Re: The "Minsky bottleneck" ...and a solution?
fritz@rodin.wustl.edu (Fritz Lehmann)
Date: Tue, 19 Jul 94 02:48:54 CDT
From: fritz@rodin.wustl.edu (Fritz Lehmann)
Message-id: <9407190748.AA03467@rodin.wustl.edu>
Newsgroups: comp.ai,sci.logic
Subject: Re: The "Minsky bottleneck" ...and a solution?
References: <30ci0m$6hl@Mercury.mcs.com> <kudzu.15.00154BFF@dnai.com> <30d3rf$bh@mercury.mcs.com>
Organization: Center for Optimization and Semantic Control, Washington University
Apparently-To: pratt@cs.stanford.edu
Apparently-To: renate.schmidt@mpi-sb.mpg.de
Apparently-To: cbrink@ucthpx.uct.ac.za
Apparently-To: srkb@cs.umbc.edu
Apparently-To: interlingua@isi.edu
Apparently-To: cg@cs.umn.edu
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk
[cc:cg@cs.umn.edu,interlingua@isi.edu,srkb@cs.umbc.edu,
cbrink@ucthpx.uct.ac.za,renate.schmidt@mpi-sb.mpg.de,pratt@cs.stanford.edu]
In the Usenet newsgroup comp.ai Jorn Barger wrote:
-----begin 1st quote-----
[. . .]
So, I have to guess, in CYC's knowledgebase there's probably a "human-
being object" that, thruout the last decade, has been needing to have
several new slots added *per day*, no doubt making Doug Lenat tear his
hair in frustration...
The way around this, I think, is to *sort* the slot/relationships, and
the way to sort them, I think, is according to the *types* of their
fillers.
-----end quote-----
In article <30d3rf$bh@mercury.mcs.com>, Jorn Barger <jorn@MCS.COM> wrote:
>Michael Sierchio <kudzu@dnai.com> wrote:
>>The way around this is a hierarchy of slots -- the details of phenomena and
>>roles become subsumed in the hierarchy, because it is the detail, or "surface
>>area", that would grow without bound otherwise.
>
>Yes... the sorting I'm suggesting would be hierarchical. But do you
>think there's any better way to sort slots into a hierarchy, than by
>the types of their fillers?
>
>j
>
Unless he has amnesia, I think Doug Lenat knows all
about sorting of slots. With Russ Greiner he co-wrote
the great RLL paper: "RLL: A Representation Language
Language", R. Greiner & D. Lenat, Proc. 1st AAAI Conference,
1980 p. 165. RLL had slots, slots of slots, facets of
facets, etc. I don't know how well CYC realizes the
"RLL Dream"; I think it came from Lenat's work in AM/Eurisko.
Sorting relations by the sorts of their relata (fillers)
is _not_ enough. There are hierarchies of relations,
ordered by subsumption, even on domains of uniform
sort (type). Algebraically these are represented by
Boolean inclusion within locally finite cylindric set
algebras. Such relational subsumption can be
_combined_ with subsumption of monadic sorts (concepts);
see "Subsumption Computed Algebraically" by Chris
Brink and Renate Schmidt in "Semantic Networks in
Artificial Intelligence", F. Lehmann, Ed., Pergamon
Press, Oxford, 1992, or the journal Computers & Mathematics
with Applications, v. 23 no.s 2-9, 1992. This was
also provided for in some KL-ONE languages, such as
NIKL and N-ary KANDOR. There is (or ought to be)
similar relational and conceptual subsumption for
higher-order relations and concepts (e.g. relations
between relations) as well as for "mixed order" relations
(e.g. the relations between a relation and its own
arguments). The first-order part has been done in
the "Peirce Algebras" of Brink, Britz, Bottner and
Schmidt (Tech. Rep. MPI-I-92-229, Max Planck Institut
fuer Informatik, Stadtwald, Saarbruecken, Germany).
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
=============================================================