[Prev][Next][Index]
CSLI Calendar, 9 March 1995, vol.10:19
-
To: friends
-
Subject: CSLI Calendar, 9 March 1995, vol.10:19
-
From: Tom Burke <burke>
-
Date: Wed, 8 Mar 1995 14:26:54 -0800
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
______________________________________________________________________________
9 March 1995 Stanford Vol. 10, No. 19
______________________________________________________________________________
A weekly publication of the Center for the Study of Language and
Information (CSLI), Stanford University, Ventura Hall, Stanford, CA 94305-4115
____________
CSLI ACTIVITIES DURING 8 -- 17 MARCH 1995
WEDNESDAY, 8 MARCH
4:00 - Seminar on Computational Learning and Adaptation
Building 370, Room 370
An Experimental and Theoretical Comparison of Model
Selection Methods
Michael Kearns, AT&T Bell Labs
Abstract below
THURSDAY, 9 MARCH
12:00 - CSLI TINLunch
Cordura Hall, Room 100
Cognitive Science's Missed Opportunity: Discussion of
Jean-Pierre Dupuy's _Aux Origines des Sciences Cognitives_
Stefano Franchi, Stanford Philosophy
Abstract below
2:15 - CSLI Seminar
Series on Visual Programming
Cordura Hall, Room 100
Directing the Behavior of Improvisational Characters
Barbara Hayes-Roth, Stanford Computer Science
Abstract below
4:15 - SSP Forum
Building 60, Room 61-H (New Room!)
Snow-Crashing through Cyberspace: Computers and the
Politics of Subjectivity
Paul N. Edwards, Stanford Science, Technology, Society Program
Abstract below
7:30 - Phonology Workshop
Building 460, Seminar Room 146
Deriving Variation from Grammar
Arto Anttila, Stanford Linguistics
Abstract below
FRIDAY, 10 MARCH
12:00 - Logic Lunch
Building 380, Room 383-N
A New Path to Pseudorecursive Varieties
Benjamin Wells, U San Francisco Mathematics and CS
Abstract below
12:30 - HCI Seminar
Terman Auditorium
Visualizing Large Information Structures Using
Focus+Context Techniques
John Lamping and Ramana Rao, Xerox PARC
Abstract below
3:15 - Philosophy Colloquium
Building 60, Room 61-H
Sovereign Performatives in the Contemporary Scene of
Utterance: Speech and Conduct in the Contemporary
Politics of Representation
Judith Butler, UC Berkeley Rhetoric & Comparative Literature
Abstract below
3:30 - Linguistics Colloquium
Cordura Hall, Room 100
Speakers, Perspectives, and Words in Acquisition
Eve Clark, Stanford Linguistics
Abstract below
MONDAY, 13 MARCH
2:15 - Syntax Seminar
Cubberley Hall, Room 229
Determiner Quantifiers and Focus
Helen de Hoop, University of Groningen
Abstract below
TUESDAY, 14 MARCH
7:30 - Phonology Workshop
306 San Mateo Drive, Menlo Park (Kiparsky)
Only the Trochee
Ruben van de Vijver, Vrije Universiteit
Abstract below
7:45 - Syntax Workshop
11 Franciscan Ridge, Portola Valley (Bresnan)
Syntax of the Gikuyu NP
John Mugane, Stanford Linguistics
THURSDAY, 16 MARCH
12:00 - CSLI TINLunch
Cordura Hall, Room 100
Vagueness, Empty Names, and Supervaluations
Ernest Lepore, Rutgers Philosophy
Abstract below
2:15 - CSLI Seminar
Series on Visual Programming
Cordura Hall, Room 100
Visual AppBuilder
Kurt Schmucker, Apple Computer
Abstract below
4:15 - SSP Forum
Building 60, Room 61-F
AI-Based Planning for Image Processing Tasks
Amy L. Lansky, NASA Ames Research Center
Abstract below
FRIDAY, 17 MARCH
12:30 - HCI Seminar
Terman Auditorium
Implicit Structures for Pen-Based Systems Within
a Freeform Interaction Paradigm
Tom Moran, Xerox PARC
3:30 - Linguistics Colloquium
Cordura Hall, Room 100
The Once and Future Dictionary
Geoff Nunberg, Xerox PARC and Stanford Linguistics
Abstract below
____________
The CSLI Calendar appears on Thursday of each week throughout the academic
year. Announcements, abstracts, and other information to appear in the
Calendar on a given Thursday should be submitted by e-mail to
incalendar@csli.stanford.edu by 5:00 p.m. on the previous Tuesday.
Past issues of the CSLI Calendar, a quarterly schedule of upcoming CSLI
events, and other information about CSLI are available on the World Wide Web:
<http://csli-www.stanford.edu/>. The Calendar, with available abstracts, is
also posted each week to the csli.bboard newsgroup.
____________
SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
on Wednesday, 8 March
4:00 p.m., Bldg 370, Room 370
An Experimental and Theoretical Comparison
of Model Selection Methods
Michael Kearns
AT&T Bell Labs
In the model selection problem, we must balance the empirical error of a
statistical model with its complexity (as measured by the number of
parameters, code length, or any other appropriate measure), in order to
minimize the resulting generalization error. Well-known proposals for solving
this problem include "penalty-based" methods (such as Rissanen's minimum
description length principle and Vapnik's guaranteed risk minimization), as
well as the widely used cross validation. In this talk I will provide both
controlled experimental evidence and theoretical results to support the
following conclusions:
(1) Even on simple model selection problems, the behavior of the
methods examined can be complex and incomparable. Furthermore,
no amount of "tuning" of the methods (such as the introduction
of constant multipliers on penalty terms, or notions of
"effective dimension") can eliminate this incomparability.
(2) It is possible to give general bounds on the generalization error
for any penalty-based method, and also for cross validation.
The quality of the penalty-based bound is determined by the
extent to which the penalty-based method under consideration
automatically limits the complexity of the model selected. The
quality of the cross-validation bound is determined by the
extent to which the underlying learning curve adheres to the
classically predicted power law behavior.
(3) The entire class of penalty-based methods is fundamentally
handicapped, in the sense that there exist two types of model
selection problems such that every penalty-based method must
incur large generalization error on at least one type, while
cross validation enjoys small generalization error on both.
This talk describes joint research with Yishay Mansour, Andrew Ng, and Dana
Ron.
The goal of this seminar is to increase communication among local researchers
with interests in computational approaches to learning and adaptation. If you
would like to be added to (or removed from) the mailing list, or if you are
interested in giving a talk in the seminar, please send email to
<langley@cs.stanford.edu>.
____________
CSLI TINLUNCH
on Thursday, 9 March
12:00 noon, in Cordura Hall, Room 100
Cognitive Science's Missed Opportunity: Discussion of
Jean-Pierre Dupuy's _Aux Origines des Sciences Cognitives_
Stefano Franchi
Stanford Philosophy
<sfranchi@parc.xerox.com>
Is there any chance to solve the age-old philosophical problems about the
nature of the mind, the structure of cognition, and the like, without falling
into the metaphysical quagmires about the nature of the subject, its
relationship with the object, and so forth? According to Jean-Pierre Dupuy,
during the late 1940s cybernetics was about to reach such a goal by developing
a new understanding of man which would eschew any metaphysics of the subject.
Yet, the opportunity was missed, and not in the least because of its founders'
lack of philosophical background.
In this context, I will consider the following two questions: Can cybernetics'
heirs -- cognitive science and contemporary philosophy of mind -- profit from
the errors of the past? Or are they bound to repeat them because of their
refusal to acknowledge any relationship whatsoever with the discredited
ancestor?
Dupuy's book, _Aux Origines des Sciences Cognitives_, is on closed reserve in
Tanner Library, Building 90.
____________
CSLI SEMINAR
Series on Visual Programming
on Thursday, 9 March
2:15 p.m., in Cordura Hall, Room 100
Directing the Behavior of Improvisational Characters
Barbara Hayes-Roth
Stanford Computer Science
<bhr@hpp.stanford.edu>
"Directed improvisation" is a paradigm for human-computer interaction.
Directors (who may be human users or other computer programs) guide the
behavior of computer characters with abstract instructions that establish
skeletal narrative structures and weak constraints on the desired behaviors.
The characters work together to improvise a course of behavior that conforms
to the structure, meets the constraints, and achieves other performance
objectives. Thus, characters follow directions, but may also enhance
performance, while surprising and engaging users with their improvisations
along the way.
Directed improvisation appears to be a promising paradigm for a variety of
applications. Our current work focuses on a particular testbed application, a
"Virtual Theater for Children," in which children direct the improvisational
behavior of animated characters to create their own stories. In this
presentation, I will give a brief introduction to the directed improvisation
paradigm and the Virtual Theater project, but I will focus on how children
direct/program characters' behavior in both interactive and scripted modes.
____________
SYMBOLIC SYSTEMS FORUM
on Thursday, 9 March
4:15 p.m., Building 60, Room 61-H
Snow-Crashing through Cyberspace: Computers and
the Politics of Subjectivity
Paul N. Edwards
Stanford Program on Science, Technology, and Society
<pedwards@pcd.stanford.edu>
In the computer age, theories, beliefs, and fictions about mind, intelligence,
and selfhood must be understood as having a political sense. They reflect a
political and technical history involving new forms of warfare, pervasive
technological systems, and the global development of capitalism and its
culture. So, too, the political constellation of the post-WWII era must be
understood as involving the subjectivity of mental machines. This talk
explores the relations among real and hypothetical thinking machines, the
history of subjectivity, and modern politics. I will examine three films --
_Blade Runner_; _The Terminator_; and _Terminator II: Judgment Day_ -- and two
novels, William Gibson's _Neuromancer_ and Neal Stephenson's _Snow Crash_.
Their themes include transformations of gender roles, the ethical challenges
posed by intelligent machines, reconciliation with cyborg partners, and the
visual representation of virtual space. The talk explores the fluidity of
selfhood in the postmodern world of cyberspace. Throughout, I ground these
interpretations in an analysis of recent political and social realities.
____________
PHONOLOGY WORKSHOP
on Thursday, 9 March
7:30 p.m., Building 460, Room 146
Deriving Variation from Grammar
Arto Anttila
Stanford Linguistics
<anttila@csli.stanford.edu>
A way to derive morphological variation from UG principles is proposed. The
specific case discussed is Finnish noun inflection where phonologically
definable classes of roots permit extensive variation in suffix allomorphy.
An example is /fyysikko/ `physicist' which has (at least) three alternative
genitive plurals: /fyysikkojen/, /fyysikoiden/ and /fyysikoitten/. The
generalization is that short words behave categorically: monosyllabic roots
choose one variant, bisyllabic roots another; trisyllables (and longer roots)
display bewildering variation which however turns out to be tightly controlled
by phonology: (i) light root-final syllables permit variation, heavy
root-final syllables do not; (ii) roots ending in the maximally sonorous /a/
strongly prefer one suffix type, roots ending in the minimally sonorous /i/
strongly prefer another suffix type, and the intermediary /o/ permits both
types to a roughly equal degree.
The claim is that categorical outputs, variable outputs and statistical
preferences follow from syllable prominence defined as a combination of
stress, weight and sonority. Assuming standard Optimality Theory, I construct
a simple grammar of syllable prominence which predicts in what circumstances
variation will and will not occur and which variant is preferred. Both
categorical cases and variation are derived from one partially ranked grammar
which yields multiple totally ranked grammars. The number of total rankings
by which a variant wins predicts its probability of occurrence.
____________
LOGIC LUNCH
on Friday, 10 March
12:00 noon, Building 380, Room 383-N
A New Path to Pseudorecursive Varieties
Benjamin Wells
U San Francisco Mathematics and Computer Science
A variety (really, its equational theory) may be recursive (or "decidable") as
in the case of the variety of all commutative semigroups, or nonrecursive
("undecidable"), as in the case of relation algebras. Intermediate between
these two examples is the pseudorecursive variety, whose theory is not
recursive, but
(*) for every n the set of equations in the theory with n variables is
recursive (for relation algebras, even the variable-free equations
in the theory form a nonrecursive set).
Informal reasoning may suggest any variety with the property (*) is decidable,
but pseudorecursive varieties can be shown to exist in abundance, even with
strict logical and algebraic limitations. Their existence urges that language
be used precisely and presents opportunities for an expanded appreciation of
decidability.
The existence of pseudorecursive varieties is in fact easily shown.
Construction of finitely based ones is a harder matter and the principle goal
of this talk. The major concrete tool is a Turing machine that records past
behavior and periodically reviews this history for accuracy, and so is termed
a recapitulator. An abstraction of this action is so closely coded by
equations in an expanded theory of semigroups, that the n-variable parts
correspond to bounded machine resources and are thus recursive, but the entire
theory matches the machine's unfettered operation, chosen to be nonrecursive.
____________
SEMINAR ON HUMAN-COMPUTER INTERACTION
on Friday, 10 March
12:30 p.m., Terman Auditorium
Visualizing Large Information Structures Using
Focus+Context Techniques
Ramana Rao and John Lamping
Xerox PARC
<{lamping,rao}@parc.xerox.com>
In the last few years, Information Visualization research at PARC and
elsewhere has explored the application of interactive graphics and animation
technology to visualizing and making sense of larger information sets than
would otherwise be practical. A common strategy of this research has been the
use of Focus+Context (or Fisheye) techniques, in which detailed views of
particular parts of an information set are blended in some way with a view of
the overall structure of the set and operations for dynamic manipulation are
provided. In this talk, we will present an account of focus+context
techniques and then spend more time on two particular visualizations for
dealing with large tables and hierarchies respectively: the Table Lens and the
Hyperbolic Tree Browser.
RAMANA RAO has worked at the Xerox Palo Alto Research Center (PARC) since
1986. His research has focused primarily on user interfaces for information
access and visualization, paper user interfaces and document imaging,
object-oriented programming and window systems. He is one of the principle
designers of the Common Lisp Interface Manager, a user interface programming
interface standard for Common Lisp. Prior to joining PARC, Ramana worked at a
startup company that developed presentation graphics applications for the IBM
PC and a consulting company that designed and built a fault-tolerant file
server for a major minicomputer company. Ramana received his BS and MS
degrees in computer science and engineering from MIT.
JOHN LAMPING has been working at PARC since he received his PhD from Stanford
in 1987 (supervised by Terry Winograd). He works on separating out various
issues that are typically entangled in computer science. His dissertation was
about allowing first class objects to retain some free parameters without
operations on the objects having to be aware of that fact. He's worked on
substitution without copying, closures without a commitment to implementation,
methods without a commitment to classes, natural language semantics without a
commitment to syntax, and graphics without a commitment to Euclidean space.
____________
PHILOSOPHY COLLOQUIUM
on Friday, 10 March
3:15 p.m., Building 60, Room 61-H
Sovereign Performatives in the Contemporary Scene of Utterance:
Speech and Conduct in the Contemporary Politics of Representation
Judith Butler
UC Berkeley Rhetoric & Comparative Literature
<jbutler@orion.oac.uci.edu>
Recent efforts to regulate hate speech on campus, the workplace, and other
public spheres rely on a conception of language as an injurious form of
action. Some legal scholars rely on Austin's conception of the performative
character of language to argue that language not only acts, but has the power
to effect certain injurious effects. Significantly, this notion is used to
support certain descriptions of hate speech and pornography, and has become
central to arguments in favor of heightened regulation. This paper considers
which view of the performative is at work in such arguments, and suggests that
an alternative view might both account for the injurious and even traumatic
character of some utterances and representations and for the possibility of
their recontextualization. The paper appreciates arguments against first
amendment absolutism, but suggests that the cultural restaging and
resignification of injurious speech is a more productive way to counter its
injurious effects than the route of regulation.
____________
LINGUISTICS COLLOQUIUM
on Friday, 10 March
3:30 p.m., Cordura Hall, Room 100
Speakers, Perspectives, and Words in Acquisition
Eve Clark
Stanford Linguistics
<eclark@psych.stanford.edu>
Speakers mark their perspective on an object, an event, an activity, or a
relation through the words they choose. The range speakers can draw on
include different levels in hierarchical taxonomies (_animal_, _mammal_,
_dog_, _spaniel_; _meal_, _lunch_; _ingest_, _eat_, _chew_) and orthogonal
domains (_teacher_, _cellist_, _parent_, _gardener_). Diversity in the
lexicon offers choices to speakers in representing their perspectives to
addressees. One consequence is that the same object or the same event may be
referred to in a number of ways on different occasions, where each expression
reflects a different perspective and hence a different goal on the part of the
speaker. There is no simple one-to-one relation between terms used for making
reference and their intended referents. This talk explores some consequences
for early lexical acquisition of such input.
____________
SYNTAX SEMINAR
on Monday, 13 March
2:15 p.m., Cubberley Hall, Room 229
Determiner Quantifiers and Focus
Helen de Hoop
University of Groningen
There is an ongoing debate in the literature on the kind of part focus plays
in the determination of the domain of quantification of different types of
quantifiers, such as adverbs, determiners, and "only" (see, e.g., Partee
1991). In this talk I will mainly be concerned with determiners, for which it
is not so clear yet to what extent focus contributes to the determination of
the quantificational structure.
With respect to quantificational adverbs, it is well-known that they are
rather flexible in their argument selection and that focus can function as a
guide in this process (cf. De Swart 1991). In the case of focus-sensitive
operators, such as "only", it can also be argued that the domain of
quantification is restricted by the set of alternatives to the focussed
constituent (cf. Rooth 1992). Yet, Vallduvi (1991) shows that strong
contextual pressure can force a reading where the argument of "only" is not
provided by focus, which supports his position that focus is a real
information-packaging primitive that has no place in semantics. He supplies a
similar argument in the case of adverbs of quantification.
An issue that has been brought up in the literature with respect to domain
selection of determiners concerns the fact that focus at an adverbial phrase
in the second argument of a determiner (normally given by the VP) will force
the non-focussed rest of the VP to get interpreted within the first argument,
the restrictor. This process can even give rise to truth-conditional
differences. This can easily be explained if we make use of the context set
variable that is present in the restrictor of determiners (Westerstahl 1985).
More specifically, in De Hoop and Sola (in prep.), we hypothesize that focus
can determine the context set variable "X" in a uniform way. A focussed
constituent supplies a C-set (following Rooth 1992). Our claim is that "X"
can be equated with the generalized union over the C-set of a focussed
constituent (see also Geilfu 1993).
It was also pointed out in the literature that focus within the first argument
part of the determiner can never make any truth-conditional differences. This
falls out nicely from our analysis, as I will show. We then have to account
for certain examples with "few" and "many" for which it has been argued in the
literature that focus in the first argument does have a truth-conditional
import (Herburger 1992). I will argue that Herburger's focus-affected reading
is nothing but an ordinary cardinal reading, which can be accounted for by
taking into consideration the vague, heavily context-dependent nature of "few"
and "many" (see also De Hoop and Sola, in prep.). Herburger's focus-affected
reading will turn out to be a special case of a cardinal reading, where the
function of focus is to express contrastiveness within contextually given
alternatives.
____________
PHONOLOGY WORKSHOP
on Tuesday, 14 March
7:30 p.m., 306 San Mateo Drive, Menlo Park
Only the Trochee
Ruben van de Vijver
Vrije Universiteit
<reuben@csli.stanford.edu>
In this talk I will argue that surface-iambs result from violations of the
constraint TROCHEE and that there is no constraint which requires that feet
are iambs.
Most theories of feet use two types of feet. A foot is either trochee (a
leftheaded foot) or it is an iamb (a rightheaded foot). In parametric
theories (cf. Hayes 1981, 1985, 1987, 1995) the choice of the type of foot was
seen as a parametric choice. The rise of Optimality Theory (Prince and
Smolensky 1993, McCarthy and Prince 1993ab, 1994) has put this understanding
in another light: Variation is seen as the result of constraint interaction.
As a consequence, variation should not be present in the constraints
themselves. This means that constraints should not contradict each other.
For if they do, their ranking is based on the content of the contradicting
constraints and not on the surface patterns.
A typical example of contradicting constraints are the foot type constraints
(Prince and Smolensky). One foot type constraint requires that feet are
trochees (leftheaded) and another constraint requires that feet are iambs
(rightheaded). I will argue that there is no need for a constraint that
requires feet to be iambic. Rightheaded feet on the surface are the result of
the violation of the constraint TROCHEE. Three iambic languages, Weri,
Hixkaryana and Chugach Alutiiq Yupik are used to illustrate this point. The
interaction of familiar constraints and concepts like NONINITIALITY, *CLASH,
*LAPSE and TROCHEE will give rise to iambic patterns. As a consequence the
foot-inventory in the interesting part of the grammar, the set of universal
constraints, will consist of one foot: the trochee.
Finally, the issue of iambic lengthening is addressed. I will argue that this
is the consequence of a tendency to be trochaic. Rhythmic considerations such
as clash- or lapse-avoidance enhance lengthening in rightheaded feet and
shortening in leftheaded feet (Prince 1990, Kager 1993).
____________
CSLI TINLUNCH
on Thursday, 16 March
12:00 noon, Cordura Hall, Room 100
Vagueness, Empty Names, and Supervaluations
Ernest Lepore
Rutgers Philosophy
<lefore@zodiac.rutgers.edu>
Sentences with vacuous names and empty predicates are common, perhaps more
common than we know. Even more evident (and inclusive) are sentences whose
presuppositions fail; and there seems to be no shortage of sentences with
vague predicates (by some author's lights -- Kamp and Partee (ms) -- "all of
them"). What ties these sentences together is that many of them are supposed
to be truth valueless. That on its face doesn't seem so bad, but many
apparently worry that once truth-value gaps are introduced into the semantics,
our beloved classical logic will have to go, in particular, the law of
excluded middle; suppose P is some sentence with either a vacuous name, empty
predicate, failed presupposition, or vague predicate (with a subject that
doesn't definitely fall in or out of its extension) and on this basis let us
suppose P without truth value. Then, presumably, so is Not-P, but then so is
[P v Not-P], and [P and Not-P]. Purists can't bear this load. Also,
providing a semantics for gappy language is not a easy task. And then there
are the paradoxes that seem to follow certain kinds of truth-value gaps.
With respect to problems about non-denoting singular terms, supervaluation
techniques have gone pretty much without criticism and have become something
of a status quo position. Supervaluation theory fares less well in
discussions about vagueness. Though it has its proponents, there is
substantial criticism in print. Still, even in the critical literature I know
of no challenge to the intelligibility of applying supervaluation techniques
to vague language and that's exactly what I want to do. If I'm right, then
there is little reason to believe supervaluation techniques have any obvious
philosophical utility. My argument hinges (mostly) on the fact that all the
known uses I'm familiar with are incoherent. In and of itself, this doesn't
impugn the intelligibility of supervaluations as mathematical objects of
study; but that's hardly a defense of its philosophical utility.
____________
CSLI SEMINAR
SERIES ON VISUAL PROGRAMMING
on Thursday, 16 March
2:15 p.m., Cordura Hall, Room 100
Visual AppBuilder
Kurt Schmucker
Apple Computer
<schmucker1@applelink.apple.com>
The goal of Novell's Visual AppBuilder product (formerly the Serius Workshop
from Serius Corporation) is to remove the programmer from the development
process for the majority of the desktop application needs of corporations.
They have tried to achieve these aims by designing a visual programming
environment in which large-grain chunks of functionality can be quickly and
easily crafted together to produce a custom application. This talk will
present this environment, examples of what can and cannot be done with it, and
an examination of its strengths and weaknesses. Demonstrations of the
environment and of applications written with it will be given.
KURT SCHMUCKER has been at Apple for over seven years and has lead a variety
of projects in its Advanced Technology Group, including the Scientific
Computing Project and portions of the Human Interface Group. Currently he is
using Prograph CPX to accelerate the transition of ATG research into Apple
products. Prior to Apple, he worked for a decade in the US Department of
Defense as a computational linguist, computer scientist, and research
mathematician.
Kurt is the author of three books, including _Object-Oriented Programming for
the Macintosh_, and is a frequent contributor on Prograph topics in both
Visual News and the MacTech Magazine. He has advanced degrees in both
mathematics and computer science. The brand of shame ("ABD") is on his
forehead, but has faded somewhat with the passing of the years.
____________
SYMBOLIC SYSTEMS FORUM
on Thursday, 16 March
4:15 p.m., Building 60, Room 61-F
AI-Based Planning for Image Processing Tasks
Amy L. Lansky
NASA Ames Research Center
<lansky@ptolemy-ethernet.arc.nasa.gov>
This talk will describe a nontraditional domain-independent planner, COLLAGE,
and its application to the the task of generating image processing plans. In
this domain, COLLAGE can be viewed as a software assistant or advisory agent
that helps Earth scientists working with remotely-sensed imagery. We believe
that this kind of task is representative of a class of fruitful domains for
planning researchers: helping humans navigate through seas of software- and
data-selection possibilities. In such domains, the human user has deep
knowledge of their domain and what they're trying to accomplish, but the
available tools and data at their fingertips may be too vast or complex (e.g.,
there may be hundreds of image processing algorithms available). The planner
helps to bridge this gap by possessing knowledge about the various algorithms
and data sets, what their requirements and purposes are, and how they can be
composed to meet task requirements.
In contrast to traditional state-based planners that use techniques based on
STRIPS action descriptors, COLLAGE uses a unique action-based planning
methodology. A suite of algorithms are utilized that focus on action
decomposition, temporal/causal relationships between actions, and propagation
of CSP-based constraints on parameter variables. One focus of this talk will
be on the utility of this type of planning for realistic domains.
DR. AMY LANSKY is a Senior Computer Scientist in the former Artificial
Intelligence Research Branch (currently Computational Sciences Division) at
NASA Ames Research Center. Since 1989, she has led the design and development
of the COLLAGE planning system. From 1983-1989, Dr. Lansky was a research
computer scientist in the Artificial Intelligence Center at SRI International.
At SRI she developed the GEMPLAN planner and also collaborated in the
development of the reactive planner PRS and its application to diagnostic
tasks for Space Shuttle.
Dr. Lansky received her Ph.D. in Computer Science from Stanford University in
1983 in the area of concurrent program verification and distributed systems.
She has been an NSF fellow, a Hertz fellow, and has been nominated as a AAAI
fellow. In computer science, her research interests include planning and
scheduling, agents, distributed systems, and programming languages. Outside
of computer science, Dr. Lansky has an avid interest alternative healing
systems and their mechanisms. She is also lead singer of the local rock and
roll band, "The Wizards."
____________
SEMINAR ON HUMAN-COMPUTER INTERACTION
on Friday, 17 March
12:30 p.m., Terman Auditorium
Implicit Structures for Pen-Based Systems
Within a Freeform Interaction Paradigm
Tom Moran
Xerox PARC
<moran@parc.xerox.com>
____________
LINGUISTICS COLLOQUIUM
on Friday, 17 March
3:30 p.m., Cordura Hall, Room 100
The Once and Future Dictionary
Geoff Nunberg
Xerox PARC and Stanford Linguistics
<nunberg.parc@xerox.com>
The Dictionary is at once the most conservative and conventionalized of print
genres, to the point where it has "grown a capital letter," and represents a
kind of ideal type. In this talk I want to start by considering the
Dictionary as a descriptive template that implies a number of claims about the
lexicon. For example, it entails a picture of compositionality as an absolute
rather than gradient phenomenon. It suggests a "Gricean" view of the relation
of lexical to encyclopedic information, where the latter is a kind of
contextually added inference. It implies a uniformitarian understanding of
the meaning relation, and it supports a notion of the lexicon as a system of
elements that participate in a single, if messy, semantic hierarchy. These
and other implications of the Dictionary form have had a considerable
influence on popular and theoretical conceptions of the lexicon and on the way
that people have tried to use the Dictionary as a model for constructing
lexical knowledge representations. And yet most of these claims are
ultimately grounded neither in empirical reality nor the "informational" needs
that lexicographers invoke when they explain their enterprise. As I'll try to
show here, they follow instead from the symbolic role of the Dictionary, by
which I mean not its prescriptive value (which is slight in any case), but its
role as the embodiment of a certain picture of how discourse operates in a
print culture, as "the book written by books." Finally, I want to argue that
the "on-line" dictionary is in a certain sense a contradiction in terms: we
should think of electronic lexical systems and utilities not simply as
improved dictionaries, but as a different sort of representation entirely.
____________