[Prev][Next][Index]

CSLI Calendar, 27 April 1995, vol.10:25




        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
______________________________________________________________________________

27 April 1995                    Stanford                      Vol. 10, No. 25
______________________________________________________________________________

      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 26 APRIL -- 5 MAY 1995

  WEDNESDAY, 26 APRIL
         4:00 - Seminar on Computational Learning and Adaptation
                Cordura Hall, Room 100
                Cellular Encoding in Learning and Automatic 
                Parallel Compilation 
                Frederic Gruau, Stanford Psychology

  THURSDAY, 27 APRIL
        12:15 - CSLI Seminar
                Series on Intelligent Agents
                Cordura Hall, Room 100
                Agent-based Interoperation
                Adam Cheyer, SRI International
                Abstract below

         2:15 - CSLI Seminar
                Cordura Hall, Room 100
                Formal Properties of Natural Language and
                Linguistic Theories
                Christopher Culy, U Iowa Linguistics
                Abstract below

         4:15 - SSP Forum
                Building 60, Room 61-F
                A Model of Human Sentence Processing: Able to
                Tell the Difference? 
                Charles Lee, Stanford Linguistics
                Abstract below

  FRIDAY, 28 APRIL
        12:00 - Logic Lunch
                Building 380, Room 383-N
                Bounded Fragments of Predicate Logic
                Johan van Benthem, Amsterdam and Stanford Philosophy
                Abstract below

        12:30 - HCI Seminar
                Skilling Auditorium
                TabWorks: Articulating a Metaphor through
                User-Centered Design
                Hector Moll-Carrillo and Matthew Marsh, IDEO
                Abstract below

         3:15 - Philosophy Colloquium
                Building 90, Room 91-A
                Movers and Elemental Motions in Aristotle
                Istvan Bodnar, Ctr Hellenistic Studies, Washington D.C.

         3:30 - Linguistics Colloquium
                Cordura Hall, Room 100
                Shaped by Some Common Contingency:  Historically but Not
                Genetically Related
                Johanna Nichols, Berkeley Linguistics
                Abstract below

  WEDNESDAY, 3 MAY
         4:00 - Seminar on Computational Learning and Adaptation
                Cordura Hall, Room 100
                Finding Propositions from Numerical Data and Neural Networks
                using Multi-Linear Functions
                Hiroshi Tsukimoto, Toshiba Corporation
                Abstract below

  THURSDAY, 4 MAY
        10:00 - STASS Seminar
                Cordura Hall, Room 100
                Interpretive Grammar and Stylized Documents
                Keith Devlin, CSLI & St. Mary's College
                Abstract below

        12:15 - CSLI Seminar
                Series on Intelligent Agents
                Cordura Hall, Room 100
                Perception, Action, and Cognition in an Integrated
                Architecture for Physical Agents
                Pat Langley, Stanford Robotics Laboratory
                Abstract below

  FRIDAY, 5 MAY
        12:30 - HCI Seminar
                Skilling Auditorium
                The Future of the Internet
                Marc Andreesen, Netscape Communications Corporation

         3:15 - Philosophy Colloquium
                Building 90, Room 91-A
                Title to be announced
                Yair Guttmann, Stanford Philosophy

         3:30 - Linguistics Colloquium
                Cordura Hall, Room 100
                Grammatical Weight
                Tom Wasow, 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 Web, URL
<http://www-csli.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, 26 April
                    4:00 p.m., Cordura Hall, Room 100
     Cellular Encoding in Learning and Automatic Parallel Compilation
                              Frederic Gruau
                           Stanford Psychology
                        <gruau@psych.stanford.edu>

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 SEMINAR
                       SERIES ON INTELLIGENT AGENTS
                          on Thursday, 27 April
                    12:15 p.m., Cordura Hall, Room 100
                        Agent-based Interoperation
                               Adam Cheyer
                            SRI International
                           <cheyer@ai.sri.com>

A number of software methodologies have evolved with the goal of encouraging
modularity and reuse of code:
   
   - Parameterized subroutines provide code reuse within an application.
   - Libraries of subroutines encourage code sharing across applications.
   - Object-oriented techniques allow tailoring of library routines through
     inheritance and polymorphism.
   - Client/Server paradigms (e.g SQL databases) permit sharing 
     of data across platforms.
   - Remote Procedure Calls (RPC) enables code to be shared across platforms.
   - Distributed object technologies (e.g. CORBA, COM) allow sharing of
     tailorable code across platforms.

Agent-based interoperation can be seen as a new paradigm for software
integration, distinguished from the above approaches in that: input requests
may be specified in terms of "what", not "how"; agents can take an active
role, monitoring real-world conditions and reacting accordingly; and agents
may be seen as holding beliefs about events in the world.

At SRI International's Artificial Intelligence Center, the Open Agent
Architecture (OAA) is being developed as a means of integrating AI and classic
systems.  The architecture provides a framework in which a distributed
community of software agents work together to access systems on behalf of the
user.  In order to facilitate the user's delegating tasks to agents, the
architecture is served by a multimodal interface, combining pen, voice and
direct manipulation.

For this talk, I will describe the Open Agent Architecture, illustrate its
features through two demonstration applications, and then speak about current
and pending work.

NOTE: A complete schedule for the CSLI seminar series on intelligent agents is
available at URL <http://www-csli.stanford.edu/>.  These talks are scheduled
for 12:15 (usually) on Thursdays throughout the Spring quarter.  If you order
beforehand, we will provide a sandwich and beverage (for $4.50, paid at the
door; quarters and small bills preferred).  Send sandwich orders (turkey, ham,
roastbeef, or vegetarian) to <bocata@csli.stanford.edu> some time on or before
Tuesday of the week of the given talk.

                               ____________

                               CSLI SEMINAR
                          on Thursday, 27 April
                    2:15 p.m., Cordura Hall, Room 100
      Formal Properties of Natural Language and Linguistic Theories
                             Christopher Culy
                      University of Iowa Linguistics
                         <culy@csli.stanford.edu>

Over the past 40 years, various formal properties of natural language and
linguistic theories have been studied, most notably weak generative capacity
and time complexity.  However, the standard closure properties (union,
intersection, concatenation, concatenation closure, homomorphism, inverse
homomorphism, and intersection with a regular set) of natural language have
not been studied.  In this paper I will show that natural language is not
closed under any of these operations (and is hence an anti-AFL).  I will
further show that these non-closure facts generally follow from very simple
substantive constraints common to all linguistic theories.

                               ____________

                          SYMBOLIC SYSTEMS FORUM
                          on Thursday, 27 April
                    4:15 p.m., Building 60, Room 61-F
    A Model of Human Sentence Processing: Able to Tell the Difference?
                               Charles Lee
                           Stanford Linguistics
                         <clee@csli.stanford.edu>

There have been numerous models proposed to model human sentence parsing
(Frazier & Fodor 1978, Church 1982, Marcus 1980, Abney 1989, Jurafsky 1993).
A brief discussion of these models and their limitations and undesirable
qualities will be presented.

This talk will introduce a model for human sentence parsing (i.e., performance
rather than competence parsing) which does not have these same limitations or
undesirable qualities.  The proposed model has the basic properties of a stack
machine with non-deterministic control.  This model is capable of making fine
distinctions between various types of sentences, which are difficult for
humans to parse, which past models fail to distinguish.  For example:

  (1) a. I gave the boy who you wanted to give the books to three books.
      b. Without her donations failed to appear.
  (2) a. I crashed the car which you thought Bill gave to you up.
      b. I gave the girl who you thought Bill liked a book.
  (3) a. After the man drank the water proved to be poisoned.
      b. Without her donations to the charity failed to appear.
  (4) a. The man gave the girl a ring impressed a watch.
      b. The horse raced past the barn fell.
  (5) a. The mouse the cat the dog barked at chased is mine.
      b. men women children dogs bark at adore love are rare.

The proposed human parsing model was implemented as a computer program and
will be demonstrated at this talk, illustrating how this model is able to make
the distinctions between the above sentences.

After graduating from Stanford with a BS ME '83, and MS ME '84, CHARLES LEE
worked at the Trace R&D Center in Madison, Wisconsin, where he designed
electronic devices and computer programs to help provide access to computers
for people who have severe physical impairment.  One of his notable
accomplishments was working with Microsoft, IBM, and Apple, to provide
features such as Sticky-Keys, Mouse-Keys, and Slow-Keys for their operating
systems (e.g., Easy Access for the Mac).  Charles returned to Stanford to
pursue a PhD degree in linguistics in 1989.  His specific interests are in
natural language processing, machine translation, and sign languages.

                               ____________

                               LOGIC LUNCH
                           on Friday, 28 April
                   12:00 noon, Building 380, Room 383-N
                   Bounded Fragments of Predicate Logic
                            Johan van Benthem
                    Amsterdam and Stanford Philosophy
                        <johan@csli.stanford.edu>

This talk will be a presentation of some recent results obtained with
Andr=E9ka and N=E9meti, on decidability and other properties of
"quantifier-bounded fragments" of predicate logic.  The methods are reduction
arguments using bisimulation-based model constructions.  Via translation, our
results also imply decidability for various logics of "generalized assignment
models" for predicate logic.

Reference: "Back and Forth Between Classical and Modal Logic", to appear in
_Bulletin of the Interest Group for Pure and Applied Logics_, London and
Saarbruecken.

                               ____________

                  SEMINAR ON HUMAN-COMPUTER INTERACTION
                           on Friday, 28 April
                     12:30 p.m., Skilling Auditorium
      TabWorks: Articulating a Metaphor through User-Centered Design
                  Hector Moll-Carrillo and Matthew Marsh
                                   IDEO
                            <hector@ideo.com>

TabWorks book metaphor enhances the standard Windows user interface, providing
an alternative way to organize applications and documents in a familiar, easy
to use environment.  The TabWorks interface was designed collaboratively by
IDEO and XSoft and was based on a concept developed at Xerox PARC.  This
briefing describes how a user-centered approach affected the design of the
TabWorks user interface: how the metaphor's visualization evolved and how
interaction mechanisms were selected and designed.

HECTOR J. MOLL-CARRILLO has been designing user interfaces for the past seven
years including interactive training applications, information kiosks and
games, electronic documentation and testing applications, and interactive
corporate/educational presentations.  He has worked on interaction design for
Macintosh and Windows environments, for products like Xerox TabWorks, as well
as custom user interfaces for medical products, visual languages for future
operating systems and consumer products like the Smith Corona LS40/42 Personal
Labeler.

Before joining IDEO, Hector designed information visuals as Senior Artist for
Computer Imaging and Graphics, in Research Triangle Park, North Carolina, and
later as a Multimedia Designer and Producer for Image Associates, in Raleigh,
North Carolina.  He has also been a freelance computer graphics consultant and
an art teacher for children from kindergarten through the ninth grade.

MATTHEW MARSH has worked as a human factors consultant for the last seven
years, providing ergonomics expertise to multi-disciplinary teams.  Recently
he has contributed towards development of medical, computer and
telecommunications products.  He is presently working on an interactive
leisure device, an insulin delivery device, and a text telephone for the
hearing impaired.

Before joining IDEO, Matt was a senior designer at Davis Associates, an
ergonomics design company in England.  While there, he had responsibility for
business development and was their principle consultant dealing with new
legislation on working with Visual Display Units.  He has also worked
extensively in transportation; trains, light rail systems and cars.  Other
work included systems design and analysis, control room design, computer
interface design and product design.

                               ____________

                          PHILOSOPHY COLLOQUIUM
                           on Friday, 28 April
                    3:15 p.m., Building 90, Room 91-A
                Movers and Elemental Motions in Aristotle
                              Istvan Bodnar
             Center for Hellenistic Studies, Washington D.C.
                        and University of Budapest

                               ____________

                          LINGUISTICS COLLOQUIUM
                           on Friday, 28 April
                    3:30 p.m., Cordura Hall, Room 100
                    Shaped by Some Common Contingency:
                 Historically but Not Genetically Related
                             Johanna Nichols
                         UC Berkeley Linguistics

Demonstrable genetic relatedness requires (among other things) that the
languages in question share one or more structures whose abstract likelihood
of occurrence is significantly less than that of any one particular language
turning up on a random draw.  The critical threshold is a probability of
0.000002 (i.e., one in 500,000, assuming approximately 5000 languages on
Earth), which for all practical purposes identifies a unique individual
language. A feature meeting this critical probability threshold is a GENETIC
MARKER; it points to a single, unique ancestor and therefore more or less
suffices to prove genetic relatedness between languages.  But there are also
features which, though shy of the critical threshold and thus not demonstrably
traceable to a single individual protolanguage, nonetheless cannot be due to
either chance or universals.  Such features can be considered HISTORICAL
MARKERS, since they show that languages must have had some shared history,
even though we cannot say that that history involved shared descent.
Historical markers are most interesting when shared between language families,
for they can demonstrate some kind of connection and hence generate historical
inferences or hypotheses even when the comparative method cannot demonstrate
strict genetic relatedness.

This paper describes the statistical, geographical, and structural properties
of genetic and historical markers, presents several examples, and maps their
geographical distribution.  Defining the statistical properties requires
showing that, for low-frequency structural features like ergativity, split
intransitivity, derived intransitivity, gender classes, articles, etc., in
some of them the low observed worldwide frequency is close to the expected
frequency, while for others it is high relative to the expected frequency.
The features that are good historical markers on statistical grounds also
exhibit large-scale geographical clusterings that identify ancient vectors of
colonization, migration, and spread.  Finally, the best historical markers are
also among some of the most revealing typological features.

                               ____________

             SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
                           on Wednesday, 3 May
                    4:00 p.m., Cordura Hall, Room 100
               Finding Propositions from Numerical Data and
               Neural Networks using Multi-Linear Functions
                            Hiroshi Tsukimoto
                           Toshiba Corporation

A method for finding propositions from numerical data and neural networks is
presented. The method is based on the space of multi-linear functions, an
extension of Boolean algebra that can be made into a Euclidean space when the
domain of the functions is {0,1} or [0,1].  Using a method such as multiple
regression analysis, numerical data can be approximated by (multi-)linear
functions, which can then be approximated by (extended) Boolean functions.
Therefore, propositions can be obtained from numerical data. When the output
function is a sigmoid function, the units of neural networks after learning
can be approximated by (multi-)linear functions, which can be approximated by
(extended) Boolean functions. Therefore, propositions can be obtained from
neural networks. A few examples of this approach are presented.

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>.

                               ____________

                              STASS SEMINAR
                            on Thursday, 4 May
                    10:00 a.m., Cordura Hall, Room 100
               Interpretive Grammar and Stylized Documents
                               Keith Devlin
                       CSLI and St. Mary's College
                        <devlin@csli.stanford.edu>

This is an extension of the talk I gave jointly with Duska Rosenberg in
January.  I will outline a grammar that applies to stylized documents
consisting of both constrained syntax slot-and-filler fields and natural
language fields, which interact in complex ways.  The grammar associates to
the document situation-theoretic structures that correspond to the strategies
a reader must use in order to interpret the document.  This is work in
progress, not a polished performance of a completed project, and I am looking
for suggestions as to how to proceed next.  The talk will be self-contained,
but a basic knowledge of situation theory will be assumed (situations, types,
and constraints, and how these are used in the study of information flow).

                               ____________

                               CSLI SEMINAR
                       SERIES ON INTELLIGENT AGENTS
                            on Thursday, 4 May
                    12:15 p.m., Cordura Hall, Room 100
                 Perception, Action, and Cognition in an
               Integrated Architecture for Physical Agents
                               Pat Langley
                       Stanford Robotics Laboratory
                        <langley@cs.stanford.edu>

In this talk I will describe Icarus, an architecture that integrates sensing,
execution, and planning in sensori-motor domains.  The framework represents
control knowledge in terms of durative states and plans that specify sequences
of states.  Icarus organizes both states and plans in concept hierarchies,
which it uses to sort experience during the recognition process.  The
execution of states can occur in the absence of planning, though the latter
process can modulate execution.  Learning involves the alteration of
probabilities about sensor conditions, successor states, and state durations,
which can in turn reduce the amount of sensing during execution.  I will
consider Icarus' operation in the domains of pole balancing and flight
control, then discuss its psychological status, its relation to other
integrated architectures, and some directions for future work.

NOTE: A complete schedule for the CSLI seminar series on intelligent agents is
available at URL <http://www-csli.stanford.edu/>.  These talks are scheduled
for 12:15 (usually) on Thursdays throughout the Spring quarter.  If you order
beforehand, we will provide a sandwich and beverage (for $4.50, paid at the
door; quarters and small bills preferred).  Send sandwich orders (turkey, ham,
roastbeef, or vegetarian) to <bocata@csli.stanford.edu> some time on or before
Tuesday of the week of the given talk.

                               ____________

                  SEMINAR ON HUMAN-COMPUTER INTERACTION
                             on Friday, 5 May
                     12:30 p.m., Skilling Auditorium
			The Future of the Internet
                              Marc Andreesen
                   Netscape Communications Corporation
                           <marca@netscape.com>

                               ____________

                          PHILOSOPHY COLLOQUIUM
                             on Friday, 5 May
                    3:15 p.m., Building 460, Room 146
                          Title to be announced
                              Yair Guttmann
                           Stanford Philosophy
                       <guttmann@csli.stanford.edu>

                               ____________

                          LINGUISTICS COLLOQUIUM
                             on Friday, 5 May
                    3:30 p.m., Cordura Hall, Room 100
                            Grammatical Weight
                               Thomas Wasow
                           Stanford Linguistics
                        <wasow@csli.stanford.edu>

Long and complex -- or "heavy" -- elements tend to be placed near the ends of
sentences (as has been known since at least the early years of this century).
I will address three questions concerning this tendency:

        (A) What structural and/or pragmatic factors contribute
            to it?  That is, how should grammatical weight be defined?
        (B) Why are heavy elements saved for the ends of sentences?
        (C) How should the tendency be captured in a generative
            grammar?

Many answers to (A) have been proposed in the literature, but almost none have
been tested systematically.  Drawing on numerous examples of heavy NP shift,
particle movement, and the dative alternation from on-line corpora, I will
compare several possible measures of weight.

Speculations regarding (B) have centered around the idea that saving heavy
elements for the end might facilitate parsing.  I will discuss this idea,
particularly the version recently worked out in detail by Hawkins, and will
argue that considerations of production may play a role in the phenomenon as
well.

Finally, I will briefly address (C) with respect to heavy NP shift, arguing
against both the traditional rightward movement analysis and the recent
leftward movement approach put forward by Larson and Kayne.

                              _____________