Re: SUO: OpenCyc Motion Open for Discussions
Hi John,
On Sat, 5 Apr 2003, John F. Sowa wrote:
> Cathy,
>
> The topic of integrating multiple ontologies has been
> discussed piecemeal over the past couple of years:
>
> > How would you like it to be integrated?
>
> I'll try to give a brief summary:
>
> 1. The major concern I have been repeating many
> times is that no single, monolithic ontology
> can ever be adequate for everything. No matter
> how big it may be, there will always be different
> perspectives, contexts, aspects, applications,
> etc., for which the given set of definitions and
> axioms is inappropriate, inadequate, or incomplete.
>
> 2. That issue has been a major theme that I stated
> in my 1984 book in Chapter 7, which had the title
> "Limits of Conceptualization" and in my 2000 book
> in Ch 6, with the title "Knowledge Soup". It's
> related to Wittgenstein's notion of family
> resemblance, to Waismann's notion of open texture,
> and to Peirce's notion that "meanings grow".
> In computational linguistics, there is a related
> slogan about syntax: "All grammars leak."
>
> 3. There has also been a great deal of ferment in
> linguistics about the inadequacy of any fixed number
> of word senses to capture the full range of meaning
> of any word. Alan Cruse, for example, has coined
> the term "microsense" to mean the particular sense
> of any word within a given context. Cruse maintains
> that there is a continuum of meanings of any word and
> that any particular list of "word senses" will always
> be incomplete. That is just one more way of repeating
> what Peirce, Wittgenstein, and Waismann were saying.
>
> 4. As an approach to accommodating such an open-ended
> range of meanings, I proposed the lattice of all possible
> theories in Ch. 7 of my KR book. I also have an article
> on the web that proposes that lattice as the conclusion:
>
> http://www.jfsowa.com/pubs/signproc.htm
Yes I read that, thanks! I seem to remember sending you feedback on the
list about it last year.
> The idea is that all possible axiomatizations are admitted
> into the (potentially infinite) lattice. Any meaning that
> anyone might ever define for any concept would be located
> somewhere in that lattice -- not as an isolated definition,
> but as part of a theory of related concepts on some topic.
>
> 5. One implication of the lattice is that the best way to
> organize an ontology is as a collection of theories, each
> of which is internally consistent, but possible inconsistent
> with many alternative ways of defining similar concepts.
Why isn't Cyc already doing this with microtheories (without the false
assertions and the tautologous top and the inconsistent bottom)?
The Cyc microtheories are organized into a partial ordering via implies,
which is how you were describing the lattice of theories in your message
to Radu.
I don't see how this fits, though, with your embrace above of
Wittgenstein's ideas about family resemblance and open texture. Don't
these ideas militate against the possibility of giving "any meaning that
anyone might ever define for any concept" a place in a set of "all
possible axiomatizations"? So, to take a concrete example, how would the
concept of a game be axiomatised in your lattice? Could you sketch this
for me? How about 'electron'?
> Both Cyc and SUMO (and many other ontologies) have been
> constructed from many diverse contributions. Putting them
> together into one big theory is OK, but it should recognized
> that there are many other ways of assembling the same components
> with other selections taken from that infinite lattice. No one
> selection can ever be ideal for all possible applications, and
> any particular selection can be, at best, a recommended version
> that is suitable for one project, but perhaps not for another.
>
> 6. The Information Flow Framework (IFF), which is already a candidate
> project for an IEEE standard, has adopted the lattice of all
> possible theories as one key idea, but they have embedded it within
> a much more elaborate framework based on category theory. That is
> much more than I had originally proposed, but it is a reasonable
> extension. IFF is a content-free framework that can accommodate
> any content whatever. OpenCyc and SUMO are two large collections
> of content. My recommendation is for the three groups to work
> together: IFF develops the framework, and the OpenCyc and SUMO
> groups populate the framework with their content.
>
> I realize that doing all the work of disentangling all the components
> that have gone into OpenCyc and SUMO and placing them within the IFF
> framework is nontrivial. I also realize that we all have "day jobs",
> and nobody is paying us to do anything along those lines. But I
> believe that it has to be done.
Since you have done so much thinking about the lattice of theories
issue and have such a clear vision of what needs to be done in this
regard, John, maybe you would be well placed to make a start on this?
> And I also believe that just proposing another collection of content,
> such as OpenCyc, without relating it to any other proposed collections
> of content is not going to address the real problems summarized in
> points #1, #2, #3, and #4 above.
I'm not sure about this sharp dichotomy between the IFF framework as
'content-free' and the Cyc framework as just 'another collection of
content'. Cyc has plenty of concepts as 'form-like' (as it were) as those
that I understand the IFF to be developing, e.g. #$Collection,
#$Thing....It seems to me that the fact that the distinction between form
and content is a matter of degree rather than what, say, Hume divided into
"relations between ideas" and "matters of fact" (and never the twain shall
meet) is one of the major insights of the Peircean pragmaticistic logical
realism you and I concur in admiring.
Cheers,
Cathy.
--------------------------------------------------------------------------
Cathy Legg, Phd Cycorp, Inc.
Ontologist 3721 Executive Center Dr., ste 100
www.cyc.com Austin, TX 78731-1615
download OpenCyc at http://www.opencyc.org
--------------------------------------------------------------------------