[Prev][Next][Index]

CSLI Calendar, 26 January 1995, vol.10:13




        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
______________________________________________________________________________

26 January 1995                  Stanford                      Vol. 10, No. 13
______________________________________________________________________________

      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 25 JANUARY -- 3 FEBRUARY 1995

  WEDNESDAY, 25 JANUARY
         4:00 - Seminar on Computational Learning and Adaptation
                Cordura Hall, Room 100
                Flexible Metric Nearest Neighbor Classification
                Jerome H. Friedman, Stanford Statistics
                Abstract below

  THURSDAY, 26 JANUARY
        10:00 - STASS Seminar
                Cordura Hall, Room 100
                Information in its Time
                Geoff Nunberg, Xerox PARC
                Abstract below

        12:00 - CSLI TINLunch
                Cordura Hall, Room 100
                Using Situation Theory to Analyze Document Structure
                Keith Devlin, CSLI and Saint Mary's College, and
                Duska Rosenberg, Brunel University Computer Science
                Abstract below

         4:15 - SSP Distinguished Speaker Series
                Building 420, Room 041
                Non-Symbolic Approaches to Intelligence
                Rodney A. Brooks, MIT AI Laboratory
                Abstract below

         7:30 - Phonology Workshop
                Building 460, Seminar Room 146
                Foot and Syllable: Constraint Interaction in Japanese
                Accentuation 
                Haruo Kubozono, Osaka University / UC Santa Cruz
                Abstract below

  FRIDAY, 27 JANUARY
        12:00 - Logic Lunch
                Building 380, Room 383-N
                On Decidable Extensions of the Second-order Lambda
                Calculus with Subtyping
                Sergei Vorobyov, Max-Planck Institut fuer Informatik
                Abstract below

        12:30 - HCI Seminar
                Terman Auditorium
                Information Playgrounds: Local Metaphors for
                Interactive Information -- A Visual Design Perspective
                Eviatar Shafrir, Hewlett-Packard

         3:30 - Linguistics Colloquium
                Cordura Hall, Room 100
                Conditions in Conflict
                Arnold Zwicky, Ohio State and Stanford Linguistics
                Abstract below

  TUESDAY, 31 JANUARY
        12:00 - Logic Seminar
                Building 380, Room 381-T
                Title and speaker to be announced

  THURSDAY, 2 FEBRUARY
         2:15 - STASS Seminar (NOTE TIME CHANGE)
                Cordura Hall, Room 100
                Language, Models, and Consequence
                Greg O'Hair, CSLI and Flinders Philosophy
                Abstract below

         4:15 - SSP Forum
                Building 60, Room 61-F
                Plausible Incremental Steps in the Origin and Evolution of
                Language: An AI Account
                Jerry Hobbs, SRI AI Center
                Abstract below

  FRIDAY, 3 FEBRUARY
        12:00 - Logic Lunch
                Building 380, Room 383-N
                Title and speaker to be announced

        12:30 - HCI Seminar
                Terman Auditorium
                Directed Improvisation by Computer Characters
                Barbara Hayes-Roth, Stanford Knowledge Systems Laboratory
                Abstract below

         3:15 - Philosophy Colloquium
                Building 90, Room 91-A
                Plato's Theory of Goods in the Laws and Philebus
                Chris Bobonich, Chicago Philosophy

         3:30 - Linguistics Colloquium
                Cordura Hall, Room 100
                Sociolinguistic Theory and Application in the African 
                American Speech Community
                John Rickford, 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, 25 January
                    4:00 p.m., Cordura Hall, Room 100
             Flexible Metric Nearest Neighbor Classification
                            Jerome H. Friedman
                           Stanford Statistics

The K-nearest-neighbor decision rule assigns an object of unknown class to the
plurality class among the K labeled "training" objects that are closest to it.
Closeness is usually defined in terms of a metric distance on the Euclidean
space with the input measurement variables as axes. The metric chosen to
define this distance can strongly affect performance.  An optimal choice
depends on the problem at hand as characterized by the respective class
distributions on the input measurement space, and within a given problem, on
the location of the unknown object in that space. In this talk new types of
K-nearest neighbor procedures will be described that estimate the local
relevance of each input variable or their linear combinations for each
individual point to be classified. This information is then used to separately
customize the metric used to define the distance from that object in finding
its nearest neighbors. These procedures are a hybrid between regular
K-nearest-neighbor methods and tree-structured recursive partitioning
techniques popular in statistics and machine learning.

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, 26 January
                    10:00 a.m., Cordura Hall, Room 100
                         Information in its Time
                              Geoff Nunberg
                                Xerox PARC
                         <nunberg@parc.xerox.com>

This will be a report on work in progress. I want to talk about the notion of
"information" -- not in the various semantic and syntactic reconstructions
that philosophers and information theorists trade in, but in the vaguer sense
that people have in mind when they use "information" a modifier before words
like "economy," "explosion," or "age." In this sense information is invested
with a number of properties that it may or not share with its technical
cousins. It is a natural category that gives us the world in a way that is
perspective-free and publicly available; it can be transferred from one agent
to another with no loss of content.  It tends to be correct (but there is no
contradiction in saying that it is false).  It is divisible into intentional
atoms -- bits, infons, morceaux -- and comes in measurable volumes (and there
is a lot more of it around than there used to be). It has a geography but no
inherent structure.

I'll suggest that in this sense information is not an eternal category but a
historical construction, and was shaped by a set of institutions and print
genres that emerged around the middle of the 19th century -- the mass
newspaper, the popular dictionary, the catalogue, the travel guide, the
railroad schedule; the museum, the department store, fiat money. I want to
consider how the abstract properties that people ascribe to information arise
out of the material properties and social contexts of the documents that
convey it, and ask whether the print-mediated conception of information may
have led us to form unwarranted expectations about how documents will function
in a digital world.

                               ____________
                                      
                              CSLI TINLUNCH
                         on Thursday, 26 January
                    12:00 a.m., Cordura Hall, Room 100
           Using Situation Theory to Analyze Document Structure
             Keith Devlin, CSLI and Saint Mary's College, and
                  Duska Rosenberg, Brunel University, UK
         <devlin@csli.stanford.edu, duska.rosenberg@brunel.ac.uk>

In two recent CSLI Reports (94-187 and 94-189), we used situation theory in a
methodological way to analyze the role played by a particular kind of stylized
document (a "problem report form" from the computer industry).  The idea was
to find a level of analysis that has some of the precision that can be
obtained using mathematical formalisms and yet takes account of the insights
of, in particular, ethnomethodologists such as Harvey Sacks.  (A precursor of
this work was our joint paper on Sacks' "baby and mommy" analysis, which we
published in the third "Situation Theory and Its Applications" volume.)  We
call our analytic method "layered formalism and zooming" (LFZ). We do not
claim it is better than other analytic techniques.  But it does appear to be
different, leading to insights that might otherwise not be obtained.  The talk
will be structured to avoid dependence on any detailed knowledge of situation
theory. We hope to have Geoff Nunberg and Herb Clark present (among others) to
comment on this work, which is ongoing.

                               ____________
                                     
              SYMBOLIC SYSTEMS DISTINGUISHED SPEAKER SERIES
                         on Thursday, 26 January
                    4:15 p.m., Building 420, Room 041
                 Non-Symbolic Approaches to Intelligence
                             Rodney A. Brooks
                            MIT AI Laboratory

Over the past ten years we have built a large number of robots at the MIT AI
Lab which do not use conventional symbol systems, conventional representation,
nor conventional reasoning, but nevertheless they are able to carry out long
term tasks, have expectations about the world, have goals, and act like they
are making and following "plans."  Now we are pushing beyond the creature
metaphor trying to get at much more human like capabilities, but still without
conventional symbol systems.  We argue that physical embodiment is key to this
approach.

RODNEY BROOKS is Professor of Computer Science and and Associate Director of
the AI Lab at MIT.  His PhD was from Stanford in computer vision.  He was a
research scientist at CMU and MIT, then a faculty member at Stanford before
joining the faculty at MIT.  His research started out in traditional
Artificial Intelligence, but after working in vision, and then in robot
planning, he adopted a bottom up approach to intelligence, through building
artificial creatures.  Along the way he founded Lucid, Inc., of Menlo Park, a
Lisp compiler and systems company, and IS Robotics of Cambridge, MA, an
artificial creatures company.  He was awarded the Computers and Thought prize
by the trustees of IJCAI in 1991.

Distinguished Speakers are invited to lecture at Stanford each year by the
Symbolic Systems Students Society and the Cognitive Science Students Society.
Past lecturers have been Daniel Dennett, Douglas Hofstadter, Patricia
Churchland, Francisco Varela, and Donald Norman.

Please contact Kevin Henry <severian@leland.stanford.edu> with any questions.
                               ____________

                            PHONOLOGY WORKSHOP
                         on Thursday, 26 January
                7:30 p.m., Building 460, Seminar Room 146
                Foot and Syllable: Constraint Interaction
                         in Japanese Accentuation
                              Haruo Kubozono
                    Osaka University and UC Santa Cruz
                         <kubozono@ling.ucsc.edu>

In this talk I discuss the prosodic structure of Japanese compound nouns and
its implications for phonological theory.  In empirical terms, I will report
the following three findings about the nature of compound accentuation in
Tokyo Japanese.  First of all, a foot-based analysis leads to a significant
generalization of the seemingly complicated accent patterns behind Japanese
compounds.  Specifically, by constructing maximally bimoraic foot structure
with the degenerate (monomoraic) feet as a last resort, compound accentuation
in Japanese can be reduced to a highly general rule, a rule that places the
accent on the penultimate foot (Kubozono and Mester 1995).  Second, I will
present evidence against Poser's (1990) idea of foot extrametricality (or
invisibility), demonstrating instead that it is the final syllable, not the
final foot, that is invisible to the compound accent rule.  Thirdly, several
cases are presented where syllable extrametricality is violated or "revocated"
(Hayes 1991).

In the second half of the talk, I will discuss the theoretical implications of
these empirical generalizations for the nature of Japanese accentuation and
its foot structure.  It will be shown, first of all, that the generalization
based on the notion of "extrametricality" can be fully expressed by the
interaction of several constraints on the well-formedness of foot/accent
structure, notably Non-Finality(syll) and Non-Finality(foot).  Second, the
cases of extrametricality revocation can also be explained as a consequence of
constraint interaction, this time between Non-Finality and Weight-to-Stress.
Lastly, I will show that the above analysis does not support Poser's argument
against "syllable integrity," suggesting instead that foot structure can be
constructed in harmony with syllable structure in Japanese.

                               ____________

                               LOGIC LUNCH
                          on Friday, 27 January
                   12:00 noon, Building 380, Room 383-N
               On Decidable Extensions of the Second-Order
                      Lambda-Calculus with Subtyping
                             Sergei Vorobyov
                   Max-Planck Institut f\"ur Informatik
                          Saarbr\"ucken, Germany

The system Fsub, the second-order lambda-calculus with subtyping, extends
Girard's system F with the additional subtype relation on polymorphic types
S<T, the possibility to impose type bounds in universal type quantification
(All A<S.T), and by more general typing rules taking the subtyping relation
into account, like

     G |- t:(All A<S.T)   G |- R<S
     ----------------------------- (Type Application)
          G |- t{R} : T[R/A]

or

     G |- t:S     G |- S<T
     --------------------- (Subsumption)
         G |- t:T

Because of the last rule, a term in Fsub may generally have (infinitely) many
types, in contrast to the Church-style simply typed lambda-calculus or the
system F.

The subtyping relation of Fsub is axiomatized by a simple collection of rules,
expressing the existence of the largest type Top, transitivity, reflexivity of
<, and co(ntra)variance of subtyping of functional and universal types, e.g.,

       G |- T1<S1        G |- S2<T2
       ---------------------------- (->)
         G |- (S1->S2) < (T1->T2)

It turns out, however, that the subtyping relation of Fsub is undecidable.
This is mainly because of the rule for subtyping universal types, analogous to
the rule (->) above. Undecidability of the Fsub subtyping implies
undecidability of the typing in Fsub.

Rather than weakening the Fsub subtyping (to attain decidability), we show
that it possesses infinitely many decidable extensions, approximating the Fsub
subtyping with any desired precision.

Decidability of subtyping does not immediately imply the decidability of the
associated typing relation. There exist systems with decidable subtyping and
open decidability problem for typing. We show how our extensions of the Fsub
subtyping could be incorporated into decidable typing systems extending Fsub.

We finally present the typing proof normalization and subject reduction
theorems for our extensions of Fsub.

                               ____________

                  SEMINAR ON HUMAN-COMPUTER INTERACTION
                          on Friday, 27 January
                      12:30 p.m., Terman Auditorium
         Information Playgrounds: Local Metaphors for Interactive
                Information -- A Visual Design Perspective
                             Eviatar Shafrir
                             Hewlett-Packard
                        <shafrir@hpuid.ptp.hp.com>

In August 1994, for the Stanford Research Workshop on the World Wide Web, I
wrote: 

   The World Wide Web has opened the door to InfoTravel.  With our favorite
   'Mosaic' or 'Cello' brand InfoCars we take off and explore InfoSpace.
   Whatever metaphor we use to describe our online adventures, we are forced
   to experience them through the windshield of this, that, or yet another
   aging Desktop.  Furthermore, the information exhibits behave as static
   Documents in Application windows. 

   While each Desktop is populated with overlapping Windows, displaying
   dynamic contents and of elastic size, our InfoCars windshields expose us
   only to Documents where the contents are tiled, scrolled, and static.  In
   addition, we are not really traveling since there is no continuous motion
   in HyperSpace.  We can only jump from one InfoStop to the next and most
   times without ample guidance or expectation of what's ahead. 

   At each InfoStop the information itself is now active, hot, and live.  This
   creates an Interaction Dissonance between Desktop behaviors (inside the
   InfoCar's Dashboard & around its windshield) and local navigation and
   interaction metaphors unique to each InfoStop (visible through the
   windshield).  Since there is no InfoGovernment to standardize highway
   signage, the Information Designer must now perform road construction as
   well as emergency road service duties for bewildered travelers. 

More than a year has elapsed since the World Wide Web made national press.  We
have become experienced InfoTravelers and have accrued enough frequent Web
Surfing Miles allowing us to recognize by audio-print alone the presidential
cat from the alley variety.

The government-owned car industry (NCSA) with its standard-issue car (Mosaic)
has been privatized resulting in widely divergent cruising experiences.  Since
last summer Netscape has introduced limited 2-D layout, faster access, and
low-and-behold: blinking! However the onus of creating meaningful experiences
with information found on the World Wide Web still rests with its creators and
designers.

As an interaction designer working with professionals creating information and
having to answer to consumers seeking to use and understand that information I
seek to address the following questions:

   What is the difference between information found online and the same
   information available in other familiar forms?

   How are we interacting with information at each InfoStop once we get there?

Today, I would like to introduce to you the concept of Information
Playgrounds.  I coined this term to describe custom information experiences
constructed for interactive online media such as the World Wide Web.

Information Playgrounds are: 
   1. information stops containing one or more live documents, 
   2. sharing a single coherent metaphor of use, 
   3. ideally containing encapsulated interactions, 
   4. presenting a well-defined scope of activity, 
   5. and producing tangible results. 

While the search goes on for an all-encompassing Unified Interaction Metaphor
Theory, distributed information webs demonstrate the need and appropriateness
for multiple local access metaphors.  Interactions are manifestations of how we
perceive information.  An ideal playground will make a single trip across the
Web, from server to browser, complete with all its interactivity.  Not a
spreadsheet - but how about a calculator? Not a chess game - but how about
Solitaire? Once inside a playground, users assume a single interaction
metaphor.  Operating accordingly, they expect a concise set of 'things to do':
here I can search and retrieve abstracts; this is a music library; and this is
a candy shop, now I can order some brownies.

It is the experience of 'playing' with live information over time, tweeking
it, poking it here, test-casing it there, that users take with them from an
Information Playground.  Users seek validation by extracting information they
already know.  They are then willing to TRUST results to questions whose
answers they didn't know.  Combining live information with time spent by users
interacting with it produces information experiences, which are the tangible
results of Information Playgrounds.

Since each Information Playgrounds has its own rules, look, feel, and
clientele, the independent, colorful, multi-facet spirit of the World Wide Web
is preserved.  At the same time it is possible to package widely varying
Information Playgrounds and 'bind' them, if you will, into published volumes
of interactive information representing large organizations communicating with
diverse visitor bases.

Using overhead and online examples my talk: 

-  will point the dissonance in metaphors found between InfoSpace and
   Desktops mentioned in the abstract from last summer, 
-  highlight some differences between information found online and that found
   in other mediums, 
-  describe navigation metaphors used to 'bind' diverse information sources and
   'publish' them on the World Wide Web, and
-  focus on the design challenges and potential of Information Playgrounds. 

EVIATAR (EV) SHAFRIR, M.Sc. (Scorpio, born 1958) is a visual interaction
designer with the User Interaction Design group at Hewlett-Packard in
Sunnyvale, California.  He holds a B.Sc. degree in Mathematics and Computer
Science from the Hebrew University of Jerusalem and a M.Sc. degree in
Engineering Management from Stanford University.  He is the senior designer
responsible for the visual appearance and behavior of Access HP.

His design interests include non-latin screen-font creation, custom software
components, and off-the-wall visual metaphors for hard-to-swallow software
interfaces.  Ev is a member of SIGCHI and of the Association for Software
Design.  His latest co-authored article regarding online information design
titled "Blazing the Trail" appears in the January 1995 issue (volume 27 number
1) of the SIGCHI Bulletin.

                               ____________

                          LINGUISTICS COLLOQUIUM
                          on Friday, 27 January
                    3:30 p.m., Cordura Hall, Room 100
                          Conditions in Conflict
                              Arnold Zwicky
                   Ohio State and Stanford Linguistics
                        <zwicky@csli.stanford.edu>

What happens when two different conditions in a grammar impose incompatible
properties on expressions?  It seems only common sense that ungrammaticality
would result, but instead it appears that conflicts are always resolved in
favor of one condition, with the other suppressed.  Expressibility holds sway,
even at the cost of surface violations.

In phonology, there are numerous instances of suppression -- in particular, of
the general by the more specific, and in counterfeeding interactions.  Indeed,
the idea that such conflicts -- between conditions from universal grammar, but
interacting in parochial ways -- are common is a cornerstone of Optimality
Theory (and of Natural Phonology).

In morphology and syntax there is one common class of cases where conflicts
never result in ungrammaticality and one common class where they always do.
The first class (of a sort by definition not possible in phonology) involves
formally incompatible realizations for the same semantics; here the conflict
is resolved in favor of one realization in some contexts, the other in others,
and sometimes both as alternatives (BEEPED/*BEPT, *SLEEPED/SLEPT,
CREEPED/CREPT).  Expressibility is essential here; otherwise, the existence of
alternative expression for the same content would always make that content
inexpressible.

Examples in the second class turn out to involve at least one condition that
merely presupposes (rather than imposes) a particular property, so that there
is no conflict, merely a violation of the presupposed condition.  For
instance, the nominal gerunds in (1a) and (2a) are ungrammatical because each
violates a presupposed condition on this construction -- in (1a), the
condition that the "subjectoid" be licensed in a possessive form; in (2a), the
condition that it be licensed by some rule as a subject for the VP in this
construction.

 (1) a. *there's being snow on the street
     b.  There was snow on the street.
 (2) a. *under the rug's being snow on the street
     b.  under the rug's being a bad place to hide a gun

In morphology, conflicts between two imposed properties are not easy to find,
though they are attested (and they involve clearly parochial conditions).  One
well-known example is from Georgian transitive verb conjugation, in which
phonological material realizing two different, but semantically compatible,
grammatical categories nevertheless fills a single slot; only one affix
occurs, but it serves as an exponent of both categories.

The focus of this paper is such conflicts in syntax.  There are at least three
types: (a) two conditions require that some word have incompatible values for
a syntactic feature; (b) two conditions require incompatible orderings of two
expressions; (c) two conditions require each of two different expressions to
be immediately adjacent to, and on the same side of, some third expression.
Examples of each type from English:

 (3) It is flying pigs that I'm afraid of.
      [person/number of BE, vis-a-vis these values for the
       subject IT and for the predicative FLYING PIGS ...
       subject agreement wins]
 (4) Who will go first?
      [ordering of WH subject WHO and auxiliary verb WILL ...
       front position for WH wins]
 (5) Who the hell else did you see?
      [adjacency to WH word WHO of both ELSE and emphatic
       THE HELL ... the emphatic wins the position]

                               ____________

                              LOGIC SEMINAR
                          on Tuesday, 31 January
                   12:00 noon, Building 380, Room 381-T
                            Title and speaker
                             to be announced

                               ____________

                              STASS SEMINAR
                         on Thursday, 2 February
                    2:15 p.m., Cordura Hall, Room 100
                          (NOTE THE TIME CHANGE)
                    Language, Models, and Consequence
                               Greg O'Hair
                       CSLI and Flinders Philosophy
                        <ohair@csli.stanford.edu>

In this talk, I reflect on some issues concerning these topics raised by John
Etchemendy's critique of the standard model-theoretic account of logical
consequence in his book, _The Concept of Logical Consequence_.  I will try to
make most of the talk relatively self-contained and informal.

NOTE: Due to scheduling conflicts, the STASS Seminar has been moved today to
2:15 p.m. from its usual 10:00 a.m. time.  For the rest of this quarter we
will meet either at 12:00 noon or at 2:15 p.m. on Thursdays, as circumstances
permit, to avoid the 10:00 a.m. conflict.

                               ____________

                          SYMBOLIC SYSTEMS FORUM
                         on Thursday, 2 February
                    4:15 p.m., Building 60, Room 61-F
              Plausible Incremental Steps in the Origin and
                   Evolution of Language: An AI Account
                               Jerry Hobbs
                    SRI Artificial Intelligence Center
                            <hobbs@ai.sri.com>

There has been a moderate revival of interest recently in the question of the
origin and evolution of language.  One of the principal difficulties that
confronts any approach to this issue is in imagining plausible, incremental,
intermediate steps.  In this talk, after reviewing the recent literature, I
will suggest what an AI theory of discourse interpretation could contribute to
this question.

I assume an abductive account of interpretation in general, in which to
interpret a situation is to find the best explanation for the observables.  I
also assume an abductive account of discourse interpretation, in which to
interpret a sentence is to find the best proof of its logical form, allowing
assumptions, with respect to a knowledge base of axioms encoding commonsense
knowledge.  Increases in knowledge of language, whether in learning,
development, or evolution, are seen as results of incremental modifications on
the axioms, axiomatizations of common proofs, and the employment of theories
motivated independently of language.  Within this framework, I describe how
two of the principal features of language could have evolved -- Gricean
nonnatural meaning and syntactic structure.  The development of Gricean
meaning is seen as the employment of increasingly complex folk theories in the
interpretation of utterances as actions.  The development of syntactic
structure is seen as increasingly specific constraints on the interpretation
of the adjacency and proximity of strings of words as expressing
predicate-argument relations.  This sort of account suggests where we should
look for confirming evidence.

                               ____________

                               LOGIC LUNCH
                          on Friday, 3 February
                   12:00 noon, Building 380, Room 383-N
                            Title and speaker
                             to be announced

                               ____________

                  SEMINAR ON HUMAN-COMPUTER INTERACTION
                          on Friday, 3 February
                      12:30 p.m., Terman Auditorium
              Directed Improvisation by Computer Characters
                            Barbara Hayes-Roth
                  Stanford Knowledge Systems Laboratory
                          <bhr@ksl.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 is similar in spirit to previous work on instructable
agents, multi-agent systems, advanced programming languages, and
object-oriented simulation.  Following this earlier work, directed
improvisation aims to employ more abstract directions to guide more
discretionary behavior by more complete computer characters.

We believe that computer characters capable of performing directed
improvisation will be useful elements of diverse applications, especially
applications in education, the arts, and entertainment.  We are developing a
technology that combines a supporting agent architecture with configurable
components to facilitate efficient and economical development of a variety of
specific characters for different applications.  Our approach incorporates
media-independent interfaces, so that a given character's "mind" can be
embodied in animation, virtual reality, text, or other media.

Our current testbed application is a "virtual theater," in which animated
characters take direction to improvise episodes of physical and verbal
behaviors.  To be effective, the characters also must exhibit simple
intelligence, life-like qualities, and improvisational expertise.  Within the
virtual theater, we conceive several specific modes of interaction: animated
puppets, animated actors, collaborative playcrafting, improv troupe, improv
direction, and interactive story.  In all of these modes, directed
improvisation offers users (in our case, children) the combined pleasures of
directing the creation, re-telling, or re-experiencing of a story, while being
delighted by the characters' unpredictable improvisations within the story
structure.  In addition to its use as an experimental testbed, the virtual
theater is a promising paradigm for several kinds of commercial applications:
experience-based learning environments, new kinds of computer games, a new
medium for artistic self-expression, or a new form for interactive story
experience.

BARBARA HAYES-ROTH has been Senior Research Scientist in Computer Science at
Stanford since 1982.  Previously, she was Senior Computer Scientist at the
Rand Corporation and Member of Technical Staff at Bell Laboratories.  She
holds BA (1971, Boston University), MS (1973, University of Michigan), and PhD
(1974, University of Michigan) degrees in Cognitive Psychology.  Dr. Hayes-Roth
developed the "dynamic control architecture" (embodied in the BB1 software),
which supports the development of computer agents that explicitly reason
about, plan, and control their own behavior in response to run-time
conditions.  Dr. Hayes-Roth has collaborated with domain experts to develop
experimental agents for: modeling protein structures, designing construction
site layouts, monitoring medical patients, monitoring plant control and
manufacturing equipment, controlling autonomous robots, and more recently,
controlling improvisational computer characters.  She has published over 100
scientific articles and is a former Councillor, current Conference Chair, and
a Fellow of the AAAI.

                               ____________

                          PHILOSOPHY COLLOQUIUM
                          on Friday, 3 February
                    3:15 p.m., Building 90, Room 91-A
             Plato's Theory of Goods in the Laws and Philebus
                              Chris Bobonich
                            Chicago Philosophy

                               ____________

                          LINGUISTICS COLLOQUIUM
                          on Friday, 3 February
                    3:30 p.m., Cordura Hall, Room 100
                Sociolinguistic Theory and Application in
                  the African American Speech Community
                              John Rickford
                           Stanford Linguistics
                       <rickford@csli.stanford.edu>

In this paper I argue that American sociolinguistics -- particularly American
quantitative sociolinguistics, its dominant subfield -- has, over the past
quarter century, drawn substantially on data from the African American Speech
Community for its descriptive, theoretical and methodolological development,
but it has given relatively little back to the community in terms of
representation or practical application.  Although this is a common (but
regrettable) state of affairs in sociolinguistics and linguistics and perhaps
academia more generally, it is particularly deleterious in the case of AAVE
because the major studies of the late 1960s which started serious research in
this field were funded by the Office of Education and were undertaken with
practical applications in mind.

The theoretical, descriptive, and methodological contributions which data from
AAVE have provided to sociolinguistics will be discussed under four
sub-headings: (1) Variable Rules; (2) Tense-Aspect Markers; (3) Social Class
and Style; and (4) the Creolization and Divergence Issues.  I will draw
primarily on work from our East Palo Alto Neighborhood Study (EPANS) and
Copula Project for this discussion.

The issue of applications (or the lack thereof) will be discussed under four
sub-headings too: (1) The induction of African American linguists into the
field; (2) The representation of African Americans in our writings; (3)
Involvement in court decisions and workplace opportunities; (4) Education,
especially in the teaching of reading, writing, and the language arts at the
elementary school level.  I will focus in particular on the educational
issues, querying in particular whether the retreat from the proposal to
experiment with dialect readers as a preliminary aid in the teaching of
reading was justified.

I will end with a plea for greater concern for what we can do to help the
speakers (from every community) who provide the data for our books, articles,
theses, and term papers.  And a plea that we go beyond concern to involvement
and action.

                               ____________

                            NEW CSLI VISITORS

    - Masaki Suwa
      IAP visiting researcher from HITACHI
      Nov 94 - Nov 95

    - Greg O'Hair
      Dept of Philosophy
      Flinders University, Australia 
      Jan 4 - March 31, 1995

    - Patricia Blanchette
      Dept of Philosophy
      University of Notre Dame
      Jan 4 - Aug 31, 1995

    - Chris Cully
      Dept of Linguistics
      University of Iowa
      Jan 4 - June 30, 1995

    - Kook Chung
      Dept of Linguistics
      Hankuk University, Seoul
      Jan 94 - Feb 96

                               ____________