systems should
do anything that is different in any substantial way:
1. The major ontological agreements required among interoperating
agents (human or computer) depend on the tasks they do.
2. The required ontology for a given task can be derived by
analyzing the messages sent among the agents that are working
together on that task. No global alignment is required for
any aspects of their ontologies outside the specific task.
3. A given agent may communicate with different agents using
different task-oriented ontologies for different purposes.
4. Any agent X that is involved in more than one task may have
a global ontology that encompasses all of the tasks. But
X need only align its ontology with other agents on a task-
by-task basis.
This kind of task-oriented approach is much easier to deal with
than any approach that depends on global alignment.
I'm not against global alignment, but I strongly recommend that
we start with the much more manageable task-oriented ontologies.
John