Wrong level of abstraction in example
fritz@rodin.wustl.edu (Fritz Lehmann)
Date: Tue, 13 Sep 94 23:49:04 CDT
From: fritz@rodin.wustl.edu (Fritz Lehmann)
Message-id: <9409140449.AA14303@rodin.wustl.edu>
To: cg@cs.umn.edu, pdoudna@aol.com, srkb@cs.umbc.edu, wisdom@mcs.com
Subject: Wrong level of abstraction in example
Cc: kennett@u.washington.edu
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk
Rob Kremer <kremer@cpsc.ucalgary.ca>, writing on my
"EnclBarr" barrier example in my recent CCAT-NEWS posting,
said:
-----begin quote----
Subject: Re: CCAT Ontologies NEWS, Format Example
Date: Tue, 13 Sep 1994 11:42:40 -0600
I was quite impressed with Fritz's format. I would like to see
a SHORT ENGLISH DESCRIPTION: in addition to the LONG LEGALISTIC
ENGLISH DESCRIPTION:. The SHORT ENGLISH DESCRIPTION:'s are what
I would like to brouse when selecting a concept/relation from the
catalogue.
I also have a "pet peeve" meta-comment: The context of the example
is clearly "all agents are human". I object! We should take
_great_ care in specifying roles as high up the hierarchy as
possible. In the example the reference to HUMAN should all be
replaced with ANIMATE, to make explicit that a EnclBarr can be
a barrier to a rat or a nanite (small robot) too.
Given the above, the relativity of the barrier to the agent type
leads to a lot of additional complexity. A barrier to a rat may
not be sufficiently high to be a barrier to a human, but a barrier
to a human, while sufficiently high, does not act as a barrier to
a rat if it has rat-sized holes in it. This can be worked out in
the context of the Size-of-person<animate!>-blocked role. However
even more complex is cases like the agent being a (flying) bird.
This has some serious effects on the LONG LEGALISTIC ENGLISH
DESCRIPTION:.
All this is very hard to handle, if you ignore it, you tend to end
up defining Enclosing-Barrier-to-Humans, Enclosing-Barrier-to-Rats,
Enclosing-Barrier-to-Nanites, Enclosing-Barrier-to-Birds, ...,
which is also unpalatable.
-----end quote--------
I completely agree that such notions as barrier
should be defined for abstract agents rather than just
for humans. This was just an example intended to
illustrate the importance of painstaking, tedious legalistic
descriptions and far-reaching use of examples and borderline
cases for each defined concept. Generally useful concepts
should be defined at the highest possible level, I agree.
But there is a complication, and that is metaphoric use.
"Barrier" is not a bad example of this problem; while on the
one hand we should define it at a high level (to account for
robots, mice, bacteria etc.), I don't think that means that
the definition should be so rarefied and abstract as to
cover all the likely or possible metaphorical uses for a
word. The name of the "Save the Rodents Act" is a barrier to
its passage in the legislature. For some, acne is a barrier
to love. Postmodernism is a barrier to intellectual progress.
I feel that there should be a separate and heavyweight theory
of metaphor, without requiring the conceptual lexicon definition
to capture every feasible metaphoric use.
Also, I agree absolutely that the short natural-language
description is needed for CCAT browsing. Let's add that
field to the CCAT concept form.
Incidentally, I forgot to put in a field for the
"canonical basis" (type restrictions on arguments) of a
relation. That should be in there too.
Yours truly, Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-733-0566 Fax:(714)-733-0506 fritz@rodin.wustl.edu
=============================================================