Re: Roles and firstname.lastname@example.org (Pat Hayes)
Content-Type: text/plain; charset="us-ascii"
Date: Thu, 28 Sep 1995 14:42:30 -0600
To: email@example.com (John F. Sowa), firstname.lastname@example.org, email@example.com,
From: firstname.lastname@example.org (Pat Hayes)
Subject: Re: Roles and dependence
Cc: email@example.com, firstname.lastname@example.org
At 7:43 AM 9/28/95 +0500, John F. Sowa wrote:
>As a starting point for defining the top level of an ontology, I find
>these six categories to be cleaner, simpler, and more comprehensive
>than anything else I've seen in the AI literature.
Well, I certainly don't (Eg. presumably numbers are in A1; what does it
mean to 'predicate' a number?)
> I certainly don't
>deny that there is an enormous amount of work necessary to bridge the gap
>between these half dozen very general categories and the thousands of
>very detailed types in natural languages and systems like Cyc. But as
>I have said in many notes to these lists, we need both top-down theoretical
>work and bottom-up empirical work. They complement one another.
Only if the theory somehow illuminates or gives insight into the empirical
data, as (for example) the 'eightfold way' did to particle physics. We can
argue endlessly about very general categories, never agree, and get
An example, from CYC as of a few years ago. I have always had a very clear
mental distinction between an individual thing and a mere patch of
space-time. CYC also had this, but the writers discovered after a lot of
work, and to their surprise, that the distinction seemed to be
unnecessary.(Doug, please comment on this in case I have it wrong.) This
was a real empirical result, on a par with CERN discovering only one
particle where theory has predicted two.
The top levels of our classification networks need to be be DISCOVERED, not
imposed from above by 'theoretical work'. None of this ontological
semiphilosophising is remotely precise enough to count as 'theory' relative
to any empirical discovery, in any case. As soon as one comes up with
slightly peculiar examples, this classification gets just as muddled as any
other a-priori top-down classification.
If your categories were the top level of Ontolongua, I would be forced into
constantly thinking about whether a timepoint (say) was in category P1 or
A1, or whether the relation of being-the-start-time-of-the-calendar should
be considered A3 or P2, and so on. But I find the description of P1 to be
incoherent, and P2 and P3 to be indistinguishable. The reference manual for
the representation language would refer me to papers written in 1885 for
further explanation. And the answer is, I don't give a damn how they are
classified in this arcane scheme, and I dont want to have to think about
it. I dont want to be forced into such a position.
My suggestion is that we simply put 'thing' at the top, and immediately
split it into the highest-level categories that make actual clear sense in
the domains we are describing. As these domains get to be more general, by
all means lets try to move the classifications upwards, but always in such
a way as to reflect the structures of the categories we have, rather than
to conform to any kind of pre-defined philosophical theory. We are in a
better position than most of the philosphers ever were to discover what the
REAL categories of human thought actually are.
>Comments like that are what incites flamage. .....
OK, I accept the rebuke. I was teasing you; sorry.
Beckman Institute (217)244 1616 office
405 North Mathews Avenue (415)855 9043 or (217)328 3947 home
Urbana, Il. 61801 (217)244 8371 fax