[Prev][Next][Index][Thread]
CSLI Calendar, 12 November 1997, vol. 13:9
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
______________________________________________________________________
12 November 1997 Stanford Vol. 13, No. 9
______________________________________________________________________
A weekly publication of the
Center for the Study of Language and Information (CSLI)
Stanford University, Ventura Hall, Stanford, CA 94305-4115
____________
ACTIVITIES DURING 12 NOVEMBER TO 21 NOVEMBER 1997
WEDNESDAY, 12 NOVEMBER
12:15pm Developmental Brownbag
Jordan Hall 420:286
Elena Lieven
University of Manchester
3:45pm Psychology Department Colloquium
Jordan Hall (420:050)
Relational reasoning in a physical symbol system
Keith Holyoak
UCLA
4:15pm EE380: Computer Systems Laboratory Colloquium
Gates B03 (NEC Auditorium)
PlaceWare Web-based Collaborative Apps Made Simple
Pavel Curtis
PlaceWare, Inc.
Abstract below
THURSDAY, 13 NOVEMBER
4:00pm Xerox PARC Forum
George Pake Auditorium, Xerox PARC
How the Direct Methanol Liquid-Feed Fuel Cell Works,
and Prospects for Ubiquity
Gerald Halpert
JPL, California Institute of Technology
4:15pm Seminar on Computational Learning and Adaptation (SCLA)
Gates 100
Bias, Variance, and Error Correcting Output Codes
for Local Learners on Binary Classification Tasks
David Aha
Naval Center for Applied Research in Artificial
Intelligence, Washington, DC.
Abstract below
4:15pm Math Colloquium
Math Corner 380:380W
Cutting up a Pea to make the Sun
Matthew Foreman
U.C.Irvine
Abstract below
7:30pm Stanford Phonology Workshop
Margaret Jacks 460:146
Destressing and Correspondence in the Clitic Group:
Evidence from Tohono O'odham
Colleen Fitzgerald
San Jose State University
Abstract below
FRIDAY, 14 NOVEMBER
12 noon Logic Lunch
Room 380:383N
Ideals as Axioms for Mathematics
Matthew Foreman
Math., U.C.Irvine
Abstract below
12:30pm Seminar on People, Computers, and Design
Gates B01 (HP classroom)
Primitive Man in the Electronic Work Environment
Anatol Holt
University of Milano
Abstract below
3:15pm Cognitive Seminar
Jordan Hall 420:100
How the brain changes during learning
Russ Poldrack
3:15pm Philosophy Department Colloquium
Bldg. 90:92Q
"Epicurus on Pleasure and Desire"
John Cooper
Princeton University
MONDAY, 17 NOVEMBER
3:30pm Psychology Social Lab
Jordan Hall 420:100
Naive Dialectism: Effects on Reasoning and Judgment about
a Contradiction
Kaiping Peng
University of California, Berkeley
4:30pm Stanford Digital Libraries Seminar
Gates B08
David Levy
Xerox PARC
WEDNESDAY, 19 NOVEMBER
3:45pm Psychology Department Colloquium
Jordan Hall (420:050)
Pathways to early conscience
Grazyna Kochanska
University of Iowa
4:15pm EE380: Computer Systems Laboratory Colloquium
Gates B03 (NEC Auditorium)
The AVR Family of Embedded Processors
Jim Panfil
4:15pm Theory Seminar
Gates 498
Dynamic Logic of Proofs
Sergei Artemov
Cornell University and Moscow University
Abstract below
THURSDAY, 20 NOVEMBER
12 noon CSLI CogLunch
Cordura Hall, Room 100
Alison Gopnik
Berkeley
4:00pm Xerox PARC Forum
George Pake Auditorium, Xerox PARC
Playing With Food: Three Centuries of Science in the
Kitchen
Harold McGee
4:15pm Seminar on Computational Learning and Adaptation (SCLA)
Gates 100
Learning the Structure of Probabilistic Models: The
Structural EM algorithm
Nir Friedman
U.C. Berkeley
Abstract below
FRIDAY, 21 NOVEMBER
12:30pm Seminar on People, Computers, and Design
Gates B01 (HP classroom)
Interaction with 3D Domains
William Buxton
Alias Wavefront
Abstract below
3:30pm Linguistics Department Colloquium
Bldg. 460:146
Satisfying Constraints on Extraction and Adjunction
Ivan Sag
Stanford University
Abstract below
7pm-10pm Symposia: Are Computers Approaching Human-Level
Creativity?
Law School, Kresge Aud.
Symposium IV: Creation of Jokes and Humor
Kim Binsted
Artificial Intelligence, U. of Edinburgh
Author of the computer program Jape, which concocts
original puns in various languages
Marvin Minsky
Media Lab and AI Lab, MIT Pioneering
researcher in artificial intelligence; author of The
Society of Mind
Steve Martin
Writer/actor, Los Angeles Director of
Roxanne, Father of the Bride, etc., and author of Picasso
at the Lapin Agile and other plays
____________
EE380 COMPUTER SYSTEMS COLLOQUIUM
on Wednesday, 12 November 1997, 4:15pm to 5:30pm
NEC Auditorium (B03), Gates Computer Science Building
[http://www.stanford.edu/class/ee380/]
The PlaceWare Platform:
Web-based Collaborative Apps Made Simple
Pavel Curtis
Principal Architect
PlaceWare, Inc.
[http://www.placeware.com]
PlaceWare, a recent spin-off from the Xerox Palo Research Center
(PARC), builds highly interactive, Web-based collaborative
environments that are accessible to anyone who has a Java-enabled
browser and a 28.8Kb modem. Our lead application is the PlaceWare
Auditorium, which allows one or more people to give an interactive,
online, multimedia presentation via the Web to hundreds or thousands
of simultaneous attendees; the presentation can include slides (made
in PowerPoint or any GIF-image editor), live annotation on the slide
images, real-time polls of the audience, live audio from the presenter
and those asking questions, private text and audio conversations in
the auditorium's "rows", and other features.
This talk, however, is not primarily about the PlaceWare Auditorium
product, but rather about the software infrastructure that made its
swift implementation possible: the PlaceWare Platform. This collection
of APIs, server structure, and support libraries for network
communication, data persistence, user authentication and
authorization, along with the PlaceWare "parts" library of reusable
distributed-application components, made it possible for the first
fully functional version of Auditorium to be built in just two
man-months.
In this talk, I'll discuss the PlaceWare system architecture and
facilities as they are made visible to application developers and
illustrate the use of some of these facilities with source code from
several sample PlaceWare applications. The PlaceWare Developer's Kit
(PDK), which includes a fully functional PlaceWare server, complete
API documentation, and full source code to many sample applications,
is now in limited public beta-testing; I'll close my talk with an open
invitation to join our free PDK evaluation program and start writing
the applications that can turn a Web site from a passive collection of
documents into a lively and productive populated place. For more
information about the PDK beta-test, see
http://www.placeware.com/product/download.html .
Biography: Pavel Curtis, an internationally recognized expert on
online communities, is a Principal Architect and co-founder of
PlaceWare, Inc.; his work there centers on investigating, designing,
and implementing applications and systems to support and further
develop the PlaceWare technology. The PlaceWare platform enables the
rapid development of highly interactive, multimedia, shared,
Java-based applications for large-group interaction and collaboration
seamlessly integrated with the World-Wide Web.
Pavel was a member of the research staff at the Xerox Palo Alto
Research Center (PARC) for over 13 years. At PARC, his focus was on
the design and implementation of programming languages, such as
Smalltalk, Interlisp, Cedar and Scheme. He is the founder and chief
administrator of LambdaMOO, one of the most popular recreational
social virtual realities on the Internet. His LambdaMOO server
software currently supports over 150 virtual communities on the
Internet, including the original 8,000-member LambdaMOO community.
Much of Pavel's research and development work was the foundation for
the Jupiter project at PARC, which served as the prototype for
PlaceWare's product and technology.
Pavel is a frequent lecturer around the world at seminars and
symposiums on virtual communities and Internet-based collaboration. He
has published a number of papers, articles, and book chapters on
online communities and collaboration.
Pavel was awarded an M.S. and Ph.D. in computer science from Cornell
University in 1983 and 1990, respectively, and a B.A. from Antioch
College in 1981.
____________
SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION (SCLA)
on Thursday, 13 November 1997, 4:15pm to 5:30pm
Gates 100
[http://www-csli.stanford.edu/cll/scla.html]
Bias, Variance, and Error Correcting Output Codes for
Local Learners on Binary Classification Tasks
David W. Aha
NCARAI/Naval Research Labs
Washington DC
(Collaborator: Francesco Ricci)
Error-correcting output codes (ECOCs) represent classes with a set of
output bits, where each bit encodes a binary classification task
corresponding to a unique partition of the classes. Algorithms that
use ECOCs learn the function corresponding to each bit, and combine
them to generate class predictions. ECOCs can reduce both variance and
bias errors for multiclass classification tasks when the errors made
at the output bits are not correlated. They work well with algorithms
that eagerly induce global classifiers (e.g., C4.5). However, previous
research explained that they cannot assist simple local classifiers
(e.g., nearest neighbor) because ECOCs yield correlated predictions
across the output bits. This is distressing because local learning
algorithms for classification are preferable to global classifiers for
some types of applications. We show that the output bit predictions of
ECOC-extended local learners can be decorrelated by selecting
different features for each bit. We present empirical evidence
suggesting that this combination of ECOCs, nearest neighbor, and
feature selection improves performance under some conditions. We
explain our modifications to the schemata racing algorithm that we
used for feature selection which improve its ability to retrieve good
feature subsets in this context. Finally, we perform a bias/variance
decomposition analysis for this algorithm for binary classification
tasks; it reveals that the local ECOC algorithm's performance
improvement is obtained by drastically reducing bias at the cost of
increasing variance.
____________
MATH COLLOQUIUM
Thursday, 13 November 1997, 4:15 PM
Math Corner 380:380W
Cutting up a Pea to make the Sun
Matthew Foreman
U.C.Irvine
The Banach-Tarski Paradox states that there is a decomposition of the
unit ball in ${\Bbb R}^3$ into a finite number of pieces that can be
rearranged by isometries to form two unit balls (or another ball of
any given size). Can the decomposition actually be done??
Surprisingly, the pieces don't have to be "that bad" and there are
versions of the Banach-Tarski paradox where the pieces can be well
approximated on a computer.
Precisely, there is a paradoxical decomposition by pieces that have
the property of Baire. As a corollary there is a "paradoxical
decomposition" using OPEN (!) sets. (This solves a problem posed by
Marczewski in 1930.) This is joint work with R. Dougherty.
____________
STANFORD PHONOLOGY WORKSHOP
on Thursday, 13 November 1997, 7:30pm
Margaret Jacks Hall 460:146
Destressing and Correspondence in the Clitic Group:
Evidence from Tohono O'odham
Colleen Fitzgerald
Dept. of Linguistics and Language development
San Jose State University
Hayes (1989) gives arguments, mainly from English poetry, that
motivate an additional level of the Prosodic Hierarchy, the clitic
group. The Prosodic Hierarchy argued for in Selkirk (1980) and Nespor
and Vogel (1982) does not include the clitic group. In this talk, I
give evidence for the clitic group from destressing Tohono O'odham
prosodic words. The stress pattern of prosodic words in Tohono
O'odham take one shape if the word is preceded by a stressed
proclitic, and another shape if the word is prefixed or preceded by an
unstressed proclitic. The presence of the stressed proclitic means
that the prosodic word begins with two unstressed syllables. This
pattern never appears in other environments. I argue that this
asymmetrical distribution of stress provides corpus-internal evidence
for the clitic group of Hayes (1989).
In this talk, I make two novel points, one empirical, and two
theoretical. First, I present data from Tohono O'odham where a
context word loses its initial primary stress. The number of
languages with so-called "rhythm rules" is small (and languages with
destressing rules are even rarer). By incorporating data from Tohono
O'odham, we enrich our typological sampling. This is particularly
important, as Hayes (1995) notes that destressing in a phrasal context
is rare, with cases found in French (Dell 1984) and Italian (Nespor
and Vogel 1989). Furthermore, Hayes (1995: 386) proposes the Strong
Domain Principle, which states that "no prosodic transformation may
apply to the head of a strong domain." Under his theory, the primary
stress of a word should not be deleted under destressing. However, we
see with Tohono O'odham that this is precisely what happens. The
empirical results presented here constitute a counterexample to the
model of phrasal stress argued for by Hayes (1995).
Second, the analysis of Tohono O'odham destressing leads to a
theoretical innovation, as Optimality Theory (McCarthy and Prince
1993, 1995 and Prince and Smolensky 1993) is used in this account of
stress clash. In particular, I show that Correspondence Theory
(McCarthy and Prince 1995) can be extended to formulate constraints to
evaluate the relationship between the stress patterns of words in
isolation and in larger context.
____________
LOGIC LUNCH
on Friday, 13 November 1997, 12 noon
Math Corner 380:383N
Ideals as Axioms for Mathematics
Matthew Foreman
Math., U.C.Irvine
Strong ideal properties give generalizations of large cardinals
consistent on small cardinals such as \aleph_1 and \aleph_2. They
settle the GCH, imply determinacy and solve most of the problems of
set theory. Are they suitable for axioms?
There will be a lunch with the speaker at the Stanford Coffee House
following his noon talk. The evening before there will be a dinner
out with him following his Math Colloquium lecture, announced
separately.
____________
SEMINAR ON PEOPLE, COMPUTER, AND DESIGN
on Friday, 14 November 1997, 12:30-2:00pm
Gates B01 (HP Classroom)
[http://www-pcd.stanford.edu/seminar/]
(SITN Channel E2)
Primitive Man in the Electronic Work Environment
Anatol Holt
University of Milano
mailto:anatolholt@iol.it
Primitive man is here to stay; he is nature (not civilization). As he
breathes, so he coordinates and communicates. An analysis of these
elementary human functions is essential to understanding the services
which computers and their networks can render.
Coordination has been widely viewed as a complex social phenomenon
properly treated in a multi-disciplinary way, mainly by the social
sciences. The phenomenon is certainly social and certainly complex;
but it is hard to see how a multi-disciplinary approach can yield
clear technical guidance to the utilization (and therefore
architecture and interface) of computers. In this talk an alternative
approach is considered; it is also demonstrated that this alternative
does lead to genuine guidance and news.
Biography: Anatol Holt began as a UNIVAC I programmer under John
W. Mauchly, in 1952. Together with the late William J. Turanski he
developed a programming environment for the UNIVAC family called
"Generalized Programming" - a very early example of CSCW. From 1963 to
1974, Holt was Principal Investigator under ARPA's IPTO sponsored
research called the "Information Systems Theory Project". During this
time Holt cooperated with CarlAdam Petri, (GMD, Bonn, Germany), in the
development of Petri nets.
Since 1974 Holt has been active in various industrial and university
settings. In 1979, together with Paul Cashman, Holt launched the
world's first "coordination program" on the ARPA net (Monitor Software
Trouble Reports). Some years later, Holt and venture capitalist Eli
Jacobs, founded "Coordination Technology, Inc." for the development of
a new system software platform to support distributed electronic work
environments.
In the last 6 years, Holt has lived and worked in Milano, Italy, where
(among other things) he has been active at the University of
Milano. In 4 of these years, he authored a book titled Organized
Activity and Its Support by Computer, recently produced by Kluwer
Academic Publishers. Holt holds an Ms in mathematics from MIT, and a
PhD in descriptive linguistics from the University of Pennsylvania.
____________
THEORY SEMINAR
Wednesday, 19 November 1997, 4:15 pm
Gates 498
Dynamic Logic of Proofs
Sergei Artemov
Cornell University and Moscow University
A derivation from hypotheses may be regarded as an operation on
proofs with the arguments standing for proofs of the hypotheses.
As was established recently every absolute (i.e. independent of
the formalism chosen) propositional operation on proofs is a "proof
polynomial" built from the variables and constants by three basic
operations: "." application, "+" non deterministic choice, "!"
proof checker. Constants here stand for the proofs of tautologies
and similar "simple facts" from a certain finite list.
The variety of proof polynomials, thus the entire notion of absolute
propositional provability, can be described within a framework of
finitely axiomatized Logic of Proofs (LP), which is supplied with
completeness, decidability and cut-elimination theorems.
Logic of Proofs provides an intended provability semantics for the
Goedel provability logic S4, as well as for some other constructions
in logic and theoretical computer science, including modal logic,
intuitionistic logic with its Brouwer-Heyting-Kolmogorov
interpretation, lambda calculus and modal lambda calculus. These
systems turned out to be natural fragments of LP. For example, every
derivation in S4 admits a realization of modalities by corresponding
proof polynomials, which discloses its hidden dynamic content.
Formation rules of the lambda terms are admissible inference rules
in the Horn fragment of LP. Similarly, the modal lambda terms are
nothing but proof polynomials in the Intuitionistic Logic of Proofs.
____________
SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION (SCLA)
on Thursday, 20 November 1997, 4:15pm to 5:30pm
Gates 100
[http://www-csli.stanford.edu/cll/scla.html]
Learning the Structure of Probabilistic Models:
The Structural EM algorithm
Nir Friedman
U.C. Berkeley
In recent years there has been a flurry of works on learning Bayesian
networks. Current state of the art methods have been shown to be
successful for two learning scenarios: learning both network structure
and parameters from complete data, and learning parameters for a fixed
network from incomplete data---that is, in the presence of missing
values or hidden variables. However, no method has yet been
demonstrated to effectively learn network structure from incomplete
data.
In this talk, I will describe a new method for learning probabilistic
models from incomplete data. This method is based on an extension of
the Expectation-Maximization (EM) algorithm for searching over models.
(In contrast to standard EM that performs parametric optimization.) I
will describe the general framework of this algorithm and prove its
convergence, I then will describe how to adapt it for learning
Bayesian networks from incomplete data, and show experimental results
that demonstrate the effectiveness of this procedure.
____________
SEMINAR ON PEOPLE, COMPUTER, AND DESIGN
on Friday, 21 November 1997, 12:30-2:00pm
Gates B01 (HP Classroom)
[http://www-pcd.stanford.edu/seminar/]
(SITN Channel E2)
Interaction with 3D Domains
William Buxton,
Alias Wavefront
mailto:buxton@aw.sgi.com
Biography: Bill Buxton is a computer scientist specializing in human
aspects of technology, human-computer interaction (especially human
input to computer systems), and computer supported collaborative work
(Telepresence). He is head of User Interface Research at Alias |
Wavefront Inc., a division of Silicon Graphics Inc. (SGI), , Chief
Scientist for SGI, and an Associate Professor in the Department of
Computer Science at the University of Toronto, where his research is
mainly sponsored by the Information Technology Research Institute of
Ontario (ITRC).
Buxton began his career in music, having done a Bachelor of Music
degree at Queen's University. He studied and taught at the Institute
of Sonology, Utrecht, Holland, for two years. After completing an
M.Sc. in Computer Science on Computer Music at the University of
Toronto, he joined the faculty as a lecturer. His early work in
designing and using computer-based tools for music composition and
performance is what led him into the area of human-computer
interaction.
In addition to his university research, Buxton had a strong connection
to industry and applied work. In particular, he had a long association
with to Xerox PARC as a consulting research scientist. He joined Alias
| Wavefront Inc. in June of 1994. In 1995, Buxton became the third
recipient of the Canadian Human-Computer Communications Society Award
for contributions to research in computer graphics and human-computer
interaction.
____________
LINGUISTICS DEPARTMENT COLLOQUIUM
on Friday, 21 November 1997, 3:30pm
Margaret Jacks Hall, Room 460:146
[http://www-linguistics.stanford.edu/colloq/colloq.html]
Satisfying Constraints on Extraction and Adjunction
Ivan Sag
Stanford University
Based on work done with
Gosse Bouma (Rijksuniversiteit Groningen)
and Rob Malouf (Stanford University)
In the tradition of G/HPSG, extraction has been treated via
inheritance of values for the feature SLASH from gap, up along the
`extraction path' to the filler. This analysis is attractive because
it makes information about extractions accessible for local selection.
One could view it as a prediction of such an analysis that there are
languages registering extraction through verb morphology (Chamorro,
Palauan), the possibility of verb-inversion (French, Icelandic), tone
downstep suppression (Kikuyu) etc. However, under past proposals, not
all verbs in an extraction domain are `slashed'. Gazdar (81) and
Pollard and Sag (94) agree that verbs whose subject has been extracted
are unslashed, and in P&S-94 adv-initial clauses are unslashed as
well. This distribution of SLASH specifications is inconsistent with
the results of Hukari and Levine (95), who show that languages
registering extraction effects also show them in subject and adverb
extraction.
This paper develops a new extraction analysis that avoids this
inadequacy. Lexemes specify a verb's dependency structure, but not its
valence or SLASH value. Every dependent is realized as a SUBJ, COMPS,
and/or SLASH member. A word's SLASH value must also be the union of
the SLASH values of its dependents. Thus a verb's object can either be
a nongap element on its COMPS list or else a gap in the verb's SLASH
value. In either case, the verb passes its SLASH value (empty in the
former case) to its mother. Gaps may be both in SLASH and the SUBJ
value, allowing us to correctly predict the behavior of the languages
discussed by H&L. Likewise, since extractable adjuncts are also
included in the DEPENDENTS list, it follows that a verb will be [SLASH
{ADV}] just in case an adverbial modifying it is extracted. Hence we
predict H&L's observation that adverb extraction is lexically
registered.
Our account not only provides better cross-linguistic coverage than
earlier G/HPSG treatments, it also accounts for subject, complement,
and adverbial extraction via a single constraint on the relation
between lexemes and words. In this presentation, I will summarize a
number of further predictions, including a number that follow directly
from the elimination of WH-trace, the `subinding' that occurs in both
English and French, and the that-trace alternation and its analogue
(que-qui alternations) in French.
____________
END MATERIAL
The CSLI Calendar appears weekly on Wednesdays throughout the academic
year. Announcements, abstracts, and other information to appear in
the Calendar should be submitted to the editor, who reserves the right
to decide what does or does not go in the calendar
[mailto:incalendar@csli.stanford.edu].
Requests to be added to the mailing list should be sent to
[mailto:majordomo@csli.stanford.edu]. With the lines in the body of
the text of either
subscribe csli-calendar
for the long form or
subscribe csli-short-calendar
for the short form. Problems with subscribing or unsubscribing should
be sent to owner-csli-calendar@csli.stanford.edu.
The full current issue is at
[http://www-csli.stanford.edu/csli/Archive/calendar/1997-98/current.html]
and the archives at
[http://www-csli.stanford.edu/csli/Archive/calendar/].
People on most of the CSLI computers can type 'help csli-calendar' to
see the current issue. They can also type 'help quip' to see the
Linguistics Department newsletter, the Sesquipedalian,
[http://www-linguistics.stanford.edu/Linguistics/].
The CSLI Calendar is also posted each week to
[news://nntp-csli.stanford.edu/csli.bboard].
Information about CSLI's research program is available at
[http://www-csli.stanford.edu/csli/].
For maps to the Stanford University campus see
[http://www.stanford.edu/home/visitors/maps.html].
____________