Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 01 No. 13
NL-KR Digest (10/09/86 09:58:36) Volume 1 Number 13
Today's Topics:
Request for References: Conjunction Parsing
Cognitive Science Colloquium
Linguistics Talk, by Professor M. Motley
New UCLA Cog Sci TRs
----------------------------------------------------------------------
Date: 07 Oct 86 12:57:36 +1000 (Tue)
From: Steven Bird <munnari!mulga.oz!steven@seismo.CSS.GOV>
Subject: Request for References: Conjunction Parsing
Treating coordination is well known to be one of the most difficult aspects
of parsing and parser writing because of the various forms of conjunction
reduction that accompany coordinated constructions. I am doing a survey of
conjunction parsing systems as part of my thesis work. Such systems either
have a meta-grammatical module that takes the job of handling conjunctions
away from the main parser, or they incorporate conjunction parsing into the
main parser. I have come across work on conjunction parsers by the following
researchers (listed alphabetically): Berwick, Boguraev, Dahl & McCord,
Fong & Berwick, Gazdar Klein Pullum & Sag, Huang, Lesmo & Torasso, Kosy,
Raze, Sager, Sedogbo, Snarr, Woods.
As I am hoping to make the survey as comprehensive as possible, I would like
to find out about any work that I have overlooked, including work in related
areas (such as treating ellipsis).
Any pointers to the literature, details about unpublished work, or general
comments, would be appreciated. Please mail me directly - I will post a
summary if there's enough interest. Thanks in advance...
[Steven Bird. Comp. Sci. Dept. Melbourne University, Parkville 3052, Australia]
[UUCP: {seismo,mcvax,ukc,ubc-vision,prlb2,enea,hplabs,tataelxsi}!munnari!steven]
[ARPA: steven%munnari.oz@seismo.css.gov CSNET: steven%munnari.oz@australia]
[JANET: steven%oz.munnari@ac.ukc PHONE: +613 344 5229 ]
------------------------------
Date: Tue, 30 Sep 86 09:49 EDT
From: William J. Rapaport <rapaport%buffalo.csnet@CSNET-RELAY.ARPA>
Subject: Cognitive Science Colloquium
UNIVERSITY AT BUFFALO
STATE UNIVERSITY OF NEW YORK
GRADUATE GROUP IN COGNITIVE SCIENCE
PRESENTS
HANS KAMP
Department of Philosophy and Center for Cognitive Science
University of Texas at Austin
SOME REMARKS ON THE REPRESENTATION OF PROPOSITIONAL ATTITUDES AND THEIR REPORTS
Many extant theories of belief sentences and other attitude reports have
failed to pay sufficient attention to the importance of internal connec-
tions that typically exist between the different attitudes of one and
the same person, and the influence these connections exert on the inten-
tional significance that can be attributed to each of those attitudes
individually. Ignoring this aspect of cognition not only leads to a
distorted understanding of it, it also prevents one from formulating
intuitively plausible solutions to many of the puzzles about attitude
reports that have preoccupied philosophers of language and mind in
recent years.
Discourse Representation Theory, which emphasizes the pervasively con-
textual nature of sentence interpretation (typically, the interpretation
of a sentence that occurs as part of a longer discourse is guided by the
interpretation already assigned to the preceding part of the discourse),
is naturally extended to a theory of attitudes in which their internal
connectedness is central, and explicitly articulated. I will outline
this theory and demonstrate its workings in connection with a few (rea-
sonably) familiar examples from the philosophical literature.
References:
(1) Asher, Nicholas, "Belief in Discourse Representation Theory,"
Journal of Philosophical Logic (1986).
(2) Kamp, Hans, "Context, Thought and Communication," Proceedings
of the Aristotelian Society (1984-85).
Monday, October 20, 1986
3:30 P.M.
Baldy 684, Amherst Campus
Co-sponsored by:
Buffalo Logic Colloquium and Department of Philosophy
-----
For further information, contact:
William J. Rapaport
Assistant Professor
Dept. of Computer Science, SUNY Buffalo, Buffalo, NY 14260
(716) 636-3193, 3180
uucp: ..!{allegra,decvax,watmath,rocksanne}!sunybcs!rapaport
csnet: rapaport@buffalo.csnet
bitnet: rapaport@sunybcs.bitnet
------------------------------
Date: Tue, 7 Oct 86 14:45:13 PDT
From: admin%cogsci.Berkeley.EDU@berkeley.edu (Cognitive Science Program)
Subject: Linguistics Talk, by Professor M. Motley
Professor Michael Motley, Professor and Chair, Department of
Rhetoric, UC Davis will give a talk on 14 October 1986, in
117 Dwinelle at 8:00 P.M. on:
"Slips of the Tongue as Clues to the Efficiency of Natural
Speech Production".
Abstract:
For almost a century, slips of the tongue have been studied
as a 'window on the mind.' Psychoanalysts, linguists, and
cognitive psychologists have disagreed, however, about what
we see when we peek into that window. By incorporating data
from both naturalistic slip collections and laboratory-
induced verbal slips into a 'spreading activiation' model of
lexical organization, we can arrive at a compromise perspec-
tive on verbal slips, per se, and also gain additional
insights to natural speech-production processing.
Sponsored by the Department of Linguistics. Leave a message
for John Ohala (ohala@cogsci) if you're interested in join-
ing the group that takes Prof. Motley to dinner before the
talk.
------------------------------
Date: Tue, 7 Oct 86 13:25:36 PDT
From: Judea Pearl <judea@LOCUS.UCLA.EDU>
Subject: New UCLA Cog Sci TRs
The following technical reports are now available from the
Cognitive Systems Laboratory
Room 4712, Boelter Hall
University of California
Los-Angeles, CA, 90024
or: judea@locus.ucla.edu
---------------------
Pearl, J., ``Bayes and Markov Networks: a Comparison of Two
Graphical Representations of Probabilistic Knowledge,'' Cognitive
Systems Laboratory Technical Report (R-46), September 1986.
ABSTRACT
This paper deals with the task of configuring effective graphical
representation for intervariable dependencies which are embedded
in a probabilistic model. It first uncovers the axiomatic basis
for the probabilistic relation `` x is independent of y , given
z ,'' and offers it as a formal definition for the qualitative
notion of informational dependency. Given an initial set of such
independence relationships, the axioms established permit us to
infer new independencies by non-numeric, logical manipulations.
Using this axiomatic basis, the paper determines those properties
of probabilistic models that can be captured by graphical
representations and compares the characteristics of two such
representations, Markov Networks and Bayes Networks. A Markov
network is an undirected graph where the links represent
symmetrical probabilistic dependencies, while a Bayes network is
a directed acyclic graph where the arrows represent causal
influences or object-property relationships. For each of these
two network types, we establish: 1) a formal semantic of the
dependencies portrayed by the networks, 2) an axiomatic
characterization of the class of dependencies capturable by the
network, 3) a method of constructing the network from either hard
data or expert judgments and 4) a summary of properties relevant
to its use as a knowledge representation scheme in inference
systems.
--------------------------------
Zukerman, I. & Pearl, J., ``Comprehension-Driven Generation of
Meta-Technical Utterances in Math Tutoring,'' UCLA Computer
Science Department Technical Report CSD-860097 (R-61).
ABSTRACT
A technical discussion often contains conversational expressions
like ``however,'' ``as I have stated before,'' ``next,'' etc.
These expressions, denoted Meta-technical Utterances (MTUs) carry
important information which the listener uses to speed up the
comprehension process. In this research we model the meaning of
MTUs in terms of their anticipated effect on the listener
comprehension, and use these predictions to select MTUs and weave
them into a computer generated discourse. This paradigm was
implemented in a system called FIGMENT, which generates
commentaries on the solution of algebraic equations.
--------------------------------
Pearl, J., ``Jeffrey's Rule and the Problem of Autonomous
Inference Agents,'' UCLA Cognitive Systems Laboratory Technical
Report (R-62), June 1986, UCLA CSD #860099, June 1986.
ABSTRACT
Jeffrey's rule of belief revision was devised by philosophers to
replace Bayes conditioning in cases where the evidence cannot be
articulated propositionally. This paper shows that unqualified
application of this rule often leads to paradoxical conclusions,
and that to determine whether or not the rule is valid in any
specific case, one must first have topological knowledge about
one's belief structure. However, if such topological knowledge
is, indeed, available, belief updating can be done by traditional
Bayes conditioning; thus, arises the question of whether it is
ever necessary to use Jeffrey's rule in formalizing belief
revision.
--------------------------------
Pearl, J., ``Distributed Revision of Belief Commitment in Multi-
Hypotheses Interpretation,'' UCLA Computer Science Department
Technical Report CSD-860045 (R-64), June 1986; presented at the
2nd AAAI Workshop on Uncertainty in Artificial Intelligence,
Philadelphia, PA., August 1986.
ABSTRACT
This paper extends the applications of belief-networks models to
include the revision of belief commitments, i.e., the categorical
instantiation of a subset of hypotheses which constitute the most
satisfactory explanation of the evidence at hand. We show that,
in singly-connected networks, the most satisfactory explanation
can be found in linear time by a message-passing algorithm
similar to the one used in belief updating. In multiply-
connected networks, the problem may be exponentially hard but, if
the network is sparse, topological considerations can be used to
render the interpretation task tractable. In general, finding
the most probable combination of hypotheses is no more complex
than computing the degree of belief for any individual
hypothesis.
------------------------------
Geffner, H. & Pearl, J., ``A Distributed Approach to Diagnosis,''
UCLA Cognitive Systems Laboratory Technical Report (R-66),
October 1986;
ABSTRACT
The paper describes a distributed scheme for finding the most
likely diagnosis of systems with multiple faults. The scheme
uses the independencies embedded in a system to decompose the
task of finding a best overall interpretation into smaller sub-
tasks of finding the best interpretations for subparts of the
net, then combining them together. This decomposition yields a
globally-optimum diagnosis by local and concurrent computations
using a message-passing algorithm. The proposed scheme offers a
drastic reduction in complexity compared with other
methods: attaining linear time in singly-connected networks and,
at worst, exp ( | cycle-cutset | ) time in multiply-connected
networks.
-------------------------------
Pearl, J., ``Evidential Reasoning Using Stochastic Simulation of
Causal Models,'' UCLA Cognitive Systems Laboratory Technical
Report (R-68-I), October 1986.
ABSTRACT
Stochastic simulation is a method of computing probabilities by
recording the fraction of time that events occur in a random
series of scenarios generated from some causal model. This paper
presents an efficient, concurrent method of conducting the
simulation which guarantees that all generated scenarios will be
consistent with the observed data. It is shown that the
simulation can be performed by purely local computations,
involving products of parameters given
with the initial specification of the model. Thus, the method
proposed renders stochastic simulation a powerful technique of
coherent inferencing, especially suited for tasks involving
complex, non-decomposable models where ``ballpark'' estimates of
probabilities will suffice.
---------------------------
Pearl, J., ``Legitimizing Causal Reasoning in Default Logics''
(note), UCLA Cognitive Systems Laboratory Technical Report (R-
69), September 1986.
ABSTRACT
The purpose of this note is to draw attention to certain aspects
of causal reasoning which are pervasive in ordinary discourse
yet, based on the author's scan of the literature, have not
received due treatment by logical formalisms of common-sense
reasoning. In a nutshell, it appears that almost every default
rule falls into one of two categories: expectation-evoking or
explanation-evoking. The former describes association among
events in the outside world (e.g., Fire is typically accompanied
by smoke.); the latter describes how we reason about the world
(e.g., Smoke normally suggests fire.). This distinction is
clearly and reliably recognized by all people and serves as an
indispensible tool for controlling the invocation of new default
rules. This note questions the ability of formal systems to
reflect common-sense inferences without acknowledging such
distinction and outlines a way in which the flow of causation can
be summoned within the formal framework of default logic.
---------------------------------
Dechter, R. & Pearl, J., ``The Cycle-Cutset Method for Improving
Search Performance in AI Applications,'' UCLA Cognitive Systems
Laboratory Technical Report (R-67); submitted to the 3rd IEEE
Conference on Artificial Intelligence Applications.
ABSTRACT
This paper introduces a new way of improving search performance by
exploiting an efficient method available for solving tree-structured
problems. The scheme is based on the following observation: If, in
the course of a backtrack search, we remove from the constraint-graph
the nodes corresponding to instantiated variables and find that the
remaining subgraph is a tree, then the rest of the search can be
completed in linear time. Thus, rather than continue the search
blindly, we invoke a tree-searching algorithm tailored to the topology
of the remaining subproblem. The paper presents this method in detail
and evaluates its merit both theoretically and experimentally.
------------------------------
End of NL-KR Digest
*******************
