[Date Prev][Date Next][Date Index]
CSLI Calendar, Wednesday, 20 August 2003, vol. 18:47
CSLI CALENDAR OF PUBLIC EVENTS
______________________________________________________________________
20 August 2003 Stanford Vol. 18, No. 47
______________________________________________________________________
A weekly publication of the
Center for the Study of Language and Information (CSLI)
Stanford University, Ventura Hall, Stanford, CA 94305-4115
http://www-csli.stanford.edu/
____________
ACTIVITIES FROM 20 AUGUST 2003 TO 29 AUGUST 2003
THURSDAY, 21 AUGUST 2003
11:00am Music 319: CCRMA Hearing Seminar
CCRMA Library, The Knoll
"Assessing Cognitive Load with Speech"
Anthony Jameson
German Research Center for Artificial Intelligence
http://mambo.ucsc.edu/psl/ccrmas/ccrmas.html
Abstract below
4:00pm PARC Forum
George Pake Auditorium at PARC
"Making the World Wide Web Fit People"
Polle Zellweger
http://www.parc.com/forum/
Abstract below
MONDAY, 25 AUGUST 2003
2:00pm CSLI Talk
Ventura 17
"How To Generate One-Anaphors"
Robert Dale
Centre for Language Technology, Macquarie University
Abstract below
TUESDAY, 26 AUGUST 2003
1:00pm MURI intern presentations
Ventura 17
http://www-csli.stanford.edu/semlab/muri/
Information below
WEDNESDAY, 27 AUGUST 2003
3:00pm Special University Oral Examination
Gates 400
"Algorithms for Data Stream Systems"
Mayur Datar
Computer Science
http://www.stanford.edu/~datar/
Abstract below
____________
Stanford Blood Center status: shortage of all types. For an
appointment: http://bloodcenter.stanford.edu/ or call 650-723-7831.
It only takes an hour of your time.
____________
ANNOUNCEMENT
http://news-service.stanford.edu/news/2003/august6/philtalk-86.html
Philosophy professors Kenneth Taylor and John Perry [who is also CSLI]
may be on their way to becoming radio's next 'Click and Clack' with
their new show "Philosophy Talk". Rather than solving callers'
automotive conundrums, however, they will be tackling bigger
questions. What is Justice? Can machines be programmed to think? Or in
the case of the upcoming pilot, is lying always bad? The show airs
August 20 at 1 p.m. on KALW (91.7 FM).
____________
MUSIC 319: CCRMA HEARING SEMINAR
on Thursday, 21 August 2003, 11:00am
CCRMA Library, The Knoll
http://mambo.ucsc.edu/psl/ccrmas/ccrmas.html
How do you tell how busy somebody is? Even better, how do you tell
how busy their brain is? How much brain power is left for a new task?
Is now a good time to interrupt somebody? How can a machine make this
judgment?
Many people suggest the talking on a cell phone while driving is
dangerous because the remote party can't tell how busy the driver is.
As interactive voice response units become more common, our machines
should tailor their interactions to the user's state. But how does a
machine tell how busy the user is, or how much brain power is left for
a new task?
Tony Jameson will be leading the discussion at next Thursday's CCRMA
Hearing Seminar. He's been doing some nice work on measuring
cognitive load based on how people speak. In simulations, a user is
given a task and then a separate task requiring the user to speak. By
measuring the quality of the speech, with simple measures, he can make
predictions of the user's cognitive load.
With the academic and commercial interest in telephony and speech
recognition here in Silicon Valley, I hope this will be a good
discussion.
Bring your ears and speech disfluencies to CCRMA, and we'll make sure
the cognitive load is interesting.
- Malcolm
"Assessing Cognitive Load and Time Pressure
on the Basis of Features of Speech"
Anthony Jameson
German Research Center for Artificial Intelligence
In two experiments involving a total of 64 subjects, we investigated
the prospects of assessing the current cognitive load and time
pressure of a speaker on the basis of features of the speech. The
experimental setting simulated a situation in which a traveler in an
airport asks questions by phone to a human assistant (e.g., "I need to
get some Mexican Pesos. Is there anywhere near here where I can change
money?"). Two independent variables were manipulated: (a)
"Navigation": whether the subject, while formulating the question, had
to move around the simulated airport terminal; and (b) "Time Pressure
for Speaking": whether the subject was instructed and motivated to
complete her utterances quickly (as opposed to aiming for maximum
clarity). In the second experiment, a further distraction was added in
the form of recorded flight announcements that were played
continuously.
The 5120 resulting utterances were transliterated and coded according
to a number of features, including: number of syllables; articulation
rate; onset latency; frequency and duration of filled and unfilled
pauses; and several other types of disfluency, such as false
starts. Multivariate analyses of variance revealed a number of
statistically significant and interpretable effects of the
experimental manipulations on the speech features. To see whether it
was possible to recognize a speaker's current cognitive state on the
basis of these features, we trained dynamic Bayesian networks on the
basis of the experimental data and checked how well a network can
recognize which experimental condition a speaker was in, given only
the features of one or more utterances. Average recognition accuracies
of roughly 65% on the basis of several utterances were
obtained. Analyses also revealed the relative diagnostic values of the
different features.
About the speaker: Anthony Jameson (http://dfki.de/~jameson/) is a
principal researcher at DFKI, the German Research Center for
Artificial Intelligence and an adjunct professor of human-computer
interaction at the International University in Germany. He earned a
B.A. at Harvard College and a PhD at the University of Amsterdam. For
25 years he has conducted multidisciplinary research on various
aspects of user modeling in interactive systems.
____________
PARC FORUM
on Thursday, 21 August 2003, 4:00pm - 5:00pm
George Pake Auditorium, PARC
http://www.parc.com/forum/
"Making the World Wide Web Fit People:
Traveling in Comfort on the Information Superhighway"
Polle T. Zellweger
MacZell Consulting
The World Wide Web represents significant progress toward Douglas
Englebart's vision of augmented human intellect from the 1960's.
People can now accomplish a wide range of information tasks online,
ranging from searching and browsing to reading, annotating,
collecting, organizing, and authoring. Unfortunately, current
information tools are often clumsy, making information tasks
unnecessarily laborious and reducing the quality of their results.
To fully realize the promise of the Information Revolution,
information tools should fit people well. They should be useful,
fitting smoothly into people's work practices. They should be easy to
use, streamlining tasks and reducing interruptions, and they should be
comfortable, matching and exploiting human capabilities.
This talk presents examples of information tool design that illustrate
the value of a central focus on these three elements of good fit.
These examples also provide a direction for further research and
development of this kind, aimed at making our 'information surround' a
useful, easy, and comfortable place for people to live, work, and
play.
About the Speaker: Polle Zellweger received her PhD in computer
science from the University of California at Berkeley, focusing on
interactive source-level debugging for optimized programs. She was a
member of the research staff at Xerox PARC from 1984 to 2001, where
she explored topics that included hypertext, multimedia, electronic
books, user interfaces and collaborative work. In 2000-2001, she was
a visiting professor at the University of Aarhus in Denmark. She has
served on the editorial board of the ACM Transactions on Information
Systems and as program chair of the ACM Multimedia95 Conference.
She is known for her work on hypertext paths, active multimedia
documents, automatic temporal formatting, and Fluid Documents.
____________
CSLI TALK
on Monday, 25 August 2003, 3:30pm
Ventura 17
"How To Generate One-Anaphors"
Robert Dale
Centre for Language Technology, Macquarie University
The problem of referring expression generation -- how to choose the
information content of an expression that picks out an intended
reference -- is a well-explored one in the natural language generation
literature. The coverage is patchy, however: for example, the
generation of references to entities by means of one-anaphoric
expressions, such as 'the small green one', is almost completely
unexplored. In this talk, I'll explore the processes that might be
involved in this generation task. It turns out that fitting this task
into the standard NLG architecture isn't easy; I'll use the case of
one-anaphora to argue for a different approach to NLG architecture,
and speculate on how this might have more general impact.
About the Speaker: Robert Dale is author or editor of five books and
around 60 papers in various aspects of natural language processing.
His current research interests include low-cost approaches to
intelligent text processing tasks; practical natural language
generation; the engineering of habitable spoken language dialog
systems; and computational, philosophical and linguistic issues in
reference and anaphora. He is editor of Computational Linguistics,
and Director of the Centre for Language Technology at Macquarie
University in Sydney, Australia.
____________
MURI INTERN PRESENTATIONS
on Tuesday, 26 August 2003, 1:00pm
Ventura 17
http://www-csli.stanford.edu/semlab/muri/
The ONR-supported MURI project on tutorial dialogue announces
presentations by its summer interns from the Symbolic Systems Program.
The talks will be in Ventura 17, starting at 1pm with a quick
introduction to the Navy damage control domain, then continuing with
approximately 20 minute presentations by each intern, as follows:
Carina Koo
"Student Modeling for SCoT-DC"
This summer, I've explored issues in student modeling by looking at
student modeling methods in several different intelligent tutoring
systems, and by exploring ways of modeling knowledge. For our tutor,
SCoT-DC (Spoken Conversational Tutor for Damage Control), the student
modeling component needs to monitor student knowledge between practice
scenarios, preserve this information for the tutor, and again monitor
student knowledge during the subsequent tutoring session. I will be
presenting an outline of the student modeling framework we've settled
on, as well as an overview of some issues specific to spoken tutoring
systems.
Jean Bogart
"An Architecture for Natural Language Generation in SCoT-DC"
In order to afford dynamic generation of dialogue in SCoT-DC, we had to
construct a means of representing and shaping the overall discourse
structure (incorporating elements of Rhetorical Structure Theory),
individual tutor utterances, and the domain entities referenced within.
This involved taking into account that a conversational tutoring system
should strive to ameliorate repetition in vocabulary and sentence
structure, as well as adapting its mode of speech to the student's
parameters (e.g., whether s/he is a Navy or Stanford user). I will outline
the architecture we have developed to support these needs within the
SCoT-DC framework, with notes on how the tutor could implement, in future,
more effective facilities for referring expression generation and resolving
referents in user dialogue.
Adrian De La Mora
"Gesture Enabled Ship Display"
In order to enhance the effectiveness of a computer-based tutor, it is
important for the tutor and student to be able to interact using both
gesture and speech. My internship, this summer, was aimed at enabling
a robust set of gestural features (both input and output) in the
tutor's current Ship Display. The main points I will address in my
presentation are:
- the various affordances and features implemented
- the graphics architecture behind the display
- the tools I used (Open Inventor API for general 2D-3D visualization).
- the problems and solutions in the implementation
- a look into the future
Trevor Austin (joint work with Mahesh Srinivasan)
"Experimentally Assessing Preferences for Graphical Indicators"
To discover the mental rules that guide the interaction of graphical
indicators like arrows and circles with various kinds of images, we
have prepared a series of experiments where users rank their
preferences and add their own indicators to photos. I describe the
challenges of preparing this experiment for the web and the graphical,
logistical, and methodological decisions involved. If you would like
to try out a beta version of the experiment, you can visit
http://www.stanford.edu/group/languagelab .
____________
SPECIAL UNIVERSITY ORAL EXAMINATION
on Wednesday, 27 August 2003, 3:00pm
Gates 400
Refreshments served at 2:45pm
http://www.stanford.edu/~datar/
"Algorithms for Data Stream Systems"
Mayur Datar
Computer Science, Stanford University
In a growing number of information processing applications, data takes
the form of "continuous data streams" rather than traditional stored
databases. These applications share several distinguishing features
like the need for real time analysis, huge volumes of data, and
unpredictable and bursty data arrivals. These applications have
spawned a considerable and growing body of research into data stream
processing, ranging from algorithms for data streams to full-fledged
data stream systems. In this talk, I will present some of my work in
this area.
In the first part of my talk, I will present algorithms for the
"sliding window model". In the sliding window model, at any given time
only the last N elements (window size) of the data stream are
considered pertinent for answering queries. We have developed a novel
data structure called Exponential Histograms that can be used to solve
a large class of problems in the sliding window model efficiently,
namely using small space and query/update time. I will explain this
data structure with the help of a simple problem: Count the number of
1's from the last N data elements of 0's and 1's.
In the second part, I will talk about some of the work that I have
done in actually building a general purpose data stream management
system (DSMS) at Stanford. My focus has been on developing algorithms
for runtime resource allocation problems in a DSMS. I will talk about
one such algorithm for online operator scheduling in DSMS that
minimizes the total memory usage of the system.
____________
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
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 (i.e., no abstracts). 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/Archive/calendar/current.shtml
and the archives at
http://www-csli.stanford.edu/Archive/calendar/
People on most of the CSLI computers can type 'help csli-calendar' to
see the current issue.
The CSLI Calendar is also posted each week to
news://nntp-csli.stanford.edu/csli.bboard.
and
news://news.stanford.edu/su.events
Information about CSLI's research program is available at
http://www-csli.stanford.edu/
For maps to the Stanford University campus see
http://www.stanford.edu/home/visitors/maps.html
____________