LAWS@SRI-AI.ARPA (04/23/85)
From: AIList Moderator Kenneth Laws <AIList-REQUEST@SRI-AI>
AIList Digest Tuesday, 23 Apr 1985 Volume 3 : Issue 50
Today's Topics:
Administrivia - Lost Messages,
Machine Translation - La Jolla,
Application - AI in Agriculture,
Seminars - Transformation of Functional Equations (MIT) &
Programming with Recurrent Equations (Penn) &
Representation, Aesthetics, Learnability (SU) &
The MIT Mobile Robot Project (Penn) &
ARLO: Representing Representation Language (MIT),
Humor - Representation Lunches & The Traveling President Problem
----------------------------------------------------------------------
Date: Mon 22 Apr 85 20:44:07-PST
From: Ken Laws <Laws@SRI-AI.ARPA>
Reply-to: AIList-Request@SRI-AI.ARPA
Subject: Lost Messages
SRI-AI had a bad system crash Sunday morning; mail sent to AIList
or AIList-Request that morning may not have gotten through.
-- Ken Laws
------------------------------
Date: 19 APR 85 14:16-N
From: PETITP%CGEUGE51.BITNET@WISCVM.ARPA
Subject: Machine translation at La Jolla
[In answer to request from goodhart@nosc AIList Digest v3 #46]
Hi!
The people you are looking for in La Jolla are probably working on
SYSTRAN, a commercial machine translation system. Here is their address:
P. Toma
WTC Inc
7854 Ivanohe Avenue
PO Box 907
La Jolla, Ca 92037
Tel: 619/459-3471
Other commercial systems I know of in the States are the Weidner System
and the ALPS system, both companies are located in Provo, Utah. (I could get
more for you if you are interested).
But none of those are proper research projects, and are based on
linguistically and computationnally "old" ideas. A more advanced project
is METAL, developed at the Linguistic Research Center, University of Texas,
in Austin. You can contact Rebecca Root (LRC.ROOT@UTEXAS.ARPA) to get more
information about it.
In Europe there are the SUSY project (Saarbruecken,Germany), and the GETA
project (Grenoble, France), and the EUROTRA project of the European Economic
Communities, to which many european universities collaborate. Here at ISSCO
in Geneva we are working on EUROTRA.
And of course there is many Japanese projects but I don't know much about
them.
A good introduction to machine translation is a paper by Jonathan Slocum,
presented at COLING-84 in Stanford: "Machine Translation: its History, Current
Status an Future Prospects". If you can't get a copy of the proceedings,
I think it was also published as a report by the LRC in Austin.
Last year ISSCO organised a tutorial on machine translation and a book will
be published by Edinburgh University Press.
Dominique Petitpierre (PETITP@CGEUGE51.BITNET)
ISSCO
54 route des acacias
CH-1227 GENEVA (Switzerland)
------------------------------
Date: Mon, 22 Apr 85 08:23 EST
From: kyle.wbst@Xerox.ARPA
Subject: Friedland's request about AI in agriculture
Some years ago, Control Data Corp in Minneapolis had a New Business
Ventures Unit that included various software systems for the farmer and
Agribusiness industry. You might check with them to see what is going on
in that area with them now. Also at about that same time frame they got
on board a former SRI type who had background with SRI's early AI
efforts. I don't know if there was any connection between that person
and the farm stuff though.
Earle Kyle.
------------------------------
Date: Tue,16 Apr 85 16:16:24 EST
From: Robyn D. Spencer <TOOTSE@MIT-MC>
Subject: Seminar - Transformation of Functional Equations (MIT)
[Forwarded from the MIT bboard by Laws@SRI-AI.]
DATE: April 22, 1985
TIME: Lecture 3:30 p.m.
PLACE: NE43-3rd floor conference room
TRANSFORMATIONS of HIGHER-TYPE FUNCTIONAL EQUATIONS
VIA THE COMPUTATION OF RETRACTS
Richard Statman
Carnegie-Mellon University
Functional equations occur in diverse branches of logic and computer
science. In type theory with the axiom of choice, every formula is equivalent
to one which asserts that a functional equation has a solution with the given
parameters. In theorem proving, unification problems are simply functional
equations which one wants to solve in all models. In programming language
semantics, programming constructs, such as fixed point operators, are
represented as solutions to functional equations.
The general functional equation
Fx=Gx
where operators F,G have type A -> B can be transformed one with
operators of type C -> D while preserving all solutions, when there is a
surjective H of type C -> A and an injective J of type B -> D. The transformed
equation is
J(F(Hy))=J(G(Hy))
We shall show that this transformation can be carried out in every model if and
only if A is a retract of C and B is a retract of D in every model. There is a
retract from C onto A if and only if the equation
lambda z. y(xz) = I
is solvable for some y in C -> A and x in A -> C. This equation is solvable in
every model if and only if it is solvable in the term model of beta-eta
conversion. Thus transformations of the above type can always be carried out by
lambda terms. We shall give some further information about when this type of
transformation can be carried out including bounds on the size of A as a
function of the size of C. The decision problem (unification problem) is open.
HOST: Albert Meyer
------------------------------
Date: Wed, 17 Apr 85 21:06 EST
From: Tim Finin <Tim%upenn.csnet@csnet-relay.arpa>
Subject: Seminar - Programming with Recurrent Equations (Penn)
PROGRAMMING WITH RECURRENT EQUATIONS
Boleslaw Szymanski (University Pennsylvania)
3:00pm April 23, 1985, 216 Moore School, Univ. of Pennsylvania
Software development tools proposed for new generation of computers are
based on assertive programming, where a program is expressed as a set of
assertions. There are two basic notations used in assertive programming:
Horn clauses of logic programming (e.g. PROLOG) and conditional equations
used in so called definitional or equational languages. Equational
languages are natural and convenient complements of PROLOG-like languages
for such applications as programming dataflow machines and modelling complex
systems.
This talk focuses on languages based on recurrent equations. Finite-
difference approximations to systems of partial difference equations lead to
such recurrence equations. Our experience indicates that such languages are
general purpose. Description of many algorithms is greatly simplified when
presented in such a form.
The talk presents new results of the MODEL project. Three MODEL language
processor components: Compiler, Configurator, and Timing System will be
discussed. The emphasis will be on the following problems: 1) optimization
of programs generated by the MODEL compiler, 2) programming parallel and/or
distributed computations with Configurator 3) use of temporal relations for
scheduling parallel components 4) distributed termination of a solution to
simultaneous equations, 5) real-time software development using Timing
System. Future research will also be outlined.
------------------------------
Date: Wed 17 Apr 85 17:11:18-PST
From: Emma Pease <Emma@SU-CSLI.ARPA>
Subject: Seminar - Representation, Aesthetics, Learnability (SU)
[Excerpted from the CSLI Newsletter by Laws@SRI-AI.]
CSLI ACTIVITIES FOR *NEXT* THURSDAY, April 25, 1985
4:15 p.m. CSLI Colloquium
Redwood Hall ``The Representational Basis for Everyday Aesthetic
Room G-19 Experience -- A Motivational Constraint on Learnable
Systems of Knowledge''
Tom Bever, Columbia University and CASBS
``The Representational Basis for Everyday Aesthetic Experience --
A Motivational Constraint on Learnable Systems of Knowledge''
The structure of everyday aesthetic judgements depends on computations
of mental representations and relations between representations.
Examination of objects of everyday aesthetic preference (e.g., simple
rhythms, shapes, and songs) affords a definition of the aesthetically
satisfying experience: such experiences involve the formation of
incompatible representations and their resolution within the framework
of an overarching representational system. The enjoyment of such
experiences follows from the extent to which they are like solving a
problem during normal cognitive development. Indigenous systems like
language must have formal properties that stimulate aesthetically
satisfying experiences as an immediate motivation for the acquisition of
abstract structures. That is, we learn a multi-levelled representational
structure for language because it is fun. --Tom Bever
------------------------------
Date: Thu, 18 Apr 85 14:59 EST
From: Tim Finin <Tim%upenn.csnet@csnet-relay.arpa>
Subject: Seminar - The MIT Mobile Robot Project (Penn)
THE MIT AI LAB MOBILE ROBOT PROJECT - Rodney A. Brooks (MIT)
3pm April 25, 23 Moore School, Univ. of Pennsylvania
We are interested in a number of questions relating to intelligent
mobile robots. These include the following. (a) How to combine a
number of early vision modules into a robust vision system which can
operate under a wide range of conditions and a wide range of scenes
through redundancy of perceptions. (b) How to make reliable maps given
that all sensors produce error laden readings, and control of the robot
is also a source of error. (c) How to apply model-based vision
techniques to the landmark selection and recognition problems. (d) How
to make a robot control and planning system which is competent and
robust enough to allow an autonomous vehicle to operate for long
periods with absolutely no assistance from a human. In support of
these goals we are building a mobile robot which will operate
autonomously for a number of hours at a time within the Artificial
Intelligence Laboratory office area. Our approach to building the
robot and its controlling software differs from that used in many other
projects in a number of ways. (1) We model the world as three
dimensional rather than two. (2) We build no special environment for
our robot and insist that it must operate in the same real world that
we inhabit. (3) In order to adequatley deal with uncertainty of
perception and control we build relational maps rather than maps
embedded in a coordinate system, and we maintain explicit models of all
uncertainties. (4) We explicitly monitor the computational performance
of the components of the control system, in order to refine the design
of a real time control system for mobile robots based on a special
purpose distributed computation engine. (5) We use vision as our
primary sense and relegate acoustic senors to local obstacle detection.
(6) We use a new architecture for an intelligent system designed to
provide integration of many early vision processes, and robust
real-time performance even in cases of sensory overload, failure of
certain early vision processes to deliver much information in
particular situations, and computation module failure.
------------------------------
Date: 19 Apr 1985 15:57 EST (Fri)
From: "Daniel S. Weld" <WELD%MIT-OZ@MIT-MC.ARPA>
Subject: Seminar - ARLO: Representing Representation Language (MIT)
[Forwarded from the MIT bboard by SASW@MIT-MC.]
AI Revolving Seminar Ken Haase
ARLO: Representing Representation Languages
Tuesday, April 23; 4:00pm; 8th Floor Playroom
ARLO is a language for describing the implementation and functionality
of frame based representation languages. A given representation
language is specified in ARLO by a collection of structures describing
how its descriptions are interpreted, defaulted, and verified. This
high level description is compiled into lisp code and ARLO structures
whose interpretation fulfills the langauge's abstract specification.
The dependencies of this compilation process (from description to
implementation) are recorded by ARLO, so that changes in the
high-level description will propogate to the generated implementation.
In addition, ARLO itself --- as a representation language for
expressing and compiling partial and complete language specifications
--- is described and interpreted in the same manner as the languages
it describes and implements.
This talk will address general issues in the definition and
implementation of representation language languages, as well as the
technical problems in implementing self-descriptive systems. Finally,
I will discuss the use of ARLO-like languages as a basis for learning
and concept formation programs like Lenat's Eurisko.
------------------------------
Date: 11 Apr 1985 13:18 EST (Thu)
From: Mike Gennert@MIT-OZ <MICHAELG%MIT-OZ@MIT-MC.ARPA>
Subject: Representation Lunches
[Forwarded from the MIT bboard by SASW@MIT-MC.]
COMPUTER AIDED CONCEPTUAL ART (CACA)
REVOLTING SEMINAR SERIES
presents
TACO: REPRESENTING REPRESENTATION LUNCHES
Mike Gennert
Mike Gerstenberger
TACO is a lunch for describing the implementation and functionality of
frame based representation lunches. A given representation lunch is
specified in TACO by a collection of fillings describing how its
descriptions are interpreted, defaulted, verified, and eaten. This
high level description is compiled into TACO shell code and TACO
fillings whose interpretation fulfills the langauge's nutritional
specification. The dependencies of this compilation process (from
description to implementation) are recorded by TACO, so that changes
in the high-level description will propogate to the generated
implementation. In addition, TACO itself --- as a representation
lunch for expressing and compiling partial and complete lunch
specifications --- is devoured and interpreted in the same manner as
the lunches it devoured and implements, i.e., with one's fingers.
This talk will address general issues in the definition and
implementation of representation lunch lunches, as well as the
technical problems in implementing self-devouring systems. Finally,
We will discuss the use of TACO-like lunches as a basis for learning
and concept formation programs like Automatic Hairstyle Generation:
The Further Adventures of Eurisko.
Friday, April 12, 12:00, 3rd Floor Lounge
------------------------------
Date: Thu, 18 Apr 85 17:26:10 est
From: Walter Hamscher <hamscher at mit-htvax>
Subject: The Traveling President Problem
[Forwarded from the MIT bboard by SASW@MIT-MC.]
COMPUTER AIDED CONCEPTUAL ART (CACA)
REVOLTING SEMINAR SERIES
:-) (-:
THE TRAVELING PRESIDENT PROBLEM
Tod Malmedy
We present a new variation of the traveling salesman problem.
There are two differences. First, the links are free but the
nodes, representing memorials, have either small positive or
large negative weights, and the goal is to maximize this
weight. Second, unlike most problems, in which a solver is
allowed to exaustively search the space of combinations for an
optimal solution, in this variation the value of the final
solution is penalized by the number of combinations tested
before finding it. Under these conditions the L/D (Leave it
to Deaver) strategy can be proven to be the worst possible.
TIME: 12 Noon Friday
PLACE: 3rd Floor Theory Playroom
HOSTS: Bhaskar Ghudaroy and Mike Beckerle
------------------------------
End of AIList Digest
********************