Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!ix.netcom.com!netcom.com!whitten
From: whitten@netcom.com (David Whitten)
Subject: Re: A question about Cyc
Message-ID: <whittenDC0zI8.8A@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
X-Newsreader: TIN [version 1.2 PL1]
References: <3u1dc3$g41@Mercury.mcs.com> <whittenDBoJ74.4qz@netcom.com> <3u65e7$160@Mars.mcs.com>
Date: Thu, 20 Jul 1995 17:34:56 GMT
Lines: 80
Sender: whitten@netcom5.netcom.com

Jorn Barger (jorn@MCS.COM) wrote:
: David Whitten <whitten@netcom.com> wrote:
: >If the database doesn't know the distinguishing fact that Drink is
: >EdibleStuff and Food, but must only be liquid, his fix of :
: >  (#%MutuallyDisjointWith #%Bread #%Drink)
: >
: >shouldn't work. (which it didn't.)
: >
: >All he is saying there is that Bread is a term that extensionally does not
: >have the same elements as Drink does (ie: an instance cannot be both a bread
: >and a drink)  He isn't saying anything about if Bread can be generalized
: >into a Drink.

: Are there cases, then, where A is a generalization of B, *and* A is
: mutually disjoint with B???  These seem contradictory...

I think part of what is complicating this discussion is that the CYC
designers have tried to tighten some of these english words' definitions
a lot more than we would.  Class A is a generalization of Class B
only holds between named Collections, not between the elements of those
Collections.  Mutually Disjoint only holds between the elements of
the Class A and Class B (The actual instancesOf the Collections)

I would expect a Collection like FrostedFoods to include Ice Cream and
hence the instance #%Ice_Cream_139. But does this instance no longer belong
in the group Ice_Cream when it melts ?  I argue that it does.
but the definition for Drinks needs a liquid, and Melted IceCream is a 
liquid.... (Of course this isn't the Bread-liquid example but close...)

: And how could Drink have gotten into the knowledgebase without mention
: of liquid???
: >But if they haven't included all the axioms that are obvious yet,
: >how can they derive the specializations?  I think that is the point of
: >the CYC effort to begin with.
: Well it starts to sound like the knowledge-enterers are putting in
: special cases, and then Lenat (or someone) is going thru and replacing
: these with the generalizations?  (It's a strategy, but it's not optimal! ;^/

I think there is some automated generalization demons running which propose
possible generalizations that have to be ok'ed by a knowledge enterer.

: I wonder, too, if this isn't an example of predicate calculus being
: too artificially constrained for realworld reasoning, demanding
: details to be explicitly spelled out (bread is not a drink) that it
: ought to be able to reasonably *guess*...

: Why can't you enter the 'food story':
: person eats food to satisfy hunger motive
: (which will actually just be an indexed node in a fractal thicket)
: and the specialization:
: person drinks liquid-food to satisfy hunger/thirst motive
: (another indexed node)

This story doesn't really handle the melted icecream case (nor does
it claim to) but I know I still feel thirsty after drinking a melted
icecream.

: and Cyc sees that bread is food, and doesn't see that bread is
: liquid, so it presumes that bread is not a drink???

: Does Cyc have to be explicitly told that *liquids and solids* are
: mutually disjoint?  I'd think this sort of 'fact' can be
: embedded in a more efficient data structure than an assertion in
: predicate calculus...

I think EVERYTHING is embedded as an assertion in the CYCL
which is an Augmented Predicate Calculus (with equality)
This is then converted into an efficient data structure to use
while reasoning.  (CYC keeps both)

: But I think, from what you're saying, that the problem here was just a
: puzzling oversight in the definition of 'drink'-- somehow nobody who
: mentioned drinks included that drinks are liquid...?
: j

I'm not sure, Guha didn't give any information about what did happen then.
Since the CYC KB is constantly changing, it might even be hard to determine
what was going on then after the fact.

David (Whitten@netcom.com) (214) 437-5255
