Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 044
AIList Digest Thursday, 6 Mar 1986 Volume 4 : Issue 44
Today's Topics:
Seminars - Acquiring Language & Computer Lexicon Use (SD SIGART) &
Commonsense Knowledge in the TACITUS Project (SU) &
Hubert Dreyfus on Being and Time (MIT) &
Intelligent Distributed Operating Systems (USC) &
Delegation and Inheritance (MIT) &
Refinement of Expert System Knowledge Bases (CMU) &
Heuristic Search: Algorithms, Theory, and Learning (CMU) &
Brains, Behavior, and Robotics (CSLI) &
Situation Calculus Planning (SRI) &
The Perspective Concept in Computer Science (CSLI)
----------------------------------------------------------------------
Date: 4 March 1986 1604-PST (Tuesday)
From: gross@nprdc.arpa (Michelle Gross)
Subject: Seminars - Acquiring Language & Computer Lexicon Use (SD SIGART)
Subject: SD SIGART-NLP meetings--Last and Next
We've been meeting the first Monday of each month.
Last night's meeting (our 3rd) covered Dr. Bob La Quey's efforts to
write a program that acquires language by determining which
grammatical rules are needed to parse incrementally more complex
text. The main difficulty with his approach seems to be how to prevent
adding spurious rules when ungrammatical sentences sneak through.
Someone suggested attaching a reliability index to each rule. The
index would be based on how often the rule has successfully helped a
parse get through. (The hope is that the ad hoc rules for
ungrammatical input would have low index values).
We also discovered that the only given rule in the
grammar (S --> N V Terminator) prevented the program from creating a
rule to parse imperative sentences (S --> V). Mallory Selfridge's 1981
IJCAI paper ``A Computer Model of Child Language Acquisition'' provided
some of the impetus for Bob's work. His talk was entitled ``A Model of
Language Acquisition.''
Our next meeting will be April 7th. The topic will be the
lexicon--how we use it and how a computer can use it. I volunteered to
present some relevant linguistic and computational literature. I plan
to discuss how the lexicon is viewed in Transformational Grammar,
Lexical Functional Grammar, and Relational Grammar (I don't know enough
about GPSG to touch on that perspective). I plan to discuss Cherry's
paper on the UNIX tool PARTS (a program from the Writer's Workbench
that assigns parts of speech by rule). I would also like to discuss
the data structures used in various dictionary projects.
Can anyone provide pointers to such information for the OED
or Webster's projects? Any other references or abstracts
you can send would only enrich our provincial San Diegan
discussions! I have a 1982 IEEE article on PARTS and Cherry's
1978 paper--are there are more recent references?
For more information on the SIG, you may contact Ed Weaver at work at
(619) 236-5963. I'll forward any electronic responses on to him.
Thanks,
Michelle gross@nprdc.ARPA ...ihnp4!sdcsvax!sdcc6!ix713 (UUCP)
Navy Personnel R&D Center UCSD Linguistics, C-008
San Diego, CA. 92152-6800 La Jolla, CA. 92093
------------------------------
Date: 03 Mar 86 1042 PST
From: Vladimir Lifschitz <VAL@SU-AI.ARPA>
Subject: Seminar - Commonsense Knowledge in the TACITUS Project (SU)
Commonsense Knowledge in the TACITUS Project
Jerry R. Hobbs
Artificial Intelligence Center
SRI International
Thursday, March 6, 4pm
MJH252
In the TACITUS project for using commonsense knowledge in the
understanding of texts about mechanical devices and their failures, we
have been developing various commonsense theories that are needed to
mediate between the way we talk about the behavior of such devices and
causal models of their operation. Of central importance in this effort
is the axiomatization of what might be called ``commonsense
metaphysics''. This includes a number of areas that figure in virtually
every domain of discourse, such as granularity, scales, cycles, time,
space, material, physical objects, shape, causality, functionality, and
force. Our effort has been to construct core theories of each of these
areas, and then to define, or at least characterize, a large number of
lexical items in terms provided by the core theories. In this talk I
will discuss our methodological principles, such as aiming for the
maximum abstraction possible in order to accommodate metaphor and
analogy, and I will describe the key ideas in the various domains we are
investigating.
------------------------------
Date: Tue, 4 Mar 1986 20:35 EST
From: AGRE%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU
Subject: Seminar - Hubert Dreyfus on Being and Time (MIT)
Artificial Intelligence Seminar
Monday, March 10, 2:30pm
545 Technology Square
(MIT Building NE43)
7th Floor Playroom
WHY YOU SHOULD READ BEING AND TIME
Hubert L. Dreyfus
Philosophy Department
UC Berkeley
The beauty of artificial intelligence is that computation keeps you honest:
mistaken approaches will simply fail. I will argue that a diagnosis of
current difficulties in AI research can be found in the work of Martin
Heidegger. Heidegger's Being and Time isolates a number of assumptions of
Western philosophy which, though subtle and pervasive, are contradicted by a
careful account of the phenomenology of everyday activity. These
assumptions and their corollaries have been implicit (and sometimes
explicit) in most AI work since the field's beginnings. The task now is to
find a positive alternative. I will start by presenting some of the basic
concepts of Heidegger's phenomenology. But Heidegger's account of everyday
practices does not directly provide an alternative to traditional methods in
AI because it offers a description rather than a mechanizable explanation.
It is difficult to reason about the ways descriptions and explanations
constrain one another. Still, I will attempt a start by outlining the
virtues and failings of some new approaches, in particular those of the
connectionist movement.
------------------------------
Date: 4 Mar 1986 12:28-EST
From: gasser@usc-cse.usc.edu
Subject: Seminar - Intelligent Distributed Operating Systems (USC)
USC Distributed Problem Solving Group Meeting
Wednesday, 3/12/86 3:00 - 5:00 PM
Seaver Science 319
John Gieser, Ph.D. Student, USC, will speak on "'Intelligent'
Operating Systems for Distributed Computing".
ABSTRACT
Recent ideas from distributed problem solving (DPS) research appear
to have merit when used to acheive cooperation in open-ended
distributed computing systems (DCS). To use these techniques, the
DCS nodes are viewed as autonomous agents in a problem-solving
situation, with each node governed by an "intelligent" operating
system (IOS). This talk will focus on some ideas for providing the
structures and mechanisms needed in the IOS to handle problems
requiring cooperation such as distributed control, load
balancing/sharing, cooperating processes, etc.
Questions: Dr. Les Gasser, (213) 743-7794, or
John Gieser (gieser@usc-cse.usc.edu)
------------------------------
Date: Tue, 4 Mar 86 16:31 EST
From: Jonathan Connell <jhc@OZ.AI.MIT.EDU>
Subject: Seminar - Delegation and Inheritance (MIT)
[Forwarded from the MIT bboard by SASW@MC.LCS.MIT.EDU.]
Thursday , March 6 4:00pm Room: NE43- 8th floor Playroom
The Artificial Intelligence Lab
Revolving Seminar Series
Delegation And Inheritance:
Two Mechanisms for Sharing Knowledge in Object-Oriented Systems
Henry Lieberman
AI Lab, MIT
When a group of objects in an object oriented programming system shares
some common behavior, how can we avoid re-programming behavior in every
object that needs it? I will explore the consequences of two mechanisms
for sharing knowledge, Inheritance and Delegation, for expressiveness
and performance of object oriented languages.
Using Inheritance, behavior common to a group of objects is encoded in a
Class object, which contains procedures for responding to messages, and
the names of variables that the procedure may access. Each class may
create a set of Instances, which share the procedures of the class, but
may have their own private values for the variables. Subclasses may
extend classes by adding additional procedures and variables.
Another way of sharing behavior is Delegation, which views each object
as a prototype capable of creating new objects by copying or reference,
removing the distinction between classes and instances. General and
specialized objects communicate using message passing rather than a
"hard wired" mechanism. Communication patterns can be determined at
message reception time rather than at compile time or object creation
time. There is a time/space tradeoff between inheritance and
delegation, delegation permitting smaller objects at the cost of
increased message traffic.
------------------------------
Date: 20 February 1986 1450-EST
From: Betsy Herk@A.CS.CMU.EDU
Subject: Seminar - Refinement of Expert System Knowledge Bases (CMU)
Speaker: Allen Ginsberg, Rutgers University
Date: Wednesday, March 5
Time: 11:30 - 1:00
Place: 5409 WeH
Title: The automatic refinement of expert system knowledge bases
Knowledge base refinement involves the generation, testing, and
possible incorporation of plausible refinements to the rules in
a knowledge base with the intention of thereby improving the
empirical adequacy of an expert system, i.e., its ability to
correctly diagnose or classify the cases in its domain of expertise.
The first part of the talk is a theoretical explication of the
basic concepts involved in knowledge base refinement -- e.g., a
precise analysis of one sense in which a refinement may be said
to be plausible is given -- and includes an overview of the
strategic goals that must be addressed by any knowledge base
refinement system. As an illustration of the general theory,
the second part of the talk focuses on the SEEK2 system for
automatic knowledge base refinement. In the last part of the
talk a brief discussion of a metalanguage for the experimental
design of refinement systems is given.
------------------------------
Date: 27 February 1986 1153-EST
From: Betsy Herk@A.CS.CMU.EDU
Subject: Seminar - Heuristic Search: Algorithms, Theory, and Learning (CMU)
Speaker: Richard Korf, Asst. Prof., Comp. Sci. Dept., UCLA
Date: Friday, March 14
Time: 1:00 - 2:30
Place: 5409 Wean Hall
Title: Heuristic search: Algorithms, theory, and learning
Abstract:
This talk will cover three new research results in the area of heuristic
search. The first is a new algorithm, called Iterative-Deepening-A*, that is
asymptotically optimal in terms of solution cost, time, and space among all
admissible heuristic tree searches. In practice, it is the only known
algorithm that is capable of finding optimal solutions to the Fifteen
Puzzle. The second is a theory which unifies the treatment of heuristic
evaluation functions in single-agent problems and two-person games. The
theory is based on the notion of a heuristic as a function that is invariant
over optimal solution paths. Based on this theory, we performed some
experiments on the automatic learning of heuristic functions. Our program
was able to learn a set of relative weights for the different chess pieces
which is different from, but competitive with, the classical values.
------------------------------
Date: Wed 5 Mar 86 16:57:49-PST
From: Emma Pease <Emma@SU-CSLI.ARPA>
Subject: Seminar - Brains, Behavior, and Robotics (CSLI)
[Excerpted from the CSLI Calendar by Laws@SRI-AI.]
Brains, Behavior, and Robotics
by James S. Albus
Discussion led by Pentti Kanerva (Kanerva@riacs.arpa)
12 noon, TINLunch, Ventura Hall Conference Room
THURSDAY, March 13, 1986
In 1950, Alan Turing wrote, ``We may hope that machines will
eventually compete with men in all purely intellectual fields. But
which are the best ones to start with? . . . Many people think that
a very abstract activity, like the playing of chess, would be best.
It can also be maintained that it is best to provide the machine with
the best sense organs that money can buy, and then teach it to
understand. . . . This process could follow the normal teaching of a
child. Things would be pointed out and named, etc. Again I do not
know what the right answer is, but I think that both approaches should
be tried.'' (Quoted by Albus on p. 5.)
``Brains, Behavior, and Robotics'' takes this ``Turing's second
approach'' to artificial intelligence, the first being the pursuit of
abstract reasoning. The book combines over a decade of research by
Albus. It is predicated on the idea that to understand human
intelligence we need to understand the evolution of intelligence in
the animal kingdom. The models developed are mathematical
(computational), but one of their criteria is neurophysiological
plausibility. Although the research is aimed at understanding the
mechanical basis of cognition, Albus also discusses philosophical and
social implications of his work.
------------------------------
Date: Wed 5 Mar 86 16:44:14-PST
From: LANSKY@SRI-AI.ARPA
Subject: Seminar - Situation Calculus Planning (SRI)
SITUATION CALCULUS PLANNING IN BLOCKS AND RELATED WORLDS
John McCarthy (JMC@SU-AI)
Stanford University
11:00 AM, MONDAY, March 10
SRI International, Building E, Room EJ228 (new conference room)
This talk will present mainly ideas rather than completed work.
Situation calculus is based on the equation s' = result(e,s),
where s and s' are situations and e is an event. Provided
one can control the deduction adequately, this is a more powerful
formalism than STRIPS. Planning a sequence of actions, or more
generally, a strategy of actions to achieve a situation with
specified properties, admits a variety of heuristics which
whittle away at the problem. In many practical situations, these
heuristics, which don't guarantee a full solution but leave a
reduced problem, are sufficient. Humans appear to use many of them
and so should computer programs. The talk therefore will concern both
epistemological and heuristic aspects of planning problems.
------------------------------
Date: Wed 5 Mar 86 16:57:49-PST
From: Emma Pease <Emma@SU-CSLI.ARPA>
Subject: Seminar - The Perspective Concept in Computer Science (CSLI)
[Excerpted from the CSLI Calendar by Laws@SRI-AI.]
SYSTEM DESCRIPTION AND DEVELOPMENT TALK
The Perspective Concept in Computer Science
12:15, Monday, March 10, Ventura Conference Room
Our topic next Monday (March 10) will be a continued discussion
(introduced by Jens Kaasboll) of the issues raised by Kristen Nygaard
in his talk about perspectives on the use of computers:
Regardless of definitions of ``perspective'', there exist many
perspectives on computers. Computers are regarded as systems, tools,
institutions, toys, partners, media, symbols, etc. Even so, there
exist system description languages but no tool, or institution, or
... languages. What do the other perspectives reflect, which make
them less attractive for language designers? Suggestive answer: The
system perspective is the definite computer science perspective in
which the processes inside the computers are regarded as the goal of
our work. Viewed through some of the other perspectives, the computer
is seen as a means for achieving ends outside the computer, i.e., the
needs of people using the computers.
------------------------------
End of AIList Digest
********************
