Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 06 No. 45
NL-KR Digest (Thu Nov 30 09:54:34 1989) Volume 6 No. 45
Today's Topics:
ATMS Implementation Query (Long)
AI Seminar
Finding Spatial Relations in the World ... (Unisys AI seminar)
CSLI Calendar, 30 November, vol. 5:10
Submissions: nl-kr@cs.rpi.edu
Requests, policy: nl-kr-request@cs.rpi.edu
Back issues are available from host archive.cs.rpi.edu [128.213.1.10] in
the files nl-kr/Vxx/Nyy (ie nl-kr/V01/N01 for V1#1), mail requests will
not be promptly satisfied. If you can't reach `cs.rpi.edu' you may want
to use `turing.cs.rpi.edu' instead.
BITNET subscribers: we now have a LISTSERVer for nl-kr.
You may send submissions to NL-KR@RPIECS
and any listserv-style administrative requests to LISTSERV@RPIECS.
[[ The Symbol Grounding Problem has once again become a hot topic
on comp.ai, and it does fall under the topics this list covers.
Because I have no way of objectively filtering the
discussion, I decided to include only the first two articles (which
were in the last issue) and point people interested in reading more
to comp.ai. For those who do not have USENET access, I am keeping
all the articles in that discussion. It is currently 52K of text and
still raging. The file is in nl-kr/sgp on the archive server. -CW ]]
-----------------------------------------------------------------
To: nl-kr@cs.rpi.edu
From: sce!sce.carleton.ca!bond@watmath.waterloo.edu (Greg Bond)
Newsgroups: comp.ai,comp.ai.nlang-know-rep
Subject: ATMS Implementation Query (Long)
Date: 29 Nov 89 01:05:25 GMT
I'm in the midst of designing an ATMS and have run into difficulties
surmising implementation details of constraint consumers from deKleer's
third paper (Problem Solving with the ATMS). The explanation of my problem
is rather involved but I would appreciate any help I can get (in the form of
advice or pointers to papers, tech reports or books) because any solution to
the problem I can think of is rather kludgy - kludges do not belong in as
elegant a mechanism as the ATMS!
My difficulties can be illustrated with a simple example: suppose I wish to
add a constraint consumer for the problem solver constraint x=y+z. Using
deKleer's notation, this amounts to a set of conjunctive class consumers y,z
|-> {x=y+z}; x,y |-> {z=x-y}; x,z |-> {y=x-z} with a type C. Given that this
constraint consumer has been installed in the ATMS and the ATMS database is
initially empty (no nodes or justifications asserted), the following
scenario unfolds: first, a premise node with datum y=2 is asserted by the
problem solver - the returned node and its datum are entered in a problem
solver lookup table to prevent the assertion of multiple ATMS nodes with the
same datum; then a premise node with datum z=3 is asserted and its node is
entered in the lookup table. At this point the conjunctive class consumer
y,z |-> {x=y+z} is enqueued (by a problem solver-supplied enqueue procedure)
for execution (by a consumer execution procedure internal to the ATMS).
When executed, the consumer derives the datum x=5 and asserts a
corresponding node in the ATMS. The datum and its node are entered in the
problem solver lookup table and the node is returned to the consumer
execution procedure where a justification for the node is asserted - in this
case, the premises with datums y=2, z=3 are the antecedents and the node
with datum x=5 is the consequent. The justification informant indicates the
consequent being a result of a consumer of type C.
Herein lie the problems. At first, I thought that conjunctive class
consumers should be checked for applicability when a node is asserted. Thus,
when a node is asserted through the standard ATMS add-node interface
procedure, any conjunctive class consumers with domain classes corresponding
to that of the newly asserted node and any existing nodes in the ATMS
database are applicable. If the node conjuncts hold in consistent
environments then the consumer is enqueued. This technique works fine for
triggering the consumer in the given example: the node for y=2 belongs to
variable class y and the node for z=3 belongs to class z - conjunctive class
consumer y,z |-> {x=y+z} is therefore applicable.
However, when the consumer asserts the newly derived node for x=5 the
remaining two conjunctive class consumers will be found to be applicable.
Since the node label for x=5 is empty until it is justified the applicable
consumers cannot be enqueued. Instead, two dummy nodes are asserted with the
respective consumers attached to them. The dummy nodes are then justified
by the respective antecedent nodes: x=5,y=2 justifying the dummy node with
consumer x,y |-> {z=x-y} attached and x=5,z=3 justifying the dummy node with
consumer x,z |-> {y=x-z} attached. When the node for x=5 is justified by
the ATMS consumer execution procedure, both of the dummy nodes' labels are
updated to non-empty but neither of the attached consumers are enqueued
because their type is the same as that justifying the antecedent node for
x=5.
The problem is that these dummy nodes and their justifications should never
have been added to the ATMS in the first place because their attached
consumers will never run. The ATMS is left with a couple of extra nodes (in
this small example) whose labels are updated along with the other node
labels for no purpose whatsoever. How is this exception handled gracefully?
Is my current interpretation of the ATMS mixed up or is it simply a matter
of passing extra parameters around (flags -> yuk!)?
- -----------------------------------------------------------------
Greg Bond -----> bond@sce.carleton.ca
Dept. of Systems and Computer Engineering, Carleton University
Ottawa, ON, Canada K1S 5B6
------------------------------
To: nl-kr@cs.rpi.edu
Date: Tue, 28 Nov 89 14:45 EDT
From: MMETEER@rcca.bbn.com
Subject: AI Seminar
BBN STC Science Development Program
AI Seminar Series Lecture
SEMANTIC INTERPRETATION AND THE LEXICON:
WHAT MAKES SENSE?
Paul S. Jacobs
AI Program, GE Research
Schenectady, NY 12301 USA
jacobs@crd.ge.com
BBN STC, 2nd Floor Conference Room
10 Moulton Street, Cambridge, Massachusetts
Tuesday, December 5, 10:30 am
Practical applications of natural language demand precision in
semantic interpretation, highlighting the problems of lexical ambiguity
and vagueness. The representation and discrimination of word meanings
is thus a key issue for language analysis, motivated especially by the
need for broad scale NL systems and by applications in information retrieval.
A successful method for distinguishing word senses, however coarsely,
could be a major contribution to natural language processing technology.
Past research does not point to a successful strategy for sense
discrimination, but it does reveal some naive approaches that won't work.
The most obvious of these is the simple search for intersections or
``lexical coherence'' among word sense categories. This twenty-year-old
approach is still popular and still destined to fail. Sense discrimination
depends on context, and context is more than the combination of the words
that appear together. Context comprises topic analysis, phrasal constructs,
complex events, and linguistic and conceptual structures. This research
focuses on accessing the power of these more complex contextual structures
in identifying word senses using a lexicon of over 10,000 roots. Semantic
and syntactic preferences, lexical relations, and other structural knowledge
combine in our approach to help with generic sense discrimination.
------------------------------
To: nl-kr@cs.rpi.edu
Date: Wed, 29 Nov 89 11:08:49 -0500
From: finin@PRC.Unisys.COM
Subject: Finding Spatial Relations in the World ... (Unisys AI seminar)
AI SEMINAR
UNISYS PAOLI RESEARCH CENTER
Finding Spatial Relations in the World
to Match our Prepositions
Annette Herskovits
Wellesley College and University of Pennsylvania
Lexical meanings are notoriously difficult to define. For any definition,
there seems to exist a counterexample. Focusing on spatial prepositions
("across", "at", "over", etc.), I will propose systematic, but quite
complex, interactions between word meaning and cognition to account for the
broad range of uses of a word.
I will assume a geometric schema associated with each spatial preposition,
but also active processes of fitting the schema onto real situations. Two
phenomena account for the flexibility of lexical use: first, the fitting
takes advantage of selections, groupings, idealizations, and tranformations
which are part and parcel of spatial cognition (rather than strictly
linguistic processes); second, approximate fits are acceptable, subject to
well-defined conditions.
However, this search for a best fitting schema cannot explain all
prepositional uses. In addition, there are some standard types of
situation, defined as functional interactions rather than strictly
spatially, for which the use of a particular preposition is required,
either by convention or because of salience.
I will discuss the consequences of this analysis for linguistics and
artificial intelligence.
11:00 pm Tuesday, December 5, 1989
BIC Conference Room
Unisys Paoli Research Center
Route 252 and Central Ave.
Paoli PA 19311
-- non-Unisys visitors who are interested in attending should --
-- send email to finin@prc.unisys.com or call 215-648-7446 --
------------------------------
To: nl-kr@cs.rpi.edu
Date: Wed, 29 Nov 89 13:43:04 PST
From: ingrid@russell.Stanford.EDU (Ingrid Deiwiks)
Subject: CSLI Calendar, 30 November, vol. 5:10
C S L I C A L E N D A R O F P U B L I C E V E N T S
- ----------------------------------------------------------------------------
30 November 1989 Stanford Vol. 5, No. 10
_____________________________________________________________________________
A weekly publication of the Center for the Study of Language and
Information (CSLI), Ventura Hall, Stanford University, Stanford, CA 94305-4115
____________
CSLI ACTIVITIES FOR THIS THURSDAY, 30 NOVEMBER 1989
12:00 noon TINLunch
Cordura 100 Stanford's Policies on Intellectual Property Rights
Jon Sandelin and David Charron, Licensing Associates
Jane McLean, Manager of the Software Distribution
Center (SDC)
Office of Technology Licensing (OTL)
Stanford University
(sandelin@popserver.stanford.edu)
Abstract in last Calendar
2:15 p.m. CSLI Seminar
Cordura 100 Models of Rational Agency 9
Michael Bratman, Martha Pollack, Stan Rosenschein
(bratman@csli.stanford.edu,
pollack@warbucks.ai.sri.com, stan@teleos.com)
Wrap-up session
Abstract in last Calendar
____________
ANNOUNCEMENT
Because of final exams and the winter break, there will be no Thursday
events and no Calendar on 7, 14, 21, and 28 December. The next
Calendar will be published on 4 January, and Thursday events will
resume on 11 January.
____________
SYMBOLIC SYSTEMS FORUM
Systematizing Our Commonsense Reasoning about Fictions
Ed Zalta
Friday, 1 December, 3:15, 60-61G
Consider this piece of reasoning:
(1) The ancient Greeks worshipped Dionysus.
(2) Dionysus is a mythical character.
(3) Mythical characters don't exist.
(4) Therefore, the ancient Greeks worshipped something that doesn't
exist.
Even though we are tempted to say that the Greeks certainly believed
that they were worshipping something that exists, if (2) and (3) are
true, what they were worshipping doesn't exist. How, then, do you
represent (4) using the resources of ordinary logic? How can you say
that there is something which doesn't exist and which the Greeks
worshipped? Haven't you contradicted yourself?
Moreover, it follows from the fact that Ponce de Leon searched for the
fountain of youth that Ponce de Leon searched for something. What is
the thing that makes "Ponce de Leon searched for something" true?
These are questions addressed in Friday's talk.
This talk is aimed at undergraduates with no assumed background.
Refreshments will be served.
____________
PHILOSOPHY DEPARTMENT COLLOQUIUM
Warranted Assertibility and Truth: Dewey's Reply to Russell
Tom Burke
Department of Philosophy
(burke@csli.stanford.edu)
Friday, 1 December, 3:15, 90-91A
Russell (1940) wants to say that a proposition expressed by the
formula `P(a)' is true just in case the object a denoted by `a' does
in fact have the property P denoted by `P'. True propositions simply
report facts. This assumes that a and P occur in the actual scheme of
things, hence the only real question as to the truth-value of the
proposition is whether or not a does in fact have property P. Russell
admits that this "involves us in metaphysics, and has difficulties
(not insuperable) in defining the correspondence which it requires for
the definition of `truth'."
Dewey assumes virtually nothing about "_the_ actual scheme of things,"
in which case a simple appeal to facts is pointless. Dewey
relativizes the propositional contents of statements to schemes of
individuation I brought to bear by agents in given situations s (to
use some current terminology). The proposition that a is of kind P
(modulo s,I) is true just in case (1) a and P belong to the
individuation scheme I brought to bear in situation s, and (2) the
activities perform _able_ in that situation to individuate object a
_would in the ideal limit_ reliably indicate that it meets the
specifications for being of kind P (whether or not those activities
are actually performed).
I will discuss several examples to motivate this pragmatist definition
of truth.
____________
COMMONSENSE AND NONMONOTONIC REASONING SEMINAR
A New Approach to Modal Operators
Matt Ginsberg
Department of Computer Science
Monday, 4 December, 3:15
Margaret Jacks Hall 252
We describe a new formalization of modal operators that views them not
in terms of Kripke's possible worlds, but as functions on an
underlying set of truth-values. Thus Moore's knowledge operator L,
where Lp means "I know that p," would correspond to a mapping taking
true into true (since we know p if we know it to be true) and taking
both false and unknown into false (since we do not know p if we either
know it to be false or know nothing about it at all). This new
approach has the following advantages over the conventional ones:
(1) intuitive simplicity;
(2) it provably generalizes both Kripke's construction and Moore's
autoepistemic logic, while making clear the distinctions between
them;
(3) it allows for easy further generalization to modal operators
that are related to temporal reasoning and to causality;
(4) the natural procedure for computing the truth-value of a sentence
involving these modal operators is "incremental" in the sense that
it computes approximate answers that "converge" to the correct one
in the large runtime limit.
This talk will concentrate on the first two of these properties; I
will discuss the third and fourth as completely as time allows.
____________
SPECIAL TALK
Nonwellfounded Sets and Their Applications
Jon Barwise
Monday, 11 December, 4:00
SRI International, Building A, Conference Room B
Nonwellfounded sets (for example, a stream of the form a = (1,(2,a)),
with the usual definition of ordered pair) were part of set theory in
the good old days, but were later banned since they were felt to be
implicated in the paradoxes. However, in recent years, they have been
slowly finding their way back into set theory, due in part to the work
of Peter Aczel (among others), who has shown that they are not
incoherent, and in part to applications that have been found for them
in modeling various kinds of circular phenomena in computer science,
AI, philosophy, and cognitive science. This talk will motivate
nonwellfounded sets by focusing on simple applications in computer
science. I will then give a summary of Aczel's work, and show how it
yields the desired applications. The talk is expository and should be
accessible to anyone familiar with basic set theory.
- --
Note for Visitors to SRI:
Please arrive at least ten minutes early in order to sign in and be
shown to the conference room.
SRI is located at 333 Ravenswood Avenue in Menlo Park. Visitors may
park in the visitors lot in front of Building A (red brick building at
333 Ravenswood Avenue, second driveway on the right, east of Laurel)
or in the conference parking area at the corner of Ravenswood and
Middlefield. The seminar room is in building A. Visitors should sign
in at the reception desk in the building A lobby.
IMPORTANT: Attendance is open, but visitors from certain countries
designated by the U.S. government must make arrangements in advance.
If you have not already made such arrangements before your arrival,
admission to the seminar will be denied. If you believe you may be
from one of these countries and if you wish to make arrangements to
attend, please call Judith Burgess at (415) 859-5924.
____________
CURRENT VISITORS AT CSLI
PETER AUSTIN, Senior Lecturer in Linguistics and Japanese Language
Coordinator, La Trobe University, Melbourne, Australia. Peter's main
research areas are syntax, semantics, and Australian Aboriginal
languages. He has published on morphosyntax, case marking, and
complex sentence constructions (including switch-reference) in
Australian languages. His most recent publication is _Complex Sentence
Constructions in Australian Aboriginal Languages_ (Benjamins, 1988).
Dates of visit: November 1989-January 1990.
SUK-JIN CHANG, Department of Linguistics, Seoul National University.
During his stay at CSLI, Suk-Jin will continue to work on developing
an information-based Korean discourse grammar (IKDG) that has been
conceived of and explored to some extent from the perspective of
natural-language processing and in the general theoretical framework
of relational theories of language as action. Specifically, he will
attempt to incorporate into IKDS the system of honorification, viewed
as a pragmatic agreement phenomenon between the discourse participants
and the information functions of topic and focus, distinct from
grammatical functions of subject and object, by keeping abreast of
ongoing studies in situation theory and situation semantics and
extending the semantic component of HPSG. Dates of visit: August
1989-July 1990.
HIROSHI KATO, Industrial Affiliates Program visiting researcher, C&C
Information Technology Laboratories, NEC Corporation, Japan. Since
1983, Hiroshi has been engaged in research and development of
educational systems, such as Computer-Based Instruction (CBI) systems
and Computer-Managed Instruction (CMI) systems. He is interested in
cognitive models of human learning and human interface. He is now
working on a research project with James Greeno on a learner's
understanding model of first-order logic through the CBI system
"Tarski's World." Dates of visit: August 1989-August 1990.
FINN KENSING, Associate Professor, Department of Computer Science,
Roskilde University Centre, Denmark. Finn is a coauther of
_Professional Systems Development: Experience, Ideas and Action_. The
book will be published by Prentice Hall this fall. While at CSLI,
Finn will be doing an inquiry into what can be learned from applying a
language/action approach on the design of systems to be used in human
work processes. He will be working closely with Terry Winograd.
Dates of visit: August 1989-February 1990.
BERNARD LINSKY, Associate Professor of Philosophy, University of
Alberta, Canada. Bernie is on sabbatical from Edmonton for the
academic year. He is here to learn about situation semantics and to
work on papers on the metaphysics of semantics, in particular, on
possible worlds and universals. Dates of visit: September 1989-August
1990.
HIDEO MIYOSHI, Industrial Affiliates Program visiting researcher,
Sharp Corporation, Japan. From 1982 to 1987, Hideo was involved in
natural-language processing as a researcher of ICOT (Japanese fifth
generation computer project), where he worked on the development of
BUP (bottom-up parser in Prolog), DUALS (an experimental
discourse-understanding system), and JPSG (Japanese Phrase-Structure
Grammar). At SHARP, he recently worked on the development of (1) a
text-retrieval system using flexible keywords, and (2) a support
system for generating controlled Japanese texts. Both projects were
sponsored by ICOT. Hideo is interested in the semantic analyses and
representations of the Japanese language. He is also interested in
the role and mechanism of knowledge in natural-language understanding.
He hopes that the STASS project will help him understand these better.
While at CSLI, he will be mainly involved in the STREP project. Dates
of visit: October 1989-October 1990.
STEPHEN NEALE, Assistant Professor of Philosophy, Princeton
University. Stephen is working on a new book on the interpretation of
plural noun phrases in natural language, and -- when time permits -- a
manuscript on free will. Dates of visit: May 1989-February 1990.
MASAYUKI NUMAO, Tokyo Institute of Technology. Masa's research topic
is: Learning in Situations. During the last few years, he's been
interested in "learning," especially program synthesis by using the
techniques of machine learning. Generally speaking, program synthesis
systems have only synthesized pure Lisp or pure Prolog programs that
do not cause "side effects," since researchers seek "general synthesis
schemes." In reality, procedural programs depend completely on each
situation, and program statements may depend on each other
unexpectedly, causing "frame problems." Masa would like to overcome
these difficulties by "analytic learning of situations."
The steps are as follows:
(1) Learning -- An abstracted environment is extracted by analyzing
the situation based on a given PASCAL interpreter, and memorized.
(2) Program Synthesis -- In similar situations, slightly modified
programs are synthesized.
Masa hopes that each situation can be given in natural language as
given in PASCAL tutorial books, but as the first step he will give it
directly. He thinks "learning" is a key factor when dealing with
"situatedness." Dates of visit: September 1989-August 1990.
GREG O'HAIR, Department of Philosophy, The Flinders University
of South Australia. Dates of visit: October 1989-June 1990.
RYO OCHITANI, Industrial Affiliates Program visiting researcher,
Fujitsu Laboratories Ltd., Japan. For the past five years, Ryo has
helped develop the machine-translation system ATLAS2 at Fujitsu
Laboratories. During the last year, he and several other people
started a new study on a quicker and easier system to classify
sentences based on the semantic view, for which they collected two
million Japanese text samples and classified several hundred sentences
manually. Ryo is interested in studying what kind of system would
utilize this information most efficiently. Dates of visit: April
1989-April 1990.
ICHIRO OHSAWA, Tokyo Institute of Technology, Japan. Ichiro's
interest is modeling on computers of intelligent and communicative
agents who can handle natural language. The two key facts about such
agents are perspectivity and situation-boundedness. That is, they
view the world and themselves from their points of view and the
information they get is given only in the situation they are in. His
current research has been aimed at a computational realization of an
agent who can interact in dialog with his own viewpoint. He looks
forward to active interactions with human intelligent agents at CSLI.
Dates of visit: September 1989-February 1990.
ELIN ROENBY PEDERSEN, Assistant Professor of Computer Science,
Department of Computer and Systems Sciences, Copenhagen School of
Business and Administration, Denmark. During her stay, Elin will
carry out a number of studies on the use of advanced work stations in
intellectual work, e.g., in system analysis and development and in
research. The aim of these studies is to get a firmer grip on the
dynamical influence of computerized tools on processes in which people
attempt to build and convey knowledge. One hypothesis is that the use
of computerized tools should be studied as a special case of the use
of strictly defined means for description, i.e., as an issue of the
psychology and philosophy of description. Elin will be working
closely with Terry Winograd and Lucy Suchman. Dates of visit:
June-December 1989.
ANDRE SCEDROV, University of Pennsylvania. For the past several
years, Andre has been working in Logical Foundations of Programming
Structures, an area that bears on logic, theoretical computer science,
and algebra and topology in mathematics. His most recent work in this
area is concerned with incorporating inheritance and other features of
object-oriented programming within type systems with static
type-checking. He is now working on refinements of type systems that
would express computational resource requirements as program
specifications. In this setting the compliance of a program with the
imposed resource bounds would be insured at compile-time. Dates of
visit: July 1989-August 1990.
HINRICH SCHUETZE, University of Stuttgart, Germany. Hinrich received
a Master's degree in Computer Science from Stuttgart University in
April 1989. His thesis deals with the treatment of plurals in
natural-language processing systems. At this time, semantics --
especially the semantics of plurals -- is his main interest. While at
CSLI, he will work on PROSIT (PROgramming in SItuation Theory), a new
programming language that has many features of situation theory built
in. He would like to extend PROSIT to a tool for the semantic
processing of natural-language input. Dates of visit: July
1989-August 1990.
HIDETOSI SIRAI, Associate Professor, Department of Psychology, Chukyo
University, Japan. Hidetosi's background is in computer science. He
has been working on Japanese Phrase-Structure Grammar (JPSG) and is
interested in Situation Semantics (SS) to describe semantic phenomena
in Japanese and English. During his stay at CSLI, he will work on
JPSG + SS + ? = NLU (Natural-Language Understanding). Dates of visit:
July 1989-February 1990.
BARBARA TVERSKY, Department of Psychology, Stanford University.
Barbara is on sabbatical this year. Dates of visit: AY 1989-90.
WILLIAM UZGALIS, Assistant Professor of Philosophy , Oregon State
University. Bill is interested in identity and individuation,
personal identity, and the debate between essentialists and
antiessentialists over such issues as the proper semantics for natural
kind terms. He works on issues in the philosophies of Locke and
Plato. Dates of visit: July 1988-August 1990.
------------------------------
End of NL-KR Digest
*******************
