Proceedings of the Twenty-third Annual Conference of the Cognitive Science Society (Cognitive Science Society (Us) Conference//Proceedings) [Pap/Cdr ed.] 9780805841527, 0805841520

Citation preview



3URFHHGLQJVRIWKH 7ZHQW\7KLUG $QQXDO&RQIHUHQFH RIWKH &RJQLWLYH6FLHQFH6RFLHW\           ²$XJXVW  8QLYHUVLW\RI(GLQEXUJK (GLQEXUJK6FRWODQG

 (GLWRUV -RKDQQD'0RRUH .HLWK6WHQQLQJ

Proceedings of the Twenty-Third Annual Conference of the Cognitive Science Society

Johanna D. Moore and Keith Stenning Editors

August 1-4, 2001 Human Communication Research Centre University of Edinburgh Edinburgh, Scotland

LAWRENCE ERLBAUM ASSOCIATES, PUBLISHERS 2001

Mahwah, New Jersey

London

c 2001 by the Cognitive Science Society Copyright All rights reserved. No part of this book may be reproduced in any form, by photostat, microform, retrieval system, or by any other means, without the prior written permission of the publisher. Distributed by Lawrence Erlbaum Associates, Inc. 10 Industrial Avenue Mahwah, New Jersey 07430 ISBN 0-8058-4152-0 ISSN 1047-1316 Printed in the United States of America

Dedicated to the memory of Herbert A. Simon, June 15, 1916 – February 9, 2001

How It All Got Put Together Once upon a time when the world was young, Oh best beloved. There came to the banks of the Monongogo River, All muddy and brown, Oh best beloved, A djinn who was one thing on the inside But many things on the outside. And he camped by the banks of the Monongogo River, All muddy and brown, Oh best beloved. And he stayed and stayed and he never went away. And he did his magic there. He had many hands, each hand with many fingers, Oh best beloved. More hands and fingers than you and I More hands than you have fingers, More fingers on each hand than you have toes. Each hand played a tune on a magic flute, Oh best beloved. And each fluted tune floated out on a separate flight. And each was a tune for a separate dance, And each was heard in a separate place, And each was heard in a separate way, And each was merged in the dance it swayed. But it was still all the same tune, For that was the magic of the djinn. Now, best beloved, listen near— Each separate place, when the world was young, Danced in a way that was all its own, Different from all of the others. But the melody told of how it could be That creatures out of an ancient sea, By dancing one dance on the inside, Could dance their own dance on the outside, Because of the place where they were in—

All of its ins and outs. For that was the magic of the djinn. And little by little, each swayed a new way, Taking the melody each its own way, But hearing the melodies far away From other places with separate dances, But the very same melody That told the dance to be done on the inside. So, each started to step in the very same way, Putting together one dance on the inside For many dances on the outside. So the melody grew, and it drifted back To the Monongogo River, all muddy and brown, And the river came clear and sweet. Ah, best beloved, I must tell the truth. The river is not yet clear and sweet, Not really so. Because putting together is a task forever. And no one—not even a djinn with kilohands and megafingers, All of which play a different-same tune— Can put all things together in a single breath, Not even a breath of fifty years. It is not all put together yet, And it never shall be, For that is the way of the world. But even so, when the world was young, Was the time of the need for the single tune To guide the dance that would move together All of the steps in all of the places. And it happened by the banks of the Monongogo River, All muddy and brown, Best beloved. And the river will never be the same. Just so.

Allen Newell Carnegie Mellon University

Foreword This volume contains the papers and posters selected for presentation at the 23rd Annual Meeting of the Cognitive Science Society in Edinburgh, August 1–4th 2001. This meeting is the first in the history of the society to be held outside North America, and reflects the increasing internationalisation of cognitive science. More than 500 submissions were received from all over the world. The breadth of topics treated, together with the evident themes that recur are testimony to the development of a distinctive field. We were reminded of the multidimensionality of the field when the several papers on topics related to categorisation proved to be the hardest of all to categorise. It is our belief that the virtue of cognitive science comes from its deep engagement with the full range of disciplines that contribute to informational theories of mind. Cognitive science began with the realisation that several disciplines studied what is ultimately the same subject matter using different concepts and methods. Observation and experiment had become separated from simulation, engineering, formal analysis, historical, cultural and evolutionary study, and philsophical speculation. It is our hope that this conference will play its small part in substantiating the vision that it is important to put back together what the disciplines have cast asunder. This multidimensionality of the field makes scheduling a major headache. It is impossible to ensure that clashes do not occur. At one point in scheduling we resorted to statistical corpus analysis on the presented papers to reveal implicit structure. (You will perhaps be relieved to hear that human analysis still appears to be ahead of LSA at this task). We hope that you enjoy the program that has resulted. We would like to acknowledge help from the following sources, without whom this event would certainly not have been possible: The Cognitive Science Society Board for inviting us to host the meeting and providing the framework, expertise and support. The Program Committee assigned submissions to referees, read their resulting reviews and made judgments on the five hundred submissions. The Reviewers (and there were more than 250 of them) reviewed the papers and gave feedback to committee and authors. Interdisciplinary reviewing is not an easy task. Submitting interdisciplinary papers sometimes feels like being tried by the legal systems of several cultures simultaneously. A necessarily imperfect process was carried out with good grace and some assurance of the quality of decisions. These tasks of assigning and performing reviews are second only to the quality of submissions in determining the calibre of the meeting. The Tutorial Chair (Frank Ritter) who was responsible for the construction and organisation of the tutorial program. The many volunteers who helped with the myriad local arrangements for a meeting of this size, and especially Jean McKendree who chaired the local arrangements committee. The meeting certainly would not have happened without Frances Swanwick who coordinated the submissions process and Jonathan Kilgour who kept the software working, or without Mary Ellen Foster’s tireless work on the Proceedings. Janet Forbes and her successor David Dougal, and their secretarial team: Margaret Prow, Eva Steel, and Yvonne Corrigan for providing administrative support. Financial support: British Academy, NSF, Erlbaum, Elsevier, Wellcome, the Glushko Foundation, and the Human Communication Research Centre. The plenary speakers Jon Elster, Wilfred Hodges and Dan Sperber. And lastly, and most importantly, the authors and symposium participants who presented their work, and made the conference what it was. Johanna Moore and Keith Stenning Conference Chairs, CogSci 2001 Human Communication Research Centre Edinburgh University

Twenty-Third Annual Conference of the Cognitive Science Society August 1-4 2001 Human Communication Research Centre University of Edinburgh Scotland

Conference Co-Chairs Johanna D. Moore, University of Edinburgh Keith Stenning, University of Edinburgh

Conference Program Committee Susan Brennan, SUNY Stonybrook Gordon Brown, Warwick Nick Chater, Warwick Peter Cheng, Nottingham Andy Clarke, Sussex Axel Cleeremans, Brussels Gary Cottrell, UCSD Matt Crocker, Saarbrucken Jean Decety, INSERM, Paris Rogers Hall, UC Berkeley Dan Jurafsky, U. Colorado, Boulder Irvin Katz, ETS, Princeton Ken Koedinger, CMU

Michiel van Lambalgen, Amsterdam Frank Ritter, Penn State Mike Oaksford, Cardiff Stellan Ohlsson, U. Illinois, Chicago Tom Ormerod, Lancaster Michael Pazzani, UC Irvine Christian Schunn, George Mason Steven Sloman, Brown University Niels Taatgen, Groningen Andree Tiberghien, CNRS, Lyon Richard Young, Hertfordshire Jiajie Zhang, U. Texas at Houston

Local Arrangments Committee Jean McKendree, Chair David Dougal Janet Forbes Ian Hughson Padraic Monaghan Peter Wiemer-Hastings Daniel Yarlett Submissions Coordinator Frances Swanwick Conference Software Maintainer Jonathan Kilgour Proceedings Mary Ellen Foster, Jonathan Kilgour Program Coordinator Michael Ramscar Registration Website Arthur Markman Website John Mateer, Jonathan Kilgour, Frances Swanwick

Marr Prize 2001 Sam Scott, Department of Cognitive Science, Carleton University Metarepresentation in Philosophy and Psychology

This conference was supported by the Cognitive Science Society, The British Academy, The Wellcome Foundation, Lawrence Erlbaum Associates Ltd, Elsevier Science, The Glushko Foundation and The Human Communication Research Center.

The Cognitive Science Society Governing Board Lawrence W. Barsalou, Emory University Jeffery Elman, University of California at San Diego Susan L. Epstein, Hunter College and the City University of New York Martha Farah, University of Pennsylvania Kenneth D. Forbus, Northwestern University Dedre Gentner, Northwestern University James G. Greeno, Stanford University Alan Lesgold, University of Pittsburgh Douglas L. Medin, Northwestern University Michael Mozer, University of Colorado Vimla Patel, McGill University Kim Plunkett, Oxford University Colleen Seifert, University of Michigan Keith Stenning, Edinburgh University Paul Thagard, University of Waterloo

Chair of the Governing Board Lawrence W. Barsalou, Emory University

Chair Elect Susan L. Epstein, Hunter College and the City University of New York

Journal Editor Robert L. Goldstone, Indiana University

Executive Officer Arthur B. Markman, University of Texas

The Cognitive Science Society, Inc., was founded in 1979 to promote interchange across traditional disciplinary lines among researchers investigating the human mind. The Society sponsors an annual meeting, and publishes the journal Cognitive Science. Membership in the Society requires a doctoral degree in a related discipline (or equivalent research experience); graduate and undergraduate students are eligible for a reduced rate membership; and all are welcome to join the society as affiliate members. For more information, please contact the society office or see their web page at http://www.cognitivesciencesociety.org/ Cognitive Science Society, University of Michigan, 525 East University, Ann Arbor MI, 48109-1109; [email protected]; phone and fax (734) 429-4286

Reviewers for the Twenty-Third Annual Conference of the Cognitive Science Society Agnar Aamodt Amit Almor Rick Alterman Eric Altmann Richard Anderson Jennifer Arnold Stephanie August Neville Austin Thom Baguley Todd Bailey Nicolas Balacheff Linden Ball Dale Barr Pierre Barrouille Renate Bartsch Rik Belew Bettina Berendt Rens Bod Lera Boroditsky Heather Bortfeld Brian Bowdle Holly Branigan Frances Brazier Bruce Bridgeman Ted Briscoe Paul Brna Andrew Brook Patty Brooks Curtis Brown Marc Brysbaert John Bullinaria Curt Burgess Bruce Burns Ruth Byrne Antonio Caballero Laura Carlson Mei Chen Morten Christiansen Ed Chronicle James Chumbley Cathy Clement Charles Clifton Tom Conlon Fred Conrad Rick Cooper Richard Cox Ian Cross Fernando Cuetos Matthew Dailey Helen deHoop Arnaud Destrebecqz Morag Donaldson Ann Dowker Ben du Boulay Reinders Duit George Dunbar

Matthew Elton Randi Engle Mary Enright Noel Enyedy Michael Erickson Martha Evens John Everatt Neil Fairley Marte Fallshore Vic Ferreira Rodolfo Fiorini Ilan Fischer Peter Flach Nancy Franklin Robert French Ann Gallagher Simon Garrod Mike Gasser Richard Gerrig David Glasspool Fernand Gobet Laura Gonnerman Barbara Gonzalez Peter Gordon Barbara Graves Wayne Gray Peter Grunwald Prahlad Gupta Karl Haberlandt Constantinos Hadjichristidis Fritz Hamm James Hampton Joy Hanna Trevor Harley Cathy Harris Nancy Hedberg Neil Heffernan Evan Heit Petra Hendriks Denis Hilton Eduard Hoenkamp Ulrich Hoffrage Douglas Hofstadter Anne Holzapfel Sid Horton Eva Hudlicka Elizabeth Ince Heisawn Jeong Michael Kac James Kahn Hans Kamp David Kaufman James Kaufman Fred Keijzer Frank Keller Gerard Kempen

Thomas King Sheldon Klein Guenther Knoblich Chris Koch Derek Koehler Boicho Kokinov Rita Kovordanyi Carol Krumhansl Pat Kyllonen Aarre Laakso Nicki Lambell Matthew Lambon-Ralph Donald Laming Alex Lamont Peter Lane Maria Lapata Michal Lavidor John Leach David Leake Christian Lebiere Jeff Lidz Brad Love Will Lowe George Luger Jose Luis Bermudez Rachel McCloy Scott McDonald Brendan McGonigle Jim MacGregor Jean McKendree Craig McKenzie Brian MacWhinney Paul Maglio Lorenzo Magnani Barbara Malt Ken Manktelow Denis Mareschal Art Markman Amy Masnick Santosh Mathan Yoshiko Matsumoto Mark Mattson Sven Mattys David Medler Monica Meijsing Paola Merlo Craig Miller Toby Mintz S Mitra Naomi Miyake Padraic Monaghan Joyce Moore Bradley Morris Paul Munro Wayne Murray Srini Narayanan

J Nerbonne David Noelle Breanndan O Nuallain Padraig O’Seaghdha Magda Osman Helen Pain Leysia Palen Barbara Partee Vimla Patel Kevin Paterson Barak Pearlmutter David Peebles Pierre Perruchet Alexander Petrov Steven Phillips Massimo Piattelli-Palmarini Martin Pickering Julian Pine Massimo Poesio Eric Postma Emmanuel Pothos Athanassios Protopapas Michael Ramscar William Rapaport Stephen Read Bob Rehder Kate Rigby Steve Ritter Bethany Rittle-Johnson Max Roberts Scott Robertson

Jenni Rodd Robert Roe Christoph Scheepers Hermi Schijif Friederike Schlaghecken Matthew Schlesinger Ute Schmid Thomas Schultz Philippe Schyns Julie Sedivy David Shanks Bruce Sherin Val Shute Asma Siddiki Derek Sleeman Peter Slezak Vladimir Sloutsky Linda Smith Cristina Sorrentino Jacques Sougne Bobbie Spellman Michael Spivey Constance Steinkuehler Suzanne Stevenson Neil Stewart Stephen Stich Rob Stufflebeam Patrick Sturt Michael Tanenhaus Heike Tappe Adam Taylor

Virginia Teller Josh Tenebaum Charles Tijus Michael Tomasello Greg Trafton David Traum Jody Underwood Ludger van Elst Ezra van Everbroeck Maarten van Someren Alonso Vera Rineke Verbrugge Gregg Vesonder Michael Waldmann Lyn Walker William Wallace Hongbin Wang Pei Wang Amy Weinberg Mike Wheeler Bob Widner Cilia Witterman Amanda Woodward Lee Wurm Takashi Yamauchi Wai Yeap Wayne Zachary Jeff Zacks Corrine Zimmerman Daniel Zizzo

Tutorial Program August 1st, 2001

How to Deal with Modularity in Formal Language Theory: An Introduction to Grammar Systems, Grammar Ecosystems and Colonies Carlos Martin-Vide, Rovira i Virgili University

APEX: An Architecture for Modeling Human Performance in Applied HCI Domains Michael Matessa, NASA Ames Research Center Michael Freed - NASA Ames Research Center John Rehling - NASA Ames Research Center Roger Remington - NASA Ames Research Center Alonso Vera - NASA Ames Research Center

An Introduction to the COGENT Cognitive Modelling Environment (with special emphasis on applications in computational linguistics) Dr. Richard Cooper, Birkbeck College Dr. Peter Yule, Birkbeck College

Eye Tracking Roger P.G. van Gompel, University of Dundee Wayne S. Murray, University of Dundee

ACT-R 5.0 John R. Anderson, Carnegie Mellon University

Tutorial Co-Chairs Frank Ritter, Penn State University Richard Young, University of Hertfordshire

Tutorial Committee Members Randy Jones, University of Michigan Todd Johnson, University of Texas, Houston Vasant Honavar Iowa State University Kevin Korb, Monash University Michail Lagoudakis, Duke University Toby Mintz, University of Southern California Josef Nerb, University of Freiberg and University of Waterloo Gary Jones, University of Derby Padraic Monaghan, University of Edinburgh

Speakers and Symposia Invited Speakers Jon Elster, Columbia University Wilfred Hodges, Queen Mary and Westfield College, University of London Dan Sperber, CNRS, Paris

Invited Symposia Emotion and Cognition Chair: Keith Stenning, University of Edinburgh Speakers: Ziva Kunda, Waterloo University Paul Seabright, Toulouse University Drew Westen, Boston University Representation and Modularity Chair: Jon Oberlander, University of Edinburgh Speakers: Lawrence Hirschfeld, University of Michigan Annette Karmiloff-Smith, Institute of Child Health, London Dylan Evans, King’s College, London

Submitted Symposia Computational Models of Historical Scientific Discoveries Chairs: Pat Langley, ISLE, Stanford Lorenzo Magnani, University of Pavia Presenters: Peter Cheng, Adrian Gordon, Sakir Kocabas, Derek Sleeman When Learning Shapes its own Environment Chair: James Hurford, University of Edinburgh Presenters: Gerd Gigerenzer, Simon Kirby, Peter Todd The Interaction of Explicit and Implicit Learning Chairs: Ron Sun, University of Missouri-Columbia Robert Matthews, Louisiana State University Presenters: Axel Cleermans, Zoltan Dienes The Cognitive Basis of Science: The View from Science Chair: Nancy Nersessian, Georgia Institute of Technology Presenters: Stephen Sich, Ronald Giere, Dedre Gentner

Herb Simon Memorial Symposium Chair: John Anderson Presenters: Pat Langley, ISLE, Stanford “Computational Scientific Discovery and Human Problem Solving” Fernand Gobet, University of Nottingham “Is Experts’ Knowledge Modular?” Kevin Gluck, Air Force Research Laboratory “The Right Tool for the Job: Information Processing Analysis in Categorisation”

“For us life is, as Shakespeare and many others have described it, a play—a very serious play whose meaning lies in living it. Like any play, in order to have meaning, it must have a beginning, a middle and an end. If an act spans about a decade, eight acts are already a very long play, making heavy demands on the dramatist (oneself) to give it shape. “Dot and I have had remarkably happy and lucky lives (the first requires the second), which continue to be interesting and challenging, and we have no urge to end them. On the other hand, the realization that these lives are likely, in fact, to end at almost any time now evokes no resentment of fate—at most, sometimes a gentle sadness. We are resigned, not in a sense of giving up or losing, but in a sense of wanting to end our years with dignity, good memories and a feeling that the play had a proper shape and ending, including a final curtain.” Herb Simon

Contents Symposia Computational Models of Historical Scientific Discoveries Pat Langley (Institute for the Study of Learning and Expertise), Lorenzo Magnani (Department of Philosophy, University of Pavia), Peter C.-H. Cheng (School of Psychology, University of Nottingham), Adrian Gordon (Department of Computing, University of Northumbria), Sakir Kocabas (Space Engineering Department, Istanbul Technical University) and Derek H. Sleeman (Department of Computing Science, University of Aberdeen) When Cognition Shapes its Own Environment Peter Todd (Center for Adaptive Behavior and Cognition, Max Planck Institute for Human Development), Simon Kirby and James Hurford (Language Evolution and Computation Research Unit, Department of Theoretical and Applied Linguistics, University of Edinburgh) The Cognitive Basis of Science: The View from Science Nancy J. Nersessian (College of Computing, Georgia Institute of Technology) The Interaction of Explicit and Implicit Learning Ron Sun (University of Missouri-Columbia), Robert Mathews (Louisiana State University, Baton Rouge)

Papers & Posters The Role of Language on Thought in Spatio-temporal Metaphors Tracy Alloway, Michael Ramscar and Martin Corley (University of Edinburgh) Coordinating Representations in Computer-Mediated Joint Activities Richard Alterman, Alex Feinman, Josh Introne and Seth Landsman (Brandeis University) An Integrative Approach to Stroop: Combining a Language Model and a Unified Cognitive Theory Erik Altmann (Michigan State University) and Douglas Davidson (University of Illinois at Urbana-Champaign) Age of Acquisition in Connectionist Networks Karen Anderson and Garrison Cottrell (University of California, San Diego) The Processing & Recognition of Symbol Sequences Mark Andrews (Cornell University) Comprehension of Action Sequences: The Case of Paper, Scissors, Rock Patric Bach, G¨unther Knoblich (Max Planck Institute for Psychological Research), Angela D. Friederici (Max Planck Institute for Cognitive Neuroscience) and Wolfgang Prinz (Max Planck Institute for Psychological Research) Toward a Model of Learning Data Representations Ryan Baker, Albert Corbett and Kenneth Koedinger (Human-Computer Interaction Institute, Carnegie Mellon University)

Referential Form, Word Duration, and Modeling the Listener in Spoken Dialogue Ellen Bard and Matthew Aylett (University of Edinburgh) The Utility of Reversed Transfers in Metaphor John Barnden (The University of Birmingham) A model theory of deontic reasoning about social norms Sieghard Beller (Department of Psychology, University of Freiburg, Germany) Cue Preference in a Multidimensional Categorization Task Patricia Berretty (Fordham University) A Perceptually Driven Dynamical Model of Rhythmic Limb Movement and Bimanual Coordination Geoffrey Bingham (Psychology Department and Cognitive Science Program, Indiana University) Inferences About Personal Identity Sergey Blok, George Newman , Jennifer Behr and Lance Rips (Northwestern University) Graded lexical activation by pseudowords in cross-modal semantic priming: Spreading of activation, backward priming, or repair? Jens B¨olte (Psychologisches Institut II) Understanding recognition from the use of visual information Lizann Bonnar, Philippe Schyns and Fr´ed´eric Gosselin (University of Glasgow) Taxonomic relations and cognitive economy in conceptual organization Anna Borghi (University of Bologna ) and Nicoletta Caramelli (University of Bologna) The Roles of Body and Mind in Abstract Thought. Lera Boroditsky (Stanford University), Michael Ramscar (Edinburgh University) and Michael Frank (Stanford University) The time-course of morphological, phonological and semantic processes in reading Modern Standard Arabic Sami Boudelaa and William Marslen-Wilson (MRC-CBU) Reference-point Reasoning and Comparison Asymmetries Brian Bowdle (Indiana University) and Douglas Medin (Northwestern University) Deference in Categorisation: Evidence for Essentialism? Nick Braisby (Open University) Meaning, Communication and Theory of Mind. Richard Breheny (RCEAL, University of Cambridge) The Effects of Reducing Information on a Modified Prisoner’s Dilemma Game Jay Brown and Marsha Lovett (Carnegie Mellon University) Mice Trap: A New Explanation for Irregular Plurals in Noun-Noun Compounds Carolyn Buck-Gengler, Lise Menn and Alice Healy (University of Colorado, Boulder) Simulating the Evolution of Modular Neural Systems John Bullinaria (University of Birmingham, UK) The Hot Hand in Basketball: Fallacy or Adaptive Thinking?

Bruce Burns (Michigan State University) Modelling Policies for Collaboration Mark Burton (ARM) and Paul Brna (Computer Based Learning Unit, Leeds University) Evaluating the Effects of Natural Language Generation Techniques on Reader Satisfaction Charles Callaway and James Lester (North Carolina State University) How Nouns and Verbs Differentially Affect the Behavior of Artificial Organisms Angelo Cangelosi (PION Plymouth Institute of Neuroscience, University of Plymouth) and Domenico Parisi (Institute of Psychology, National Research Council) Learning Grammatical Constructions Nancy C. Chang (International Computer Science Institute) and Tiago V. Maia (State University of New York at Buffalo) A Model of Infant Causal Perception and its Development Harold Chaput and Leslie Cohen (The University of Texas at Austin) The Effect of Practice on Strategy Change Suzanne Charman and Andrew Howes (School of Psychology, Cardiff University) A Potential Limitation of Embedded-Teaching for Formal Learning Mei Chen (Concordia University) Drawing out the Temporal Signature of Induced Perceptual Chunks Peter Cheng, Jeanette McFadzean and Lucy Copeland (ESRC Centre for Research in Development, Instruction and Training, Department of Psychology, University of Nottingham, U.K.) Modeling Tonality: Applications to Music Cognition Elaine Chew (University of Southern California) Causal Information as a Constraint on Similarity Jessica Choplin, Patricia Cheng and Keith Holyoak (University of California, Los Angeles) Hemispheric Lateralisation of Length effect Yu-Ju Chou and Richard Shillcock (Division of Informatics, University of Edinburgh) Integrating Distributional, Prosodic and Phonological Information in a Connectionist Model of Language Aquisition Morten Christiansen and Rick Dale (Southern Illinois University, Carbondale) Using Distributional Measures to Model Typicality in Categorization Louise Connell (University College Dublin) and Michael Ramscar (University of Edinburgh) Young Children’s Construction of Operational Definitions in Magnetism:the role of cognitive readiness and scaffolding the learning environment Constantinos Constantinou, Athanassios Raftopoulos and George Spanoudis (University of Cyprus) Testing a computational model of categorisation and category combination: Identifying diseases and new disease combinations Fintan Costello (Dublin City University) Exploring Neuronal Plasticity: Language Development in Pediatric Hemispherectomies

Stella de Bode and Susan Curtiss (UCLA, Neurolinguistics Laboratory) ’Does pure water boil, when it’s heated to 100C?’: The Associative Strength of Disabling Conditions in Conditional Reasoning Wim De Neys, Walter Schaeken and G´ery d’Ydewalle (KULeuven) When Knowledge is Unconscious Because of Conscious Knowledge and Vice Versa Zoltan Dienes (Sussex University) and Josef Perner (University of Salzburg) What Can Homophone Effects Tell Us About the Nature of Orthographic Representation in Visual Word Recognition? Jodi Edwards (Department of Linguistics, University of Calgary) and Penny Pexman (Department of Psychology, University of Calgary) Memory Representations of Source Information Reza Farivar (McGill University), Noah Silverberg and Helena Kadlec (University of Victoria) Testing Hypotheses About Mechanical Devices Aidan Feeney (University of Durham) and Simon Handley (University of Plymouth) An Influence of Spatial Language on Recognition Memory for Spatial Scenes Michele Feist and Dedre Gentner (Northwestern University) The Origin of Somatic Markers: a Suggestion to Damasio’s Theory Inspired by Dewey’s Ethics Suzanne Filipic (Universit´e de Paris III-Sorbonne Nouvelle ) Investigating Dissociations Between Perceptual Categorization and Explicit Memory Marci Flanery, Thomas Palmeri and Brooke Schaper (Vanderbilt University) Development of Physics Text Corpora for Latent Semantic Analysis Donald Franceschetti , Ashish Karnavat , Johanna Marineau , Genna McCallie , Brent Olde, Blair Terry and Arthur Graesser (University of Memphis) Modeling Cognition with Software Agents Stan Franklin and Arthur Graesser (Institute for Intelligent Systems, The University of Memphis) Reversing Category Exclusivities in Infant Perceptual Categorization: Simulations and Data Robert French, Martial Mermillod (University of Li`ege, Belgium), Paul Quinn (Washington and Jefferson University, U.S.A.) and Denis Mareschal (Birkbeck College, U.K.) Adaptive Selection of Problem Solving Strategies Danilo Fum and Fabio Del Missier (Department of Psychology, University of Trieste) Self-Organising Networks for Classification Learning from Normal and Aphasic Speech Sheila Garfield, Mark Elshaw and Stefan Wermter (University of Sunderland) Rational imitation of goal-directed actions in 14-month-olds Gy¨orgy Gergely (Institute for Psychology, Hungarian Academy of Sciences), Harold Bekkering (Max Planck Institute for Psychological Research) and Ildik´o Kir´aly (Institute for Psychology, Hungarian Academy of Sciences) The Right Tool for the Job: Information-Processing Analysis in Categorization

Kevin Gluck (Air Force Research Laboratory), James Staszewski, Howard Richman, Herb Simon and Polly Delahanty (Carnegie Mellon University) Is Experts’ Knowledge Modular? Fernand Gobet (School of Psychology, University of Nottingham) Strategies in Analogous Planning Cases Andrew Gordon (IBM TJ Watson Research Center) Superstitious Perceptions Fr´ed´eric Gosselin, Philippe Schyns, Lizann Bonnar and Liza Paul (University of Glasgow) Words and Shape Similarity Guide 13-month-olds Inferences about Nonobvious Object Properties Susan Graham, Cari Kilbreath and Andrea Welder (University of Calgary) The Emergence of Semantic Categories from Distributed Featural Representations Michael Greer (Centre for Speech and Language, Department of Experimental Psychology, University of Cambridge), Maarten van Casteren (MRC Cognition and Brain Sciences Unit, Cambridge, UK), Stuart McLellan, Helen Moss, Jennifer Rodd (Centre for Speech and Language, Department of Experimental Psychology, University of Cambridge), Timothy Rogers (MRC Cognition and Brain Sciences Unit, Cambridge, UK) and Lorraine Tyler (Centre for Speech and Language, Department of Experimental Psychology, University of Cambridge) Belief Versus Knowledge: A Necessary Distinction for Explaining, Predicting, and Assessing Conceptual Change Thomas Griffin and Stellan Ohlsson (University of Illinois at Chicago) Randomness and coincidences: Reconciling intuition and probability theory Thomas Griffiths and Joshua Tenenbaum (Department of Psychology, Stanford University) Judging the Probability of Representative and Unrepresentative Unpackings Constantinos Hadjichristidis (Department of Psychology, University of Durham), Steven Sloman (Department of Cognitive & Linguistic Sciences, Brown University) and Edward Wisniewski (Department of Psychology, University of North Carolina at Greensboro) On the Evaluation of If p then q Conditionals Constantinos Hadjichristidis, Rosemary Stevenson (Department of Psychology, University of Durham), David Over (School of Social Sciences, University of Sunderland ), Steven Sloman (Department of Cognitive & Linguistic Sciences, Brown University), Jonathan Evans (Centre for Thinking and Language, Department of Psychology, University of Plymouth) and Aidan Feeney (Department of Psychology, University of Durham) Very Rapid Induction of General Patterns Robert Hadley (Simon Fraser University) Similarity: a transformational approach

Ulrike Hahn, Lucy Richardson (Cardiff University) and Nick Chater (University of Warwick) A Parser for Harmonic Context-Free Grammars John Hale and Paul Smolensky (Department of Cognitive Science, The Johns Hopkins University) Models of Ontogenetic Development for Autonomous Adaptive Systems Derek Harter, Robert Kozma (University of Memphis, Institute for Intelligent Systems, Department of Mathematical Sciences) and Arthur Graesser (University of Memphis, Institute for Intelligent Systems, Department of Psychology) Representational form and communicative use Patrick G.T. Healey (Department of Computer Science, Queen Mary, University of London.), Nik Swoboda (Deprtment of Computer Science, Indiana University.), Ichiro Umata and Yasuhiro Katagiri (ATR Media Integration and Communications Laboratories.) Pragmatics at work: Formulation and interpretation of conditional instructions Denis Hilton, Jean-Franc¸ois Bonnefon (Universit´e Toulouse 2) and Markus Kemmelmeier (University of Michigan) The Influence of Recall Feedback in Information Retrieval on User Satisfaction and User Behavior Eduard Hoenkamp and Henriette van Vugt (Nijmegen Institute for Cognition and Information) Modelling Language Acquisition: Grammar from the Lexicon? Steve R. Howell and Suzanna Becker (McMaster University) The strategic use of memory for frequency and recency in search control Andrew Howes and Stephen J. Payne (Cardiff University) Conceptual Combination as Theory Formation Dietmar Janetzko (Institute of Computer Science and Social Research Dep. of Cognitive Science, University of Freiburg) Combining Integral and Separable Subspaces Mikael Johannesson (Department of Computer Science, Univeristy of Sk¨ovde, Sweden, and Lund University Cognitive Science, Lund, Sweden) Distributed Cognition in Apes Christine M. Johnson and Tasha M. Oswald (Department of Cognitive Science, UC San Diego) Cascade explains and informs the utility of fading examples to problems Randolph Jones (Colby College and Soar Technology) and Eric Fleischman (Colby College) Modelling the Detailed Pattern of SRT Sequence Learning F.W. Jones and Ian McLaren (University of Cambridge) Where Do Probability Judgments Come From? Evidence for Similarity-Graded Probability Peter Juslin, H˚akan Nilsson and Henrik Olsson (Department of Psychology, Ume˚a University) Similarity Processing Depends on the Similarities Present Mark Keane (University College Dublin), Deirdre Hackett (Educational Research Centre) and Jodi Davenport (MIT)

Constraints on Linguistic Coreference: Structural vs. Pragmatic Factors Frank Keller (Computational Linguistics, Saarland University) and Ash Asudeh (Department of Linguistics, Stanford University) Training for Insight: The Case of the Nine-Dot Problem Trina Kershaw and Stellan Ohlsson (University of Illinois at Chicago) Theory-based reasoning in clinical psychologists Nancy Kim (Yale University) and Woo-kyoung Ahn (Vanderbilt University) Effect of Exemplar Typicality on Naming Deficits in Aphasia Swathi Kiran, Cynthia Thompson and Douglas Medin (Northwestern University) Visual Statistical Learning in Infants Natasha Kirkham, Jonathan Slemmer and Scott Johnson (Cornell University) Episode Blending as Result of Analogical Problem Solving Boicho Kokinov and Neda Zareva-Toncheva (New Bulgarian University) Dissecting Common Ground: Examining an Instance of Reference Repair Timothy Koschmann (Southern Illinois University), Curtis LeBaron (University of Colorado at Boulder), Charles Goodwin (UCLA) and Paul Feltovich (Southern Illinois University) Kinds of kinds: Sources of Category Coherence Kenneth Kurtz and Dedre Gentner (Northwestern University) Learning Perceptual Chunks for Problem Decomposition Peter Lane, Peter Cheng and Fernand Gobet (University of Nottingham) The Mechanics of Associative Change Mike Le Pelley and Ian McLaren (Department of Experimental Psychology, Cambridge University) Representation and Generalisation in Associative Systems Mike Le Pelley and Ian McLaren (Department of Experimental Psychology, Cambridge University) Costs of Switching Perspectives in Route and Survey Descriptions Paul Lee and Barbara Tversky (Stanford University) A Connectionist Investigation of Linguistic Arguments from the Poverty of the Stimulus: Learning the Unlearnable John Lewis (McGill University) and Jeff Elman (University of California, San Diego) Ties That Bind: Reconciling Discrepancies Between Categorization and Naming Kenneth Livingston, Janet Andrews and Patrick Dwyer (Vassar College) Effects of multiple sources of information on induction in young children Yafen Lo (Rice University) and Vladimir Sloutsky (Ohio State University) Activating verb semantics from the regular and irregular past tense. Catherine Longworth, Billi Randall, Lorraine Tyler (Centre for Speech and Language, Dept. Exp. Psychology, Cambridge, UK.) and William Marslen-Wilson (MRC Cognition and Brain Sciences Unit, Cambridge, UK)

Towards a Theory of Semantic Space Will Lowe (Center for Cognitive Studies, Tufts University) Individual Differences in Reasoning about Broken Devices: An Eye Tracking Shulan Lu, Brent Olde, Elisa Cooper and Arthur Graesser (The University of Memphis) Modeling Forms of Surprise in an Artificial Agent Luis Macedo (Instituto Superior de Engenharia de Coimbra) and Amilcar Cardoso (Departamento de Engenharia Informatica da Universidade de Coimbra) Modeling the interplay of emotions and plans in multi-agent simulations Stacy Marsella (USC Information Sciences Institute) and Jonathan Gratch (USC Institute for Creative Technologies) Elementary school children’s understanding of experimental error Amy Masnick and David Klahr (Carnegie Mellon University) Interactive Models of Collaborative Communication Michael Matessa (NASA Ames Research Center) Testing the Distributional Hypothesis: The Influence of Context on Judgements of Semantic Similarity Scott McDonald and Michael Ramscar (Institute for Communicating and Collaborative Systems, University of Edinburgh) Activating Verbs from Typical Agents, Patients, Instruments, and Locations via Event Schemas Ken McRae (U. of Western Ontario), Mary Hare (Bowling Green State University), Todd Ferretti and Jeff Elman (U. California San Diego) Spatial Experience, Sensory Qualities, and the Visual Field Douglas Meehan (CUNY Graduate Center) How Primitive is Self-consciousness?: Autonomous Nonconceptual Content and Immunity to Error through Misidentification Roblin Meeks (The Graduate School and University Center of The City University of New York) Automated Proof Planning for Instructional Design Erica Melis (DFKI Saarbr¨ucken), Christoph Glasmacher (Department of Psychology; Saarland University), Carsten Ullrich (DFKI Saarbr¨ucken) and Peter Gerjets (Department of Psychology; Saarland University) Modeling an Opportunistic Strategy for Information Navigation Craig Miller (DePaul University) and Roger Remington (NASA Ames) Emergence of effects of collaboration in a simple discovery task Kazuhisa Miwa (Nagoya University) Effects of Competing Speech on Sentence-Word Priming: Semantic, Perceptual, and Attentional Factors Katherine Moll, Eileen Cardillo and Jennifer Utman (University of Oxford) The consistency of children’s responses to logical statements: Coordinating components of formal reasoning Bradley J. Morris and David Klahr (Carnegie Mellon University) Working-memory modularity in analogical reasoning

Robert Morrison, Keith Holyoak and Bao Truong (University of California, Los Angeles) Emotional Impact on Logic Deficits May Underlie Psychotic Delusions in Schizophrenia Lilianne Mujica-Parodi, Tsafrir Greenberg (New York State Psychiatric Institute), Robert Bilder (Nathan S. Kline Institute for Psychiatric Research) and Dolores Malaspina (New York State Psychiatric Institute) Interactions between Frequency Effects and Age of Acquisition Effects in a Connectionist Network Paul Munro (University of Pittsburgh) and Garrison Cottrell (University of California, San Diego) Modality preference and its change in the course of development Amanda Napolitano, Vladimir Sloutsky and Sarah Boysen (Ohio State Univeristy) Clustering Using the Contrast Model Daniel Navarro and Michael Lee (Department of Psychology, University of Adelaide) Active Inference in Concept Learning Jonathan Nelson (Cognitive Science Department, U. of California, San Diego), Joshua Tenenbaum (Psychology Department, Stanford University) and Javier Movellan (Cognitive Science Department, U of California, San Diego) Addition as Interactive Problem Solving Hansj¨org Neth and Stephen J. Payne (School of Psychology, Cardiff University) On the Normativity of Failing to Recall Valid Advice David Noelle (Center for the Neural Basis of Cognition) How is Abstract, Generative Knowledge Acquired? A Comparison of Three Learning Scenarios Timothy Nokes and Stellan Ohlsson (University of Illinois at Chicago) The Age-Complicity Hypothesis: A Cognitive Account of Some Historical Linguistic Data Marcus O’Toole, Jon Oberlander and Richard Shillcock (University of Edinburgh) Singular and General Causal Arguments Uwe Oestermeier and Friedrich Hesse (Knowledge Media Research Center) Roles of Shared Relations in Induction Hitoshi Ohnishi (National Institute of Multimedia Education) A model of embodied communications with gestures between humans and robots Tetsuo Ono, Michita Imai (ATR Media Integration & Communications Research Laboratories) and Hiroshi Ishiguro (Faculty of Systems Engineering, Wakayama University) Remembering to Forget: Modeling Inhibitory and Competitive Mechanisms in Human Memory Mike Oram (University of St. Andrews) and Malcolm MacLeod (University of St Andrews) The origins of syllable systems : an operational model. Pierre-yves Oudeyer (Sony CSL Paris) Prototype Abstraction in Category Learning?

Thomas Palmeri and Marci Flanery (Vanderbilt University) The role of velocity in affect discrimination Helena M. Paterson, Frank E. Pollick and Anthony J. Sanford (Department of Psychology, University of Glasgow) Graph-based Reasoning: From Task Analysis to Cognitive Explanation David Peebles and Peter Cheng (University of Nottingham) The Impact of Feedback Semantics in Visual Word Recognition: Number of Features Effects in Lexical Decision and Naming Tasks Penny Pexman (Department of Psychology, University of Calgary), Stephen Lupker (Department of Psychology, University of Western Ontario) and Yasushi Hino (Department of Psychology, Chukyo University) Category learning without labels-A simplicity approach Emmanuel Pothos (Depatment of Psychology, University of Edinburgh) and Nick Chater (Department of Psychology, University of Warwick) Neural Synchrony Through Controlled Tracking Dennis Pozega and Paul Thagard (University of Waterloo) The Conscious-Subconscious Interface: An Emerging Metaphor in HCI Aryn Pyke and Robert West (Carleton University, Ottawa, Canada) Cognitive Uncertainty in Syllogistic Reasoning: An Alternative Mental Models Theory Jeremy Quayle (University of Derby) and Linden Ball (Lancaster University) Using a Triad Judgment Task to Examine the Effect of Experience on Problem Representation in Statistics Mitchell Rabinowitz and Tracy Hogan (Fordham University) Perceptual Learning Meets Philosophy: Cognitive Penetrability of Perception and its Philosophical Implicationseption Athanassios Raftopoulos (Department of Educational Sciences, University of Cyprus) The influence of semantics on past-tense inflection Michael Ramscar (University of Edinburgh) The Emergence of Words Terry Regier, Bryce Corrigan, Rachael Cabasaan, Amanda Woodward (University of Chicago) and Michael Gasser and Linda Smith (Indiana University) A Knowledge-Resonance (KRES) Model of Category Learning Bob Rehder (New York University) and Gregory Murphy (University of Illinois) Regularity and Irregularity in an Inflectionally Complex Language: Evidence from Polish Agnieszka Reid and William Marslen-Wilson (MRC Cognition and Brain Sciences Unit) Cats could be Dogs, but Dogs could not be Cats: What if they Bark and Mew? A Connectionist Account of Early Infant Memory and Categorization Robert A.P. Reuter (Cognitive Science Research Unit, Free University of Brussels (ULB)) Motor Representations in Memory and Mental Models: Embodiment in Cognition

Daniel Richardson, Michael Spivey and Jamie Cheung (Cornell University) Language is Spatial : Experimental Evidence for Image Schemas of Concrete and Abstract Verbs Daniel Richardson, Michael Spivey, Shimon Edelman and Adam Naples (Cornell University) Efficacious Logic Instruction: People Are Not Irremediably Poor Deductive Reasoners Kelsey Rinella, Selmer Bringsjord and Yingrui Yang (Rensselaer Polytechnic Institute) Using cognitive models to guide instructional design: The case of fraction division Bethany Rittle-Johnson and Kenneth Koedinger (Carnegie Mellon University) For Better or Worse: Modelling Effects of Semantic Ambiguity Jennifer Rodd (Centre for Speech and Language, Department of Experimental Psychology, Cambridge University ), Gareth Gaskell (Department of Psychology, University of York) and William Marslen-Wilson (MRC Cognition and Brain Sciences Unit, Cambridge) A Comparative Evaluation of Socratic Versus Didactic Tutoring Carolyn Ros´e (Learning Research and Development Center, University of Pittsburgh), Johanna Moore (HCRC, University of Edinburgh), Kurt VanLehn (Learning Research and Development Center, University of Pittsburgh) and David Allbritton (Deptartment of Psychology, DePaul University) Mental Models and the Meaning of Connectives: A Study on Children, Adolescents and Adults Katiuscia Sacco, Monica Bucciarelli and Mauro Adenzato (Centro di Scienza Cognitiva, Universita’ di Torino) A Selective Attention Based Method for Visual Pattern Recognition Albert Ali Salah, Ethem Alpaydın and Lale Akarun (Bogazici University Computer Engineering Department) Solving arithmetic operations: a semantic approach Emmanuel Sander (University Paris 8 - ESA CNRS 7021) Do Perceptual Complexity and Object Familiarity Matter for Novel Word Extension? Catherine Sandhofer and Linda Smith (Indiana University) Decomposing interactive behavior Michael Schoelles and Wayne Gray (George Mason University) The Influence of Causal Interpretation on Memory for System States Wolfgang Schoppek (University of Bayreuth) Metarepresentation in Philosophy and Psychology Sam Scott (Carleton University) Connectionist modelling of surface dyslexia based on foveal splitting: Impaired pronunciation after only two half pints Richard Shillcock and Padraic Monaghan (University of Edinburgh) Assessing Generalization in Connectionist and Rule-based Models Under the Learning Constraint Thomas Shultz (McGill University) Clinging to Beliefs: A Constraint-satisfaction Model

Thomas Shultz (McGilll University), Jacques Katz (Carnegie Mellon University) and Mark Lepper (Stanford University) Semantic Effect on Episodic Associations Yaron Silberman (Interdisciplinary Center for Neural Computation, The Hebrew University of Jerusalem), Risto Miikkulainen (Department of Computer Science, The University of Texas at Austin) and Shlomo Bentin (Department of Psychology, The Hebrew University of Jerusalem) Representation: Where Philosophy Goes When It Dies Peter Slezak (University of New South Wales) Effects of linguistic and perceptual information on categorization in young children Vladimir Sloutsky and Anna Fisher (Ohio State University) The Interaction of Explicit and Implicit Learning: An Integrated Model Paul Slusarz and Ron Sun (University of Missouri-Columbia) Preserved Implicit Learning on both the Serial Reaction Time Task and Artificial Grammar in Patients with Parkinson’s Disease Jared Smith, Richard Siegert, John McDowall (Victoria University of Wellington, New Zealand) and David Abernethy (Wellington School of Medicine, University of Otago, New Zealand) On choosing the parse with the scene: The role of visual context and verb bias in ambiguity resolution Jesse Snedeker, Kirsten Thorpe and John Trueswell (Institute for Research in Cognitive Science/University of Pennsylvania) Synfire chains and catastrophic interference Jacques Sougn´e and Robert French (University of LIEGE) Human Sequence Learning: Can Associations Explain Everything? Rainer Spiegel and Ian McLaren (University of Cambridge, Department of Experimental Psychology) Effect of Choice Set on Valuation of Risky Prospects Neil Stewart, Nick Chater and Henry Stott (University of Warwick) The Fate of Irrelevant Information in Analogical Mapping Christiopher Stilwell and Arthur Markman (University of Texas, Austin) Visual Expertise is a General Skill Maki Sugimoto (HNC Software, Inc.) and Garrison Cottrell (University of California, San Diego, Department of Computer Science and Engineering) The Role of Feedback in Categorisation Mark Suret and Ian McLaren (Department of Experimental Psychology, University of Cambridge, UK) An Analogue of The Phillips Effect Mark Suret and Ian McLaren (Department of Experimental Psychology, University of Cambridge, UK) Cue-Readiness in Insight Problem-Solving

Hiroaki Suzuki, Keiga Abe (Department of Education, Aoyama Gakuin University), Kazuo Hiraki (Department of Systems Science, The University of Tokyo) and Michiko Miyazaki (Department of Human System Science, Tokyo Institute of Technology) Extending the Past-tense Debate: a Model of the German Plural Niels Taatgen (University of Groningen, department of artificial intelligence) The Modality Effect in Multimedia Instructions Huib Tabbers, Rob Martens and Jeroen van Merri¨enboer (Open University of the Netherlands, Educational Technology Expertise Centre ) Real World Constraints on the Mental Lexicon: Assimilation, the Speech Lexicon and the Information Structure of Spanish Words Monica Tamariz (Department of Linguistics, University of Edinburgh) and Richard Shillcock (Department of Cognitive Science, University of Edinburgh) The rational basis of representativeness Joshua Tenenbaum and Thomas Griffiths (Stanford University) A connectionist account of the emergence of the literal-metaphorical-anomalous distinction in young children Michael Thomas (Neurocognitive Development Unit, Institute of Child Health), Denis Mareschal (Centre for Brain and Cognitive Development, Birkbeck College) and Andrew Hinds (Department of Psychology, King Alfreds College, Winchester) A new model of graph and visualization usage Greg Trafton (NRL) and Susan Trickett (George Mason University) That’s odd! How scientists respond to anomalous data Susan Trickett (George Mason University), Greg Trafton (Naval Research Lab), Christian Schunn and Anthony Harrison (George Mason University) Spoken Language Comprehension Improves the Efficiency of Visual Search Melinda Tyler and Michael Spivey (Cornell University) “Two” Many Optimalities ` Oscar Vilarroya (Centre de Recerca en Cincia Cognitiva) Generalization in simple recurrent networks Marius Vilcu and Robert Hadley (School of Computing Science, Simon Fraser University) A Computational Model of Counterfactual Thinking: The Temporal Order Effect Clare R. Walsh and Ruth M.J. Byrne (University of Dublin, Trinity College) The Semantic Modulation of Deductive Premises Clare R. Walsh (University of Dublin, Trinity College) and P.N. Johnson-Laird (Princeton University) The Appearance of Unity: A Higher-Order Interpretation of the Unity of Consciousness

Josh Weisberg (CUNY Graduate Center) How to Solve the Problem of Compositionality by Oscillatory Networks Markus Werning (Erfurt University) A Model of Perceptual Change by Domain Integration Gert Westermann (Sony Computer Science Lab) Imagery, Context Availability, Contextual Constraint and Abstractness Katja Wiemer-Hastings, Jan Krug and Xu Xu (Northern Illinois University) Rules for Syntax, Vectors for Semantics Peter Wiemer-Hastings and Iraide Zipitria (University of Edinburgh) Did Language Give Us Numbers? Symbolic Thinking and the Emergence of Systematic Numerical Cognition. Heike Wiese (Humboldt University Berlin) Selection Procedures for Module Discovery: Exploring Evolutionary Algorithms for Cognitive Science Janet Wiles, Ruth Schulz, Scott Bolland, Bradley Tonkes and Jennifer Hallinan (University of Queensland) How learning can guide evolution in hierarchical modular tasks Janet Wiles, Bradley Tonkes and James Watson (University of Queensland) Supporting Understanding through Task and Browser Design Jennifer Wiley (Department of Psychology, University of Illinois at Chicago) Access to Relational Knowledge: a Comparison of Two Models William Wilson, Nadine Marcus (University of New South Wales, Sydney, Australia) and Graeme Halford (University of Queensland, Brisbane, Australia) What does he mean? Maria Wolters (Rhetorical Systems Ltd. ) and David Beaver (Department of Linguistics, Stanford University) Structural Determinants of Counterfactual Reasoning Daniel Yarlett and Michael Ramscar (School of Cognitive Science, University of Edinburgh) Competition between linguistic cues and perceptual cues in children’s categorization: English- and Japanese-speaking children Hanako Yoshida, Linda Smith, Cindy Drake, Joy Swanson and Leanna Gudel (Indiana University) Base-Rate Neglect in Pigeons: Implications for Memory Mechanisms Thomas Zentall and Tricia Clement (University of Kentucky)

Member Abstracts Explanations of words and natural contexts: An experiment with childrens limericks Greg Aist (Carnegie Mellon University) Understanding death as the cessation of intentional action: A cross-cultural developmental study H. Clark Barrett (Max Planck Institute for Human Development) Working Memory Processes During Abductive Reasoning Martin Baumann and Josef F. Krems (Department of Psychology, Chemnitz University of Technology) Organizing Features into Attribute Values

Dorrit Billman, Carl Blunt and Jeff Lindsay (School of Psychology, Georgia Institute of Technology) Attention Shift and Verb Labels in Event Memory Dorrit Billman (School of Psychology, Georgia Institute of Technology) and Michael Firment (Department of Psychology, Kennesaw State University) The semantics of temporal prepositions: the case of IN David Br´ee (University of Manchester) Thoughts on the Prospective MML-TP: A Mental MetaLogic-Based Theorem Prover Selmer Bringsjord and Yingrui Yang (RPI) Hemispheric Effects of Concreteness in Pictures and Words Daniel Casasanto, John Kounios and John Detre (University of Pennsylvania) Learning Statistics: The Use of Conceptual Equations and Overviews to Aid Transfer Richard Catrambone (Georgia Institute of Technology) and Robert Atkinson (Mississippi State University) Infants? Associations of Words and Sounds to Animals and Vehicles Eliana Colunga and Linda Smith (Indiana University) A Connectionist Model of Semantic Memory: Superordinate structure without hierarchies George Cree and Ken McRae (University of Western Ontario) Concept Generalization in Separable and Integral Stimulus Spaces Nicolas Davidenko and Joshua Tenenbaum (Stanford University) Linguistic Resources and “Ontologies” across Sense Modalities: A Comparison between Color, Odor, and Noise and Sound Daniele Dubois (LCPE/ CNRS) and Caroline Cance (Universit´e de Paris 3 & LCPE) What was the Cause? Children’s Ability to Catgeorize Inferences Michelle Ellefson (Southern Illinois University) Structural Alignment in Similarity and Difference of Simple Visual Stimuli Zachary Estes and Uri Hasson (Princeton University) Music Evolution: The Memory Modulation Theory Steven Flinn (ManyWorlds, Inc.) Language affects memory, but does it affect perception? Michael Frank and Lera Boroditsky (Stanford University) Pragmatic Knowledge and Bridging Inferences Raymond, W. Gibbs (University of California, Santa Cruz) and Tomoko Matsui (International Christian University) The AMBR Model Comparison Project: Multi-tasking, the Icarus Federation, and Concept Learning Kevin Gluck and Michael Young (Air Force Research Laboratory) Does Adult Category Verification Reflect Child-like Concepts? Robert Goldberg (University of Pittsburgh) Imagining the Impossible James Hampton, Alan Green (City University, London) and Zachary Estes (Princeton University) Understanding Negation - The Case of Negated Metaphors

Uri Hasson and Sam Glucksberg (Princeton University) Neural Networks as Fitness Evaluators in Genetic Algorithms: Simulating Human Creativity Vera Kempe (University of Stirling), Robert Levy and Craig Graci (State University of New York at Oswego) Modeling the Effect of Category Use on Learning and Representation Kenneth Kurtz, John Cochener and Douglas Medin (Northwestern University) Towards a Multiple Component Model of Human Memory:A Hippocampal-Cortical Memory Model of Encoding Specificity Kenneth Kwok and James McClelland (Carnegie Mellon University and Center for the Neural Basis of Cognition) Categorical Perception as Adaptive Processing of Complex Visuo-spatial Configurations in High-level Basket-ball Players Eric Laurent (University of the Mediterranean), Thierry Ripoll (University of Provence) and Hubert Ripoll (University of the Mediterranean) Configural and Elemental Approaches to Causal Learning Mike Le Pelley, S. E. Forwood and Ian McLaren (Department of Experimental Psychology, Cambridge University) Levels of Processing and Picture Memory: An Eye movement Analysis Yuh-shiow Lee (Dept. of Psychology, National Chung-Cheng University) An Alternative Method of Problem Solving: The Goal-Induced Attractor William Levy and Xiangbao Wu (University of Virginia) Sub Space: Describing Distant Psychological Space Eliza Littleton (Aptima, Inc.), Christian Schunn (George Mason University) and Susan Kirschenbaum (Naval Undersea Warfare Center) A Criticism of the Conception of Ecological Rationality Daniel Hsi-wen Liu (Providence University) Thnking through Doing: Manipulative Abduction? Lorenzo Magnani (University of Pavia, Pavia, Italy and Georgia Institute of Technology, Atlanta, USA) Spatial priming of recognition in a virtual space Gareth Miles and Andrew Howes (Cardiff University) The frequency of connectives in preschool children’s language environment Bradley J. Morris (Carnegie Mellon University) A Soar model of human video-game players Hidemi Ogasawara (School of Computer and Cognitive Science, Chukyo University) and Takehiko Ohno (Communication Science Laboratories, NTT) Practical Cognition in the Assessment of Goals Luis Angel P´erez-Miranda (The University of the Basque Country (UPV-EHU)) Exceptional and temporal effects in counterfactual thinking Susana Segura (University of Malaga) and Rachel McCloy (University of Dublin) Children’s Algorithmic Sense-making through Verbalization

Hajime Shirouzu (School of Computer and Cognitive Sciences, Chukyo University) Prosodic Guidance: Evidence for the Early Use of A Capricious Parsing Constraint Jesse Snedeker and John Trueswell(Institute for Research in Cognitive Sciene/University of Pennsylvania) Learning and Memory: A Cognitive Approach About The Role of Memory in Text Comprehension Adriana Soares and Carla Corrˆea (Universidade Estadual do Norte Fluminense) SARAH: Modeling the Results of Spiegel and McLaren (2001) Rainer Spiegel and Ian McLaren (University of Cambridge, Department of Experimental Psychology) The Relationship between Learned Categories and Structural Alignment Daisuke Tanaka (Department of Psychology, University of Tokyo) Timing and Rhythm in Multimodal Communication for Conversational Agents Ipke Wachsmuth (University of Bielefeld) Training Task-Switching Skill in Adults with Attention Deficit Hyperactivity Disorder Holly White (University of Memphis) and Priti Shah (University of Michigan) Advantages of a Visual Representation for Computer Programming Kirsten Whitley, Laura Novick and Doug Fisher (Vanderbilt University) Mass and Count in Language and Cognition: Some Evidence from Language Comprehension Heike Wiese (Humboldt-University Berlin) and Maria Pi˜nango (Yale University) Inhibition mechanism of phonological short-term memory in foreign language processing Takashi Yagyu (Department of psychology, University of Tokyo) Odd-Even effect in multiplication revisited: The role of equation presentation format Michael Yip (School of Arts & Social Sciences, The Open University of Hong Kong, Hong Kong SAR)

Symposium Abstracts

Computational Models of Historical Scientific Discoveries Pat Langley, Institute for the Study of Learning and Expertise Lorenzo Magnani, Department of Philosophy, University of Pavia Peter C.-H. Cheng, School of Psychology, University of Nottingham Adrian Gordon, Department of Computing, University of Northumbria Sakir Kocabas, Space Engineering Department, Istanbul Technical University Derek H. Sleeman, Department of Computing Science, University of Aberdeen The discovery of scientific knowledge is one of the most challenging tasks that confront humans, yet cognitive science has made considerable progress toward explaining this activity in terms of familiar cognitive processes like heuristic search (e.g., Langley et al., 1987). A main research theme relies on selecting historical discoveries from some discipline, identifying data and knowledge available at the time, and implementing a computer program that models the processes that led to the scientists’ insights. The literature on computational scientific discovery includes many examples of such studies, but initial work in this tradition had some significant drawbacks, which we address in this symposium. One such limitation was that early research in law discovery ignored the influence of domain knowledge in guiding search. For example, Gordon et al. (1994) noted that attempts to fit data from solution chemistry in the late 1700s took into account informal qualitative models like polymerization and dissociation. They have developed Hume, a discovery system that draws on such qualitative knowledge to direct its search for numeric laws. Hume utilizes this knowledge not only to rediscover laws found early in the history of solution chemistry, but also to explain, at an abstract level, the origins of other relations that scientists proposed and later rejected. Early discovery research also downplayed the role of diagrams, which occupy a central place in many aspects of science. For example, Huygens’ and Wren’s first presentations of momentum conservation took the form of diagrams, suggesting they may have been instrumental in the discovery process. In response, Cheng and Simon (1992) have developed Huygens, a computational model for inductive discovery of this law that uses a psychologically plausible diagrammatic approach. The system replicates the discovery by manipulating geometric diagrams that encode particle collisions and searching for patterns common to those diagrams. The quantitative data given to the system are equivalent to those available at the time of the original discovery. Another challenge concerns the computational modeling of extended periods in the history of science, rather than isolated events. To this end, Kocabas and Langley (1995) have developed BR4, an account of theory revision in particle physics that checks if the current theory is consistent (explains observed reactions) and complete (forbids unobserved reactions), revises quantum values

and posits new particles to maintain consistency, and introduces new properties to maintain completeness. BR-4 models, in abstract terms, major developments in particle physics over two decades, including the proposal of baryon and lepton numbers, postulation of the neutrino, and prediction of numerous reactions. Background knowledge about symmetry and conservation combine with data to constrain the search for an improved theory in a manner consistent with the incremental nature of historical discovery. We hope this symposium will encourage additional research that extends our ability to model historical scientific discoveries in computational terms.

References Cheng, P. C.-H. and Simon, H. A. (1992). The right representation for discovery: Finding the conservation of momentum. In Proceedings of the Ninth International Conference on Machine Learning, pages 62–71, San Mateo, CA. Morgan Kaufmann. Gordon, A., Edwards, P., Sleeman, D., and Kodratoff, Y. (1994). Scientific discovery in a space of structural models. In Proceedings of the Sixteenth Annual Conference of the Cognitive Science Society, pages 381– 386, Atlanta. Lawrence Erlbaum. Kocabas, S. and Langley, P. (1995). Integration of research tasks for modeling discoveries in particle physics. In Proceedings of the AAAI Spring Symposium on Systematic Methods of Scientific Discovery, pages 87–92, Stanford, CA. AAAI Press. Langley, P., Simon, H. A., Bradshaw, G. L., and Zytkow, J. M. (1987). Scientific discovery: Computational explorations of the creative processes. MIT Press, Cambridge, MA.

Symposium: When Cognition Shapes its Own Environment Peter Todd ([email protected]) Center for Adaptive Behavior and Cognition, Max Planck Institute for Human Development, Berlin, Germany.

Simon Kirby ([email protected]) and James R Hurford ([email protected]) Language Evolution and Computation Research Unit, Department of Theoretical and Applied Linguistics, University of Edinburgh, 40 George Square, Edinburgh, EH8 9LL, UK.

Introduction Cognitive mechanisms are shaped by their environments, both through evolutionary selection across generations and through learning and development within lifetimes. But by making decisions that guide actions which in turn alter the surrounding world, cognitive mechanisms can also shape their environments in turn. This mutual shaping interaction between cognitive structure and environment structure can even result in coevolution between the two over extended periods of time. In this symposium, we explore how simple decision heuristics can exploit the information structure of the environment to make good decisions, how simple language-learning mechanisms can capitalize on the structure of the ”spoken” environment to develop useful grammars, and how both sorts of cognitive mechanisms can actually help build the very environment structure that they rely on to perform well.

Programme There will be three talks, as follows: 1. Peter Todd, “Simple Heuristics that exploit environment structure”, Traditional views of rational decision making assume that individuals gather, evaluate, and combine all the available evidence to come up with the best choice possible. But given that human and animal minds are designed to work in environments where information is often costly and difficult to obtain, we should instead expect many decisions to be made with simple ”fast and frugal” heuristics that limit information use. In our study of ecological rationality, we have been exploring just how well such simple decisionmaking heuristics can do when they are able to exploit the structure of information in specific environments. This talk will outline the research program pursued by the Center for Adaptive Behavior and Cognition as developed in the book, Simple Heuristics That Make Us Smart (Oxford, 1999), and highlight how the match between cognitive mechanism structure and environment structure allows the Recognition heuristic and Take The Best heuristic to perform on par with traditionally rational decision mechanisms.

2. Simon Kirby, “The Iterated Learning Model of Language Evolution”, The past decade has seen a shift in the focus of research on language evolution away from approaches that rely solely on natural selection as an explanatory mechanism. Instead, there has been a growing appreciation of languages (as opposed to the language acquisition device) as complex adaptive systems in their own right. In this talk we will present an approach that explores the relationship between biologically given language learning biases and the cultural evolution of language. We introduce a computationally implemented model of the transmission of linguistic behaviour over time: the Iterated Learning Model (ILM). In this model there is no biological evolution, natural selection, nor any measurement of the success of communication. Nonetheless, there is significant evolution. We show that fully syntactic languages emerge from primitive communication systems in the ILM under two conditions specific to Hominids: (i) a complex meaning space structure, and (ii) the poverty of the stimulus. 3. Peter Todd, Simon Kirby and Jim Hurford, “Putting the Models Together: how the environment is shaped by the action of the recognition heuristic”, To explore how cognitive mechanisms can exert a shaping force on their environment and thus affect their own performance, we begin by considering the actions of a very simple cognitive mechanism, the recognition heuristic for making choices. This heuristic specifies that when choosing between two options, one of which is recognized and one not, the recognized option should be selected. The recognition heuristic makes good choices, in environments where recognition is correlated with the choice criterion. Many natural environments have this structure, but such structure can also be “built”: By using the recognition heuristic, agents can create an environment in which some objects are much more often and “talked about” and recognized than others. An agent-based simulation is used to show what behavioral factors affect the emergence of this environmental structure.

The Cognitive Basis of Science: The View from Science Session Organizer: Nancy J. Nersessian ([email protected]) College of Computing, 801 Atlantic Drive Atlanta, GA 30332 USA The issue of the nature of the processes or “mechanisms” that underlie scientific cognition is a fundamental problem for cognitive science. A rich and nuanced understanding of scientific knowledge and practice must take into account how human cognitive abilities and limitations afford and constrain the practices and products of the scientific enterprise. Reflexively, investigating scientific cognition opens the possibility that aspects of cognition previously not observed or considered will emerge and require enriching or even altering significantly current understandings of cognitive processes.

The Baby in the Lab Coat: Why child development is an inadequate model for understanding the development of science Stephen P. Stich, Department of Philosophy, Rutgers University In two recent books and a number of articles, Alison Gopnik and her collaborators have proposed a bold and intriguing hypothesis about the relationship between scientific cognition and cognitive development in childhood. According to this view, the processes underlying cognitive development infants and children and the processes underlying scientific cognition are identical. One of the attractions of the hypothesis is that, if it is correct, it will unify two fields of investigation – the study of early cognitive development and the study of scientific cognition – that have hitherto been thought quite distinct, with the result that advances in either domain will further our understanding of the other. In this talk we argue that Gopnik’s bold hypothesis is untenable. More specifically, we will argue that if Gopnik and her collaborators are right about cognitive development in early childhood then they are wrong about science. The minds of normal adults and of older children, we will argue, are more complex than the minds of young children, as Gopnik portrays them. And some of the mechanisms that play no role in Gopnik’s account of cognitive development in childhood play an essential role in scientific cognition.

Scientific Cognition as Distributed Cognition Ronald N. Giere, Center for Philosophy of Science, University of Minnesota I argue that most important cases of cognition in contemporary science are best understood as examples of distributed cognition. Here I focus exclusively on the acquisition of new knowledge as the paradigm of scientific cognition. Scientific cognition, then, does not reduce to mere distributed computation. The simplest case is that in which

two people cooperate in acquiring some knowledge that is not directly acquired by either one alone. It is even possible that neither person could physically perform the task alone. This is an example of what has been called “socially shared cognition” (Resnick) or “collective cognition” (Knorr). The most elaborate example is the case of experimental highenergy physics at CERN, as described by the sociologist, Karin Knorr in her recent book, Epistemic Cultures. I go beyond Knorr’s analysis to include the particle accelerator and related equipment as part of a distributed cognitive system. So here the cognition is distributed both among both people and artifacts. Such artifacts as diagrams and graphics and even abstract mathematical constructions are also included as components of distributed cognitive systems. This makes it possible to understand the increasing power of science since the seventeenth century as in large measure due to the creation of increasing powerful cognitive systems, both instrumental and representational.

The Cognitive Basis of Model-based Reasoning in Science Nancy J. Nersessian, Program in Cognitive Science, Georgia Institute of Technology Although scientific practice is inherently “socially shared cognition,” the nature of individual cognitive abilities and how these constrain and facilitate practices still needs to be figured into the account of scientific cognition. This presentation will focus on the issue of the cognitive basis of the model-based reasoning practices employed in creative reasoning leading to conceptual change across the sciences. I will first locate the analysis of model-based reasoning within the mental modeling framework in cognitive science and then discuss the roles of analogy, visual representation, and thought experimenting in constructing new conceptual structures. A brief indication of the lines along which a fuller account of how the cognitive, social, and material are fused in the scientist’s representations of the world will be developed. That the account needs to be rooted in the interplay between the individual and the communal in the model-based reasoning that takes place in concept formation and change. Modeling is a principal means through which a scientist transports conceptual resources drawn from her wider cultural milieu into science and transmits novel representations through her community. Scientific modeling always takes place in a material environment that includes the natural world, socio-cultural artifacts (stemming from both outside of science and within it), and instruments devised by scientists and communities to probe and represent that world. Symposium Discussant: Dedre Gentner, Department of Psychology, Northwestern

The Interaction of Explicit and Implicit Learning Ron Sun ([email protected]) University of Missouri-Columbia Columbia, MO 65203

Robert Mathews ([email protected]) Louisiana State University, Baton Rouge Baton Rouge, LA

The Focus of the Symposium The role of implicit learning in skill acquisition and the distinction between implicit and explicit learning have been widely recognized in recent years (see, e.g., Reber 1989, Stanley et al 1989, Willingham et al 1989, Anderson 1993), Although implicit learning has been actively investigated, the complex and multifaceted interaction between the implicit and the explicit and the importance of this interaction have not been universally recognized; to a large extent, such interaction has been downplayed or ignored, with only a few notable exceptions. 1 Research has been focused on showing the lack of explicit learning in various learning settings (see especially Lewicki et al 1987) and on the controversies stemming from such claims. Similar oversight is also evident in computational simulation models of implicit learning (with few exceptions such as Cleeremans 1994 and Sun et al 2000). Despite the lack of studies of interaction, it has been gaining recognition that it is difficult, if not impossible, to find a situation in which only one type of learning is engaged (Reber 1989, Seger 1994, but see Lewicki et al 1987). Our review of existing data has indicated that, while one can manipulate conditions to emphasize one or the other type, in most situations, both types of learning are involved, with varying amounts of contributions from each (see, e.g., Sun et al 2000; see also Stanley et al 1989, Willingham et al 1989). Likewise, in the development of cognitive architectures (e.g., Rosenbloom et al 1993, Anderson 1993), the distinction between procedural and declarative knowledge has been proposed for a long time, and advocated or adopted by many in the field (see especially Anderson 1993). The distinction maps roughly onto the distinction between the explicit and implicit knowledge, because procedural knowledge is generally inaccessible while declarative knowledge is generally accessible and thus explicit. However, in work on cognitive architectures, focus has been almost exclusively on “top-down” models (that is, learning first explicit knowledge and then implicit knowledge on the basis of the former), the bottom-up direction (that is, learning first implicit knowl1 By the explicit, we mean processes involving some form of generalized (or generalizable) knowledge that is consciously accessible.

edge and then explicit knowledge, or learning both in parallel) has been largely ignored, paralleling and reflecting the related neglect of the interaction of explicit and implicit processes in the skill learning literature. However, there are a few scattered pieces of work that did demonstrate the parallel development of the two types of knowledge or the extraction of explicit knowledge from implicit knowledge (e.g, Willingham et al 1989, Stanley et al 1989, Sun et al 2000), contrary to usual top-down approaches in developing cognitive architectures. Many issues arise with regard to the interaction between implicit and explicit processes, which we need to look into if we want to better understand this interaction: How can we best capture implicit processes computationally? How can we best capture explicit processes computationally? How do the two types of knowledge develop along side each other and influence each other’s development? Is bottom-up learning (or parallel learning) possible, besides top-down learning? How can they (bottom-up learning, top-down learning, and parallel learning) be realized computationally? How do the two types of acquired knowledge interact during skilled performance? What is the impact of that interaction on performance? How do we capture such impact computationally?

Titles of the Talks Axel Cleeremans: “Behavioral, neural, and computational correlates of implicit and explicit learning” Zoltan Dienes: “The effect of prior knowledge on implicit learning” Bob Mathews: “Finding the optimal mix of implicit and explicit learning” Ron Sun: “The synergy of the implciit and the explicit”

Papers and Posters

The Roles of Thought and Experience in the Understanding of Spatio-temporal Metaphors Tracy Packiam Alloway ([email protected]) Department of Psychology University of Edinburgh, 7 George Square Edinburgh EH8 9JZ, UK

Michael Ramscar ([email protected]) School of Cognitive Science, Division of Informatics University of Edinburgh, 2 Buccleuch Place, Edinburgh, EH8 9LW

Martin Corley ([email protected]) Department of Psychology University of Edinburgh, 7 George Square Edinburgh EH8 9JZ, UK

Abstract Spatial and temporal metaphors are often used interchangeably, and thus, offer a unique way of exploring the relationship between language and thought. Both spatial and temporal speaking incorporates two systems of motion. The first is an ego-moving system, when the individual moves from one point to another, spatially, or from the past to the future, temporally. The second is the object- (or time-) moving system, when the individual is stationary and observes objects, or time, moving towards him/her. This study explored the effect of a spatial environment on the ambiguous temporal question: Next Wednesday’s meeting has been moved forward two days--What day is the meeting now? Results reveal that when participants are immersed in an ego-moving spatial environment, such as a virtual reality game, and receive a prime that causes them to think in an object-moving way, they are more likely to perform a target task in a way consistent with the way they have been primed to think, although it contradicts the spatial motion they subsequently experience in the testing environment.

Introduction What is the relationship between language and sensory experience? According to one recent claim (Lakoff and Johnson, 1999), abstract concepts, such as time, are substrated in concrete concepts like space that can be experienced directly. The representations of these concrete concepts are formed directly, by experience. Thus, our spatial experiences form "a neural structure that is actually part of, or makes use of, the sensorimotor system of our brains. Much of conceptual inference is, therefore, sensorimotor inference" (Lakoff and Johnson, 1999, p. 20). On this view, our understanding of concepts such as time is predicated on our spatial experiences, and thus the idea of motion in time relies on our understanding of motion in space.

There is evidence for this relationship between motion in space and time in the structure of language. We can talk of putting things forward in time, as well as moving forward through space (see Lakoff & Johnson, 1999; 1980). According to Lakoff and Johnson’s (1980; 1999) Conceptual Metaphor hypothesis, metaphors are not justa manner of speaking but a deeper reflection of human thought processes. Metaphoric speaking is reflective, say Lakoff and Johnson, of deeper conceptual mappings that occur in our thinking and is depicted as an over-arching and general metaphor termed asthe Conceptual Metaphor. Consider the following statements: Your claims are indefensible. He attacked every weak point in my argument. He shot down all of my arguments. According to the Conceptual Metaphor (metaphoric representation) hypothesis when we use statements such as these we are making use of a larger conglomerate metaphor, in this instance, ARGUMENT IS WAR.1 The thrust of the Conceptual Metaphor argument is as follows: arguments are similar to wars in that there are winners and losers, positions are attacked and defended, and one can gain or lose ground. The theory of Conceptual Metaphor suggests that we process metaphors by mapping from a base domain to a target domain. In this particular example, the base domain is ARGUMENT IS WAR and the target domain is asubordinate metaphor such as Your claims are indefensible.

Motion in Space and Time Lakoff and Johnson extend the idea of Conceptual Metaphor to spatio-temporal metaphors by invoking the 1

Following Lakoff and Johnson’s convention (1980), all Conceptual Metaphors are typed in the uppercase to distinguish them from the subordinate metaphors

locative terms of FRONT/BACK to represent how we view time and space. FRONT is assigned on the assumption of motion (Fillmore, 1978). According to this theory, in the ego-moving system, FRONT is used to designate a future event because the ego is moving forward and encounters the future event in front of him. In the time-moving system, the FRONT term denotes a past event where the ego or the individual is stationary but the events are moving. Thus it is possible to define (at least) two schemas of motion in space. 1) Object-Moving Metaphor (OM) In this schema of motion,the individual is seen as stationary and objects seem to come towards him/her. For an example of this schema, consider an individual waiting at a bus stop and observing vehicles coming towards him/her. In this schema of motion, the individual assigns the term FRONT to the object closest towards him. In the diagram below, the term FRONT would be assigned to the white rock.

1) Time Moving metaphor (TM) The motion of time provides the framework in which temporal metaphors are comprehended. In this schema, front, or ahead is determined by thefuture moving to the past. For example, in the month of February, Christmas is now in the future. In time it will move to the present and then to the past (e.g. Christmas is coming). The individual is a stationary observer as times "flows" past. This schema is the temporal equivalent of the OM metaphor in the domain of space. 2) Ego-Moving metaphor (EM) The ego or the individual moves from the past to the future such as the sentence His vacation to the beach lay ahead of him. In this metaphor, the observer is seen as moving forward through time, passing temporal events that are seen as stationary points. It is thus the temporal equivalent of the spatial EM system, where the observer moves forward through space When discussing motion in time, temporal events are viewed as points or locations in space, and a similar rationale is used when assigning deictic terms such as front and back. For example, in the EM system, FRONT is used to designate a future event because the ego is moving forward and encounters the future event in front of him, while in the TM system the FRONT term denotes a past event where the ego or the individual is stationary but the events are moving.

Figure 1 2) Ego-Moving Metaphor (EM) Inthis schema of motion, the objects are stationary and it is the individual that is in motion. Here, the term FRONT would be assigned to the object furthest away from the individual. In the picture below, it is the black rock that would be labeled as FRONT.

Figure 2 Thus in the EM system, front is used to designate an object furthest away from the individual, as the trajectory of motion is in that direction. While in the OM system, the term front is assigned to the object closest to the individual.

Motion in Time The schemas of motion represented in the domain of time reflect the representation of motion in the domain of space.

Studies of Spatio-temporal Metaphors Gentner and Imai (1992), and McGlone and Harding (1998) confirmed the idea that the different schemas of motion (EM and TM in the domain of time) are indeed psychologically real systems.Gentner and Imai found that participants responded faster to questions that were schema consistent with regards to temporal schemas in priming than to questions that were inconsistent with their primes. Gentner and Imai argue that this supports the theory that metaphors are mapped in distinct schemas: the shift from one schema to another causes a disruption in the processing, reflected in increased processing time. They argue that their study indicates that the relations between space and time are reflective of a psychologically real conceptual system as opposed to an etymological relic.2 A study by McGlone and Harding (1998) involved participants answering questions about days of the week - relative to Wednesday - which were posed in either the ego-moving or the time-moving metaphor. Egomoving metaphor trials comprised statements such as We passed the deadline two days ago, whilst timemoving metaphor trials involved statements such as The deadline passed us two days ago; in each case, 2

Although McGlone and Harding (1998) criticised some aspects of Gentner and Imais methodology, their corrected replication of the original study confirms its findings.

participants read the statements and were then asked to indicate the day of the week that a given event had occurred or was going to occur. At the end of each block of such priming statements, participants read an ambiguous statement, such as The reception scheduled for next Wednesday has been moved forward two days 3 and then were asked to indicate the day of the week on which this event was now going to occur. Participants who had answered blocks of priming questions about statements phrased in a way consistent with theego-moving metaphor tended to disambiguate moved forward in a manner consistent with the egomoving system (they assigned forward - the front - to the future, and hence thought the meeting had been rescheduled for Friday), whereas participants who had answered blocks of questions about statements phrased a way consistent with the time-moving metaphor tended to disambiguate moved forward in a manner consistent with the time-moving system (they assigned forward - the front - to the past, and hence thought the meeting had been re-scheduled for Monday). This work has been further developed in a recent set of experiments by Boroditsky (2000) which explicitly explored the relationship between the domains of space and time. Boroditsky found that temporal priming significantly influenced temporal reasoning in a crossdomain extension of the paradigm used in earlier experiments. Spatially priming participants with the ego moving schema led them to infer that an ambiguous meeting ("Next Wednesday’s meeting has been moved forwards two days") had been moved to Friday, whereas spatially priming participants with the object moving schema led them to assign the meeting to Monday. This study provides good evidence to support the notion that our representation of motion in space is mapped on to our understanding of motion in time, although it leaves open the question of what is directing this representational mapping spatial representations that are contiguous with our embodied experience, or functionally separable, abstract conceptual representations of space and time.

Experiment 1 This experiment directly explores the claim that our embodied experiences in space direct our conceptual understanding of time. Participants were immersed in an embodied environment, a virtual reality game, and were presented with an ambiguous spatial task, either after either a purely embodied prime, or after embodied priming during which a linguistic prime had cued them to think in terms of a contrary spatial schema. The experiment was designed to explore the role of experience and thought between the two schemas of motion in the domain of space. 3

All trials were conducted on a Wednesday.

Participants 61 University of Edinburgh students volunteered to take part in this experiment.

Materials In order to create a particularly convincing Ego Moving environment, participants played a slightly modified version of a pre-existing section of the virtual reality computer game, U n R e a l . This is a first person perspective game and involves the participant walking through a courtyard environment to complete a task. All monsters and other artifacts of the game that were not relevant to the experiment were removed from this section of the game. The objects in the target task appeared upon completion of the game. These were two chests, with no discernible front or back (unlike other objects, such as a car, or a TV), one of which was closer to the player than the other. The game was projected onto a 368cm by 282cm size screen in order to magnify the virtual effects of the game.

Procedure Pre-Test 25 participants were tested individually seated in front of the projector screen. The game was set at the point in front of the two chests. Participants did not play the game and were only instructed to Move to the front chest . In this condition, the target task was performed in isolation, and the results provided a baseline for how the term front in this task is interpreted. Out of the twenty-five subjects, twelve of them interpreted the term front to refer to the chest closest to them, while the rest assigned front to the chest furthest from them, confirming the ambiguity of the assignment of front in the target task. Experimental Conditions 36 participants were tested individually. They were asked to fill in a brief questionnaire requesting demographic information, as well as familiarity with video games and computers. At the end of the questionnaire were the following instructions:Your task is to find the location of a young woman. Try your best to navigate around the environment in order to find her. During this game, it is important to try to remember some key landmarks, such as a pair of brightly coloured pillars as you enter a path, as well as the doors on the buildings. After you have been playing for some time, you will hear a question requiring a true or false answer. This question will be about the game. Try to answer it correctly and speak your answer loudly." The participants were then shown how to use the arrow keys on the keyboard when navigating through the environment and then left alone to play the game. (The experimenter was on hand, should the volunteers

have any difficulty maneuvering around the environment; however, all volunteers seemed adequately proficient at navigating around the environment.) There were two experimental conditions. In the first condition, volunteers received a pre-recorded true/false question specific to the assignment of the term front approximately four minutes into playing the game. The question they were posed -- During the game, the green pillar is in front of the red pillar — prompted them to think in an Object Moving manner about space (the green pillar was closer to the participants than the red pillar in the game environment, thus this question is true from an OM perspective).4 We were interested to see if the thinking in an OM way in answering the question would result in a different assignment of front from the EM perspective that was embodied in the game. The order of the pillars in the question was reversed for half of the participants to counter-act an affirmative response bias. Thus, half the participants answered the true/false question:During the game, the red pillar is in front of the green pillar. The answer to this question was false from an OM perspective. In the second condition, volunteers received a prerecorded non-spatial question rather than a spatial prime approximately four minutes into playing the game. They had to provide a true or false answer to the following question: During the game, most of the doors are open". The correct answer to this question was true, however, the amount of doors the volunteer saw depended on the route he or she chose in navigating around the environment to complete the task. However, the question was also presented in the inverse to avoid any particular response bias, and half of the participants in this condition answered the following question: During the game, most of the doors are closed". The question in this condition served as a control to ensure that simply answering a question would not cause people to re-represent their perspective of front or back (but that rather a question must cause people to specifically think in a way that involves a representation of front/back for this to occur). Playing the EM game served as the embodied prime in this condition. Once the participants had completed the task, the virtual young woman they sought congratulated them and they were asked to complete the target task: "Move to the front chest". The two chests were located on the left of the virtual woman and were added from the 4

Pre-testing had shown that this question, which is true from an OM perspective, was unambiguous and ordinarily answered from the OM perspective. Out of 20 participants, 90% allocated the term front in an OM perspective. A binomial test confirmed this as significant; p variance(dim 3) > variance(dim 2), with the variance for the remaining 6 dimensions equal to the variance along dimension 3. All uni-dimensional rules that best separate two categories with the same mean on the other two relevant dimensions have an accuracy of 90% (i.e., category overlap along each pair of dimensions was set to 10%). Procedure Participants were told that they were to learn five different categories that were equally represented during each learning session. Participants were instructed that they may or may not need to use all the dimensions available to them. Participants were run over consecutive days until learning curves leveled off. Each day consisted of 20 blocks with 50 trials per block (for a total of 1000 trials per day). Stimulus display was response terminated and corrective feedback was given after every trial. Thus, if a subject responded ‘A’ to an exemplar from category B, a low tone sounded followed by a ‘B’ appearing on the screen. In addition, overall percent correct was given after every learning block. A cue preference task (Experiment 1b) was administered to participants the day after learning ended. The cue preference day began with a practice block in which participants simply categorized 50 stimuli as they had done on previous days. The practice block was followed by twelve blocks, each consisting of 50 trials. Each trial began with the presentation of one of the three relevant dimensions. Participants then made a categorization judgment based on only that one dimension. After making a judgment, participants chose another dimension and then made another categorization judgment. Thus, two judgments were made for the same stimulus. The first judgment was based on only one experimenter-chosen dimension, while the second judgment was based on two dimensions. No feedback was given during the last twelve blocks of the test day.

Stimuli and Materials Stimuli were generated using the GRT Toolbox (Alfonso-Reese, 1995). Values along every dimension were transformed from number of dots per square into actual screen coordinates. Each dimension was represented as a texture in one of nine possible squares on a computer screen. The location of the three relevant dimensions was different for each subject with the constraint that the center square (in a 3x3 grid) will never be one of the relevant dimensions. Stimuli were presented on a SuperMac Technology 17T Color Display driven by a Power Macintosh G3 running a Psychophysics Toolbox (Brainard, 1997) and lowlevel VideoToolbox (Pelli, 1997) within MATLAB (The MathWorks, Inc., 1998). Each participant sat 18 inches from the monitor. The height of the center square of the stimuli was constrained such that visual angle was less than 2°.

Results and Discussion Experiment 1A Learning for three of the four participants reached asymptote after five days, while the fourth participant required six days. Participants 1, 2, and 3 achieved an overall accuracy of approximately 70% by the last day, while Participant 4 only achieved an overall accuracy of approximately 60% on the last day. The optimal percent correct was 81.9%. Participants’ responses for the last day (without the first block) were randomly split into two halves (training and testing sets) five times. Each split was constrained to contain approximately the same number of stimuli from each category. The Categorization by Elimination algorithm, the Deterministic Generalized Context Model (see Ashby & Maddox, 1993), and six versions of Decision Bound Theory were fit to each participant’s training set responses. For CBE, low and high values of each bin along each dimension, as well as the cue order, were estimated from the responses in the training set. The parameters estimated for GCM were the sensitivity parameter, an attention weight for each dimension, the bias towards each category, and the gamma parameter (which is a measure of response selection). For fitting the GCM, a Euclidean-Gaussian distance-similarity metric was used (see Maddox & Ashby, 1998). The six versions of DBT were all Independent Decisions Classifiers, which is a special case of Decision Bound Theory in which each dimension is assumed to be independent of the other dimensions (see Ashby & Gott, 1988; Ashby & Maddox, 1990). This version of DBT was used since the best fitting bound (to separate the categories) is perpendicular to each of the three relevant dimensions. In the versions of the Independent Decisions Classifier tested here, one criterion is placed along one dimension. Two criteria are then placed along a second dimension and four criteria are placed along the third dimension. All

Table 1: AIC Scores for Experiment 1A

GCM DBT CBE

P1 Train 585.4 594.74 646.28

P1 Test 633.6 638.16 643.59

P2 Train 739.42 742.63 638.32

P2 Test 823.08 780.87 640.36

possible combinations of the three relevant dimensions were tested. As mentioned earlier, all three models were fit to part of the data set (the training set) and the best fitting parameters estimates were obtained. These parameters were then used to determine the mo dels’ accuracy on the remaining data (the testing set). A potential problem with multi-parameter models is that these models may be prone to overfit the data. That is, they actually fit the noise present in the data in order to achieve high accuracy. Training the model on a subset of the data and testing the model on the rest of the data may assess a model’s “true” performance. The AIC goodness-of-fit statistic was used to compare the fits of the three models. AIC(M i ) = -2lnLi + 2v i Where lnLI refers to the negative log likelihood value for model M i obtained through maximum likelihood estimation and v i refers to the number of free parameters in model Mi . The smaller the AIC score, the closer the model is to the “true” model (Ashby, 1992). Goodness-of-fit values for each participant (averaged over the five training and five testing sets) are shown in Table 1. Each row corresponds to one of the three models while each column refers to each participant’s training and testing sets. The generalized context model was best able to account for Participant 1’s training and testing data. Categorization by elimination was best able to account for Participant 2, Participant 3, and Participant 4’s training and testing data. Experiment 1B Experiment 1b was designed to answer two questions. First, how well can humans perform in a categorization task when dimensionality is reduced? Second, what are the properties of the dimensions preferred by humans? Obviously, one of the most important features of a cue is how accurate that cue is in categorizing objects when used alone. Another property of cues is the range of values possible, that is, the variance of a cue. It seems reasonable to assume that humans are able to learn the accuracy of various cues and would use those cues that are more accurate. Given this assumption, all three of the relevant dimensions are equally accurate when used alone. However, the question of whether humans prefer to use cues with more or less variance is addressed by having different variances for the three relevant dimensions.

P3 Train 647.33 645.32 624.5

P3 Test 687.14 665.22 634.86

P4 Train 814.4 809.55 656.04

P4 Test 835.24 824.54 646.85

In Experiment 1b (conducted after performance asymptotes) participants were given one dimension and asked for a categorization judgment.2 Participants then chose a second dimension (from the remaining eight dimensions) and made a categorization judgment based on only those two dimensions. Only the three relevant dimensions for the categorization task were used in Experiment 1b as the first cue presented to the participant. Both high and low values of these dimensions were given to the participants. Dimension values were selected from the categories such that the values were always less than (or greater than) the best fitting criteria values for that dimension (i.e., only dimensional values from nonoverlapping category regions were presented). The first major result to notice from this experiment is the overall percent correct participants achieved, which is shown in Table 2. The optimal percent correct possible with only two categories is 51.6%. Participant 3 was very close to optimal, while Participants 2 and 4 actually performed better than would be expected. In addition, Participant 4 actually performed better in Experiment 1b than in Experiment 1a! Table 2: Overall Percent Correct in Experiment 1B

1 Percent Correct

42.67

Participant 2 3 55.23

49.83

4 64.5

The results from Experiment 1b indicate that participants did indeed learn which of the cues in Experiment 1a were relevant. All four participants chose (nearly always, if not always) one of the three relevant dimensions as their second cue in Experiment 1b (see Table 3). This indicates that participants were not using any of the other dimensions during Experiment 1a 3 . 2

Participants were given the first cue to insure that all three of the relevant dimensions would be chosen. If participants were allowed to choose the first cue to use, it is possible that the same cue would be used first for each trial. 3 This does not rule out the possibility that participants were using other dimensions in Experiment 1a, but preferred to use one of the three relevant dimensions when limited in the number of dimensions available to them. However, verbal

Table 3: Dimension Preference for Participants 1-4 Dimension Presented 1 2 3

1 2 3

1 2 3

1 2 3

Dimension Chosen by Participant 1 1 2 3 4-9 23 150 25 0 188 9 2 0 186 11 0 0 Dimension Chosen by Participant 2 1 2 3 4-9 9 80 103 1 86 3 100 5 91 88 7 7 Dimension Chosen by Participant 3 1 2 3 4-9 16 162 22 0 162 5 27 0 186 9 4 0 Dimension Chosen by Participant 4 1 2 3 4-9 15 45 134 0 113 0 87 0 133 59 8 0

According to CBE when dimension 1 is presented, dimension 3 should be chosen and when dimension 2 or 3 is presented, dimension 1 should be chosen. When dimension 1 was presented first two of the participants preferred the dimension with the highest probability of success (dimension 3). When dimension 2 was presented first, three of the participants preferred the dimension with the highest probability of success (dimension1). All four participants preferred the dimension with the highest probability of success (dimension 1) when dimension 3 was presented first. Overall, the participants generally chose the second dimension in accord with predictions made by CBE.

Learning Relevant Cues Given the difficulty of the task in Experiment 1a, it is remarkable that the participants were able to learn the relevant cues. As shown above, all four participants chose (nearly always, if not always) the three relevant dimensions as their second cue in Experiment 1b. But how did cue use progress as the participants learned the different categories in Experiment 1a? To answer this question three different versions of MDS were fit to the participants’ category confusion matrices from each half of each day in order to determine how many cues were used by each participant for a particular data set. MDS1 uses only one dimension, MDS2 uses two dimensions, and MDS3 uses three dimensions to protocol collected at the end of the experiment indicated that participants were only using three dimensions during Experiment 1a.

account for the participants’ confusions . A χ2 analysis was performed on the differences between the fit values for models differing in one dimension. These results are reported in Table 4. For participant 1, an MDS choice model using two dimensions did fit the responses better than an MDS choice model using only one dimension for day 2. By day 4, an MDS choice model using three dimensions did obtain a significantly higher fit value than an MDS choice model using only two dimensions. These results indicate that participant 1 used only one dimension on day 1, two dimensions on days 2 and 3, and three dimensions on days 4 and 5.4 Similarly, the MDS analysis indicates that participants 2 and 3 used only one dimension on the first half of day 1, two dimensions on the second half of day 1, and three dimensions after day 1. Participant 4 appeared to use only one dimension on the first half of day 1, two dimensions on days 2 and 3, and three dimensions on days 4 through 6. Taken with the results from Experiment 1b, it appears that participants not only increased over days the number of cues used when categorizing, but also learned the correct (or relevant) cues to use to accurately categorize. Given a task consisting of many dimensions, it is clear that participants begin by using only one dimension. Additional dimensions are then learned in a sequential fashion. What is remarkable from these data, is that participants learned to use all three dimensions. Dimension 1 had more variance than any of the other eight dimensions while dimension 2 had less variance than any of the other eight dimensions. Therefore, it is not surprising that participants were able to learn these two dimensions (i.e., the two dimensions out of nine that had differing variances). Dimension 3 on the other hand, had the same amount of variance as the six irrelevant dimensions, yet participants learned by the end of the experiment that this dimension was necessary for accurate categorization.

Conclusion In conclusion, the studies reported here show that humans are able to learn artificial multidimensional categories. It was also shown that people are able to distinguish relevant from irrelevant dimensions in multidimensional categorization tasks. Results from such a task indicate that a satisficing model is best able to account for the participants’ responses. In addition, the predictions made by the satisficing model regarding cue preference were shown to be in accord with the cue 4

Note, that on the last half of day 5, the increase in parameters used by and MDS choice model with three dimensions did not fit the data significantly better than an MDS choice model with less parameters (i.e., less dimensions).

Table 7: Χ2 diff Values for Participants 1 Participant 1 Day/Half

Participant 2

Participant 3

Participant 4

1/1

MDS1 MDS2 8.34

MDS2 MDS3 0.08

MDS1 MDS2 3.26

MDS2 MDS3 0.3

MDS1 MDS2 8.62

MDS2 MDS3 3.6

MDS1 MDS2 1.02

MDS2 MDS3 2.84

1/2

6.56

6.76

27.18*

12.3

102.9*

18.84*

35.7*

5.08

2/1

83.3*

13.8

71.28*

18.96*

92.78*

9.94

86.16*

0.64

2/2

140.44*

2.56

69.94*

6.54

136.76*

30.16*

117.28*

3.62

3/1

214.98*

9.42

78.76*

22.04*

183.38*

29.14*

109.98*

0.38

3/2

174*

11.14

98.18*

35.86*

140.16*

21.1*

80.2*

4.8

4/1

244.36*

28.54*

116.86*

37.6*

155.02*

35.3*

74.56*

11.92*

4/2

146.22*

22.7*

149.28*

30.82*

196.44*

33.72*

80.36*

22.78*

5/1

151.78*

23.48*

116.8*

38.18*

113.6*

41.34*

80.48*

30.94*

5/2

201.98*

14.5

147.96*

34.34*

193.02*

39.92*

143.76*

18*

6/1

--

--

--

--

--

--

132.96*

37.92*

6/2

--

--

--

--

--

--

155.54*

33.08*

preferences of the participants. Finally, the new experimental design proposed provides a method for further testing the properties of dimensions (cues) that humans prefer (or are constrained?) to use.

References Aeberhard, S., Coomans, D., & de Vel, O. (1994). Comparative analysis of statistical pattern recognition methods in high dimensional settings. Pattern Recognition, 27(8), 1065-1077. Alfonso-Reese, L. A. (1995). General Recognition Theory toolbox for MATLAB. [Macintosh computer software], Santa Barbara, CA. Ashby, F. G., & Gott, R. (1988). Decision rules in the perception and categorization of multidimensional stimuli. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 33-53. Ashby, F. G., & Maddox, W. T. (1993). Comparing decision bound and exemplar models of categorization. Perception & Psychophysics, 53(1), 49-70. Ashby, F. G., & Maddox, W. T. (1990). Integrating information from separable psychological dimensions. Journal of Experimental Psychology: Human Perception and Performance, 16(3), 598-612. Ashby, F. G., & Perrin, N. A. (1988). Toward a unified theory of similarity and recognition. Psychological Review, 95(1), 124-150. Berretty, P. M., Todd, P. M., & Blythe, P. W. (1997). Categorization by elimination: A fast and frugal approach to categorization. In M. G. Shafto & P. Langley (Eds.), Proceedings of the Nineteenth Annual Conference of the Cognitive Science Society. Mahwah, NJ: Erlbaum.

Berretty, P.M, Todd, P.M., and Martignon, L. (1999). Using few cues to choose: Fast and Frugal Categorization. In G. Gigerenzer & P.M Todd (Eds.), Simple heuristics that make us smart. Oxford University Press. Brainard, D. H. (1997). The Psychophysics Toolbox, Spatial Vision, 10, 443-446. Maddox, W. T., & Ashby, F. G. (1998). Selective attention and the formation of linear decision boundaries: Comment on McKinley and Nosofsky (1996). Journal of Experimental Psychology: Human Perception and Performance, 24(1), 301-321. Martin, G. L., & Pittman, J. A. (1991). Recognizing handprinted letters and digits using backpropagation learning. Neural Computation, 3(2), 258-267. Nosofsky, R. M. (1986). Attention, similarity, and the identification-categorization relationship. Journal of Experimental Psychology: General, 115(1), 39-57. Nosofsky, R. M., Palmeri, T. J., & McKinley, S. C. (1994). Attention, similarity, and the identificationcategorization relationship. Journal of Experimental Psychology: General, 115(1), 39-57. Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: Transforming numbers into movies, Spatial Vision, 10, 437-442. Posner, M. I.; Keele, S. W. (1968) On the genesis of abstract ideas. Journal of Experimental Psychology, 77(3), 353-363. Quinlan, J. R. (1993). C4.5: Programs for machine learning. Los Altos: Morgan Kaufmann. Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79, 281-299.

A Perceptually Driven Dynamical Model of Rhythmic Limb Movement and Bimanual Coordination Geoffrey P. Bingham ([email protected]) Department of Psychology and Cognitive Science Program, 1101 E. 10th St. Indiana University Bloomington, IN 47405-7007 USA

Abstract We review the properties of coordinated rhythmic bimanual movements and previous models of those movements. Those models capture the phenomena but they fail to show how the behaviors arise from known components of the perception/ action system and in particular, they do not explicitly represent the known perceptual coupling of the limb movements. We review our own studies on the perception of relative phase and use the results to motivate a new perceptually driven model of bimanual coordination. The new model and its behaviors are described. The model captures both the phenomena of bimanual coordination found in motor studies and the pattern of judgments of mean relative phase and of phase variability found in perception studies.

Introduction In coordination of rhythmic bimanual movements, relative phase is the relative position of two oscillating limbs within an oscillatory cycle. For people without special skills (e.g. jazz drumming), only two relative phases can be stably produced in free voluntary movement at preferred frequency (Kelso, 1995). They are at 0° and 180°. Other relative phases can be produced on average when people follow metronomes, but the movements exhibit large amounts of phase variability (Tuller & Kelso, 1989). They are unstable. Preferred frequency is near 1 Hz. As frequency is increased beyond preferred frequency, the phase variability increases strongly for movement at 180° relative phase, but not at 0° (Kelso, 1990). If people are given an instruction not to correct if switching occurs, then movement at 180° will switch to movement at 0° when frequency reaches about 3-4 Hz (Kelso, 1984; Kelso, Scholz & Schöner, 1986; Kelso, Schöner, Scholz & Haken,1987). With the switch, the level of phase variability drops. There is no tendency to switch from 0° to 180° under any changes of frequency. These phenomena have been captured by a dynamical model formulated by Haken, Kelso and Bunz (1985). The HKB model is a first order dynamic written in terms of the relative phase, φ, as the state variable. The equation of motion, which describes the temporal rate of change in φ, that is, φ-dot, is derived from a potential function, V(φ), which captures the two stable relative phases as attractors as show in Figure 1. The attractors are wells or local minima in the potential layout. As the dynamic evolves, relative phase is attracted to the bottom of the wells at 0° and 180°. A noise term in the model causes the

The HKB model: V(φ) = –a cos(φ) – b cos(2φ) 0

+180

0

+180

0

+180

increasing frequency Figure 1. The HKB model. The parameters a and b are varied to model changes in the potential as a function of increases in frequency of movement. relative phase to depart stocastically from the bottom of a well. The effect of an increase in frequency is represented by changes in the potential. The well at 180° becomes progressively more shallow so that the stochastic variations in relative phase produce increasingly large departures in relative phase away from 180°. These departures eventually take the relative phase into the well around 0° at which point, the relative phase moves rapidly to 0° with small variation.

Investigating Phase Perception We wondered: what is the ultimate origin of the potential function in this model? Why are 0° and 180° the only stable modes and why is 180° less stable than 0° at higher frequencies? To answer these questions, we investigated the perception of relative phase because the bimanual movements are coupled perceptually, not mechanically (Kelso, 1984; 1995). The coupling is haptic when the two limbs are those of a single person. Schmidt, Carello and Turvey (1990) found the same behaviors in a visual coupling of limb movements performed by two different people. Similar results were obtained by Wimmers, Beek, and van Wieringen (1992). To perform these tasks, people must be able to perceive relative phase, if for no other reason, than to comply with the instruction to oscillate at 0° or 180° relative phase. For reasons discussed at length by Bingham, Zaal, Shull, and Collins (2001), we investigated the visual perception of mean relative phase and of phase variability using both actual human movements (Bingham, Schmidt & Zaal, 1998) and simulations (Bingham, et al., 2001; Zaal, Bingham & Schmidt, 2000) to generate displays of two oscillating balls viewed side on or in depth. Observers judged mean phase or phase variability on a 10 point scale. We found that judgments of phase variability (or of the stability of

Judgment Means

5

Frequency = 1.25 hz

4 3 2 Frequency = .75 hz

1 0

coordination. The asymmetric inverted-U pattern of the judgments is essentially the same as the potential function of the HKB model. The potential represents the relative stability of coordination or the relative effort of maintaining a given relative phase. The two functions match not only in the inverted-U shape centered around 90° relative phase, but also in the asymmetry between 0° and 180°. 180° is less stable than 0°. This congruence of the movement and perception results supports the hypothesis that the relative stability of bimanual coordination is a function of the stability of phase perception. So, we developed a new model of bimanual coordination in which the role of phase perception is explicit.

Modelling the single oscillator

0

30 60 90 120 150 180 Mean Relative Phase (deg)

Figure 2. Judgments of phase variability. Mean judgments of phase variability for movements with 0° phase SD (Standard Deviation) and at 7 mean phases from 0° to 180° relative phase. Filled circles: Movement at a freequency of .75 hz. Filled squares: Movement at 1.25 hz. movement) followed an asymmetric inverted-U function of mean relative phase, even with no phase variability in the movement as shown in Figure 2. Movement at 0° relative phase was judged to be most stable. At 180°, movement was judged to be less stable. At intervening relative phases, movement was judged to be relatively unstable and maximally so at 90°. Levels of phase variability (0°, 5°, 10°, and 15° phase SD) were not discriminated at relative phases other than 0° and 180° because those movements were already judged to be highly variable even with no phase variability. The standard deviations of judgments followed this same asymmetric inverted-U pattern. We found that judgments of mean relative phase varied linearly with actual mean relative phase. However, as phase variability increased, 0° mean phase was increasingly confused with 30° mean phase and likewise, 180° was increasingly confused with 150°. Also, the standard deviations of judgments of mean relative phase followed the same asymmetric invertedU function found for the means and standard deviations of judgments of phase variability. Finally, we investigated whether phase perception would vary in a way consistent with the finding in bimanual coordination studies of mode switching from 180° to 0° relative phase when the frequency was sufficiently increased. In addition to mode switching, increases in the frequency of movement yielded increases in phase variability at 180° relative phase but not at 0° relative phase. As shown in Figure 2, Bingham, et al. (in press) found that as frequency increased (even a small amount), movements at all mean relative phases other than 0° were judged to be more variable. This was true in particular at 180° relative phase. Frequency had no effect on judged levels of phase variability at 0° mean phase. Results from our phase perception studies are all consistent with the findings of the studies on bimanual

The HKB model is a first order dynamical model in which relative phase is the state variable. That is, the model describes relative phase behavior directly without reference to the behavior of the individual oscillators. The model was derived from a model formulated by Kay, Kelso, Saltzman and Schöner (1987) that does describe the oscillation of the limbs explicitly. In this latter model, the state variables are the positions and velocities of the two oscillators. To develop this model, Kay, et al. (1987) first modelled the rhythmic behavior of a single limb. In this and a subsequent study (Kay, Saltzman & Kelso, 1991), they showed that human rhythmic limb movments exhibit limit cycle stability, phase resetting, an inverse frequencyamplitude relation, a direct frequency-peak velocity relation, and, in response to perturbation, a rapid return to the limit cycle in a time that was independent of frequency. A dimensionality analysis showed that a second-order dynamic with small amplitude noise is an appropriate model. The presence of a limit cycle meant the model should be nonlinear and a capability for phase resetting entailed an autonomous dynamic. (Note: Phase resetting means that the phase of the oscillator was different after a perturbation than it would have been if not perturbed. An externally driven or non-autonomous oscillator will not phase reset because the external driver enforces its phase which is unaffected by perturbation of the oscillator.) Kay, et al. (1987) captured these properties in a 'hybrid' model that consisted of a linear damped mass-spring with two nonlinear damping (or escapment) terms, one taken from the van der Pol oscillator and the other taken from the Rayleigh oscillator (hence the 'hybrid') yielding: x¨ + b ˙x + α ˙x3 + γ x2 ˙x + k x = 0

(1 )

This model was important because it captured the principle dynamical properties exhibited by human rhythmical movements. However, the relation between terms of the model and known components of the human movement system was unclear. The damped mass-spring was suggestive of Feldman's λ-model of limb movement (also known as the equilibrium point or mass-spring model). The λ-model represents a functional combination of known muscle properties and reflexes. Nevertheless, in the hybrid model, the functional realization of the nonlinear damping terms was unknown. Following a strategy described by Bingham (1988),

Bingham (1995) developed an alternative model to the hybrid model. All of the components of the new model explicitly represented functional components of the perception/action system. The model also incorporated the λ-model, that is, a linear damped mass-spring. However, in this case, the mass-spring was driven by a perceptual term. Limb movements are known to exhibit organizations that are both energetically optimal and stable (e.g. Diedrich & Warren, 1995; Margaria, 1976; McMahon, 1984). Both energy optimality and stability are achieved by driving a damped mass-spring at resonance, that is, with the driver leading the oscillator by 90°. Accordingly, Hatsopoulos and Warren (1996) suggested that this strategy might be used in driving the Feldman mass-spring organization to produce rhythmic limb movements. However, a driver that is explicitly a function of time would yield a nonautonomous dynamic, that is, a dynamic that would not exhibit phase resetting. Bingham (1995) solved this problem by replacing time in the driver by the perceived phase of the oscillator. That is, instead of Fsin(t), the driver is Fsin(φ), where φ is the phase. Because φ (= f[x, dx/dt]) is a (nonlinear) function of the state variables, that is, the position and velocity of the oscillator, the resulting dynamic is autonomous. The perceptually driven model is: x¨ + b ˙x + k x = c sin[ φ ]

limit cycle after a brief perturbing pulse. As also shown, the model exhibits the inverse frequency-amplitude and direct frequency-peak velocity relations as frequency was increased from 1 hz to 6 hz. Finally, the model exhibits a pattern of phase resetting that is similar to that exhibited by the hybrid oscillator. Goldfield, Kay and Warren (1993) found that human infants were able to drive a damped mass-spring at resonance. The system consisted of the infant itself suspended from the spring of a "jolly bouncer" which the infant drove by kicking. This instantiates the model and shows that even infants can use perceived phase to drive such an oscillator at resonance. We hypothesize that all rhythmic limb movements are organized in this way. Once again, the components are the Feldman mass-spring (composed of muscle and reflex properties) and a driver that is a function of the perceived phase of the oscillator.

Modeling Coupled Oscillators With this model of a single oscillating limb, we were ready to model the coupled system. Kay, et al. (1987) had modeled the coupled system by combining two hybrid oscillators via a nonlinear coupling: x¨ 1 + b ˙x1 + α ˙x31 + γ x21 ˙x1 + k x1 =

(2 )

( x˙ 1 − ˙x2 )[ a + b ( x 1 − x2 ) 2 ]

where

φ

 x˙ n  = arctan , ˙xn = ˙x/√ k and c = c ( k ) .  x 

The amplitude of the driver is a function of the stiffness. Bingham (1995) showed that this oscillator yields a limit cycle. This is also shown in Figure 3 by rapid return to the 20 15

Velocity

10

5 0 -5

-10

( x˙ 2 − ˙x1 )[ a + b ( x 2 − x1 ) 2 ]

(3 )

This model required that people simultaneously perceive the instantaneous velocity difference between the oscillators as well as the instantaneous position differences so that both could be used in the coupling function. This model did yield the two stable modes (namely, 0° and 180° relative phase) at frequencies near 1 hz, and mode switching from 180° to 0° relative phase at frequencies between 3 hz and 4 hz. We propose an alternative model in which two phase driven oscillators are coupled by driving each oscillator using the perceived phase of the other oscillator multiplied by the sign of the product of the two drivers (Ρ). This sign simply indicates at each instant whether the two oscillators are moving in the same direction (sign = +1) or in opposite directions (sign = –1). The model is: x¨ 1 + b ˙x1 + k x1 = c sin( φ 2 ) Ρ ij

-15 -20

-2

x¨ 2 + b ˙x2 + α ˙x32 + γ x22 x˙ 2 + k x2 =

-1

0 Position

1

2

Figure 3. Phase portrait of the single perceptually driven oscillator. Movement starts at 1 hz and increases gradually to 6 hz. Early in the movement while still at 1 hz, the movement was perturbed by a 50ms pulse. Rapid return to the limit cycle within about 1 cycle is shown. Also shown is the decrease in amplitude and the increase in peak velocity that accompanies the increase in frequency.

x¨ 2 + b ˙x2 + k x2 = c sin( φ 1 ) Ρji

(4 )

where Ρ = sgn( sin ( φ 1 ) sin( φ 2 ) + α ( x˙ i − ˙xj ) Ν t )

(5 )

P represents the perceived relative phase. As shown in equation (5), the product of the two drivers is incremented by a gaussian noise term with a time constant of 50 ms and a

400

different tasks, one is a coordinated movement task and the other is a judgment task. Equation (5) represents the way the perception of relative phase plays a role in the coordinated movement task. This is in terms of the momentary value of Ρ, that is, whether the oscillators are perceived to be moving in the same or in opposite directions at a given moment in time. Equations (6) and (7) represent the way the perception of relative phase plays a role in the judgment tasks. In this case, the behavior of Ρ is assessed (that is, integrated) over some window of time that is large enough to span one or two cycles of movement. So, the two tasks are connected by a single perceptible property, but the way the property is evaluated and used is task-specific.

200

0 -200 -400 0

5

10

15

20

Time

Figure 4. Continuous relative phase from a run of the perceptually coupled model starting at 1 hz and 180° relative phase. Frequency was increased progressively to over 4 hz. Relative phaase became progressively more variable and switched to 360° = 0° at 4 hz. (Note: Frequecy = sqrt(Time+1).) results for judgments of mean relative phase and of phase variability. (See e.g. Figure 5.) Judged mean phase is produced by integrating Ρ over a moving window of width σ (= 2 s) to yield PJM:

P JM

∫ =

t t– σ

Ρ dt (6 )

σ

Judged phase variability is predicted by integrating (Ρ − PJM)2 over the same window to yield PJV:

P JV =

∫

t t– σ

2

[Ρ − PJM] dt σ

(7 )

PJM varies linearly with actual mean phase and PJV yields an asymmetric inverted-U as a function of actual mean phase. There are two aspects of the perceptual portions of the model that should be emphasized. First, there are actually two perceptible properties entailed in the model. The two are very closely related, but they are distinct. The first is the phase of a single oscillator. The perception thereof is entailed in the single oscillator model. This is, of course, incorporated into the coupled oscillator model. The second perceptible property is relative phase. This latter property brings us to the second aspect of the model to be noted. This is especially important. This model is being used to model performance in two

Judged Phase Variability

Relative Phase (degs)

variance that is proportional to the velocity difference between the oscillators. This noise term reflects known sensitivities to the directions of optical velocities (De Bruyn & Orban, 1988; Snowden & Braddick, 1991) and is motivated by results from phase perception experiments (Collins & Bingham, 2000). This model also yields only two stable modes (at 0° and 180° relative phase) at frequencies near 1 hz, and, as shown in Figure 4, yields mode switching from 180° to 0° relative phase at frequencies between 3 hz and 4 hz. Furthermore, the model predicts our

100 80 60 40 20

3 hz

2 hz

1 hz

0 -20 0 20 40 60 80 100 120 140 160 180 200

Measured Relative Phase Figure 5. Model predictions of judgments of phase variability at a number of different mean relative phases and at three different frequencies of movement. The model was forced to relative phases other than 0° and 180° to obtain these results. Conclusions The model captures both the movement and the perception results. It exhibits the fundamental properties of human rhythmic movements. It builds on the previous taskdynamic modeling results of Kay et al. (1987) and Kay et al. (1991) which revealed fundamental dynamic properties of human movement. Those properties are captured by the new model as they were by previous models. However, unlike the previous models, the new model’s components are interpretable in terms of known components of the perception/action system. It explicitly represents the perceptual coupling that is well recognized to be fundamental to the coordination task and the resulting bimanual behaviors. This is important because we can now proceed to investigate the perception component (no less important than the properties of muscle in the Feldman component) to discover the origin of some of the dynamic properties of these perception/action systems. This is an explicit perception/action model. Finally, although its behaviors are extremely complex,

the model itself is relatively simple and elegant. Two relatively simple equations (4) capture limit cycle stability, phase resetting, inverse frequency-amplitude and direct frequency-peak velocity relationships, the stable modes and mode transitions and the increasing patterns of instability leading up to mode transition. With the addition of two more simple equations (6) and (7) computing a mean and a variance, the model accounts for the results for perceptual judgments of mean relative phase and of phase variability and the ways these vary with the frequency of movement. All this from a model with 5 parameters (k, b, c, α, and σ), four of which are fixed and one, k, is varied to generate variations in frequency of movement. (Note: because c=f(k), c varies with k but once the scaling of c is fixed, this does not represent an extra degree of freedom.) The model is representative of nonlinear dynamics: complex behavior emergent from simple dynamic organization.

Acknowledgments This research was supported in part by NEI grant # EY11741-01A1 and by NIMH grant # 5T32MH19879-07. The author is grateful for assistance provided by David R. Collins in performing simulations and some of the phase perception studies that have constrained the model. The studies reported herein were reviewed and approved by the Human Subjects Committee at Indiana University. All participants gave their informed consent prior to participation in the experiments.

References Bingham, G.P. (1988). Task specific devices and the perceptual bottleneck. Human Movement Science, 7, 225264. Bingham, G.P. (1995). The role of perception in timing: Feedback control in motor programming and task dynamics. In E.Covey, H. Hawkins, T. McMullen & R. Port (Eds.) Neural representation of temporal patterns. New York: Plenum Press. Bingham, G. P., Schmidt, R. C., Zaal, F. T. J. M. (1998). Visual perception of relative phasing of human limb movements. Perception & Psychophysics, 61, 246-258. Bingham, G.P., Zaal, F.T.J.M., Shull, J.A., Collins, D.R. (2001). The effect of frequency on the perception of the relative phase and phase variability of two oscillating objects. Experimental Brain Research, 136, 543-552. Collins, D.R. & Bingham, G.P. (2000). How continuous is the perception of relative phase? InterJournal: Complex Systems, MS # 381. De Bruyn, B. & Orban, G.A. (1988). Human velocity and direction discrimination measured with random dot patterns. Vision Research, 28, 1323-1335. Diedrich, F.J. & Warren, W.H. (1995). Why change gaits? Dynamics of the walk-run transition. Journal of Experimental Psychology: Human Perception and Performance, 21, 183-202. Goldfield, G., Kay, B.A. & Warren, W.H. (1993). Infant bouncing: The assembly and tuning of an action system. Child Develoment, 64, 1128-1142. Haken, H., Kelso, J. A. S., & Bunz, H. (1985). A theoretical model of phase transitions in human hand movements. Biological Cybernetics, 51, 347-356.

Hatsopoulos, N.G. & Warren, W.H. (1996). Resonance tuning in arm swinging. Journal of Motor Behavior, 28, 3-14. Kay, B.A., Kelso, J.A.S., Saltzman, E.L. & Schöner, G. (1987). Space-time behavior of single and bimanual rhythmical movments: Data and limit cycle model. Journal of Experimental Psychology: Human Perception and Performance, 13, 178-192. Kay, B.A., Saltzman, E.L. & Kelso, J.A.S. (1991). Steady-state and perturbed rhythmical movements: A dynamical analysis. Journal of Experimental Psychology: Human Perception and Performance, 17, 183-197. Kelso, J. A. S. (1984). Phase transitions and critical behavior in human bimanual coordination. American Journal of Physiology: Regulation, Integration, and Comparative Physiolology, 15, R1000-R1004. Kelso, J. A. S. (1990). Phase transitions: Foundations of behavior. In H. Haken and M. Stadler (Eds.), Synergetics of cognition. Springer Verlag, Berlin. Kelso, J. A. S. (1995). Dynamic patterns: The selforganization of brain and behavior. MIT Press, Cambridge, MA. Kelso, J. A. S., Scholz, J. P., Schöner, G. (1986). Nonequilibrium phase transitions in coordinated biological motion: Critical fluctuations. Physics Letters A, 118, 279-284. Kelso, J. A. S., Schöner, G., Scholz, J. P., Haken, H. (1987). Phase-locked modes, phase transitions and component oscillators in biological motion. Physica Scripta, 35, 79-87. Margaria, R. (1988). Biomechanics and energetics of muscular exercise. Oxford: Clarendon Press. McMahon, T.A. (1984). Muscles, reflexes, and locomotion. Princeton, N.J.: Princeton University Press. Schmidt, R. C., Carello, C., Turvey, M. T. (1990). Phase transitions and critical fluctuations in the visual coordination of rhythmic movements between people. Journal of Experimental Psychology: Human Perception and Performance, 16, 227-247. Snowden, R.J. & Braddick, O.J. (1991). The temporal integration and resolution of velocity signals. Vision Research, 31, 907-914. Tuller, B., Kelso, J. A. S. (1989). Environmentally specified patterens of movement coordination in normal and split-brain subjects. Experimental Brain Research, 75, 306-316. Wimmers, R. H., Beek, P. J., van Wieringen, P. C. W. (1992). Phase transitions in rhythmic tracking movements: A case of unilateral coupling. Human Movement Science 11, 217-226. Zaal, F.T.J.M., Bingham, G.P., Schmidt, R.C. (2000). Visual perception of mean relative phase and phase variability. Journal of Experimental Psychology: Human Perception and Performance, 26, 1209-1220.

InferencesAboutPersonalIdentity SergeyBlok(

[email protected]) [email protected])

GeorgeNewman(

Behr( [email protected])

Jennifer

LanceJ.Rips(

[email protected])

DepartmentofPsychology,NorthwesternUniversity, Evanston,Illinois60208USA Abstract

FeaturesofPersonIdentity We investigate the features people use in making in ferences about continuity of individual persons. Using a tra nsformationparadigm,weshowthatpeopleweighbothconti nuityof the brain and continuity of mental content. Across experiments,wedocumentinstancesinwhichparticipants aremore likely to assert individual continuity than continu ity of personhood. We discuss these results in terms ofa hie rarchical viewofconceptsandphilosophicalworkonpersonal identity.

Introduction Peoplearesensitivetotheeffectsthattransforma tionshave on membership in basic-level categories (e.g., Gelm an & Wellman,1991;Keil,1989;Rips,1989).Forexample ,Rips (1989) asked participants to read stories about cre atures of one category (e.g., birds) that came to resemble th ose of another(e.g.,insects). Ifthetransformationwas duetoaccidental factors, participants believed that the cr eature remained a member of the original category. If the transformation was part of normal maturation, howev er, participants judged the creature a member of the se cond category. In general, transformations that alter a n object’s core properties also tend to change the object’s ca tegory, while transformations that alter an object’s surfac e or incidentalpropertiestendtopreserveitscategory . Despitetherelativelylargenumberofstudiesthat address questions of category membership continuity, there have beenfewstudiesaddressingreasoningaboutindivid ualcontinuity (see Hall, 1998, Johnson, 1990 and Liittsch wager, 1994, for exceptions in the developmental literatur e). The central question is howwe decide thata particular individual – for example, my dog Fido – is the same indivi dual (still Fido) across transformations. This question is potentially different from the one about membership – wh ether thisindividualisstilladog. We investigate here two issues concerning reasoning about individuals. First, we explore the kinds of features people use in judging continuity of identity. Seco nd, we contrastthe waysinwhichpeoplereasonaboutclas smembershipandaboutidentitycontinuity.

Whatpropertiesdoesapersonattime t1 needtosharewith one attime t2 inorder for thatindividualto be thesame at both temporal markers? In making such judgments, pe ople may be phenomenalists, relying on continuity of appearance. Inapreliminaryexperiment,wecreatedstoriestha tvaried the type of transformation that a hypothetical targ et person undergoes. One set of participants -- the Plastic S urgery group--readascenarioaboutJim,amaleaccounta nt,who receivesplasticsurgerytoalterhisappearanceco smetically toresemblethatofMarsha,afemaleactress.Anot hersetof participants -- the Brain Transplant group -- read a similar story in which Jim’s brain is replaced with that of Marsha. After reading the story, both groups supplied judgm ents of Jim’s identitychange –whethertheindividualwasstillJim orhadbecomeMarshaaftersurgery.Resultsindicat edthata greater proportion of participants in the Brain Tra nsplant groupbelievedJim’sidentityhadchangedthanint hePlas2 (1, 39) = tic Surgery group (45% and 15%, respectively, 4.29, p‘ key, once the stimulus had gone from the screen. If no response was required during training, then there was a two second inter-stimulus interval. Training consisted of 30 presentations of A and 30 of B in a random order. All participants were told that the chequerboards that they were to be shown could be divided into two groups, and that this was their task. After the final training trial, the instructions concerning the test phase were displayed. The test phase was identical for all participants regardless of their training condition and consisted of 130 trials which displayed a chequerboard on its own. The instructions asked each participant to continue placing the stimuli into two groups in the same way as before, but were informed that there would now be no label, if there had been one before, and that a response was required for each chequerboard. In conditions where a label had been presented to the participants a key mapping was provided, for example, “Press ‘x’ if it’s an A”. If no label had been present during training, participants were simply asked to carry on placing the chequerboards into the most relevant group. The instructions for the Mental Decision group required them to make an arbitrary assignment of the groups they had formed during training to the keys to be used during the test phase. Once a response to a test stimulus had been made, it was immediately replaced by another stimulus. If any other key apart from the two that were designated was pressed, the computer beeped, and another response was required. Participants were asked to focus more on accuracy rather than speed in this part of the experiment, and there was no explicit time-out procedure, so participants cannot be considered to be under any time pressure. The design of the test phase was such that participants were shown stimuli along the continuum from the master pattern (A) to the second prototype (B). Each test stimulus had a multiple of 10 squares changed from A, to make it more like B. As A and B differed by 120 squares, there are 13 steps along such a continuum and with each point being sampled 10 times, this gives 130 test trials. The 10 stimuli for any given point on the continuum were all different, and generated as described above. At the end of the experiment the computer automatically recorded responses and reaction times from the training phase where possible, and from the test phase in a data file. Each participant was paid for their time and thanked for their participation.

Results The two measures of performance recorded during the generalisation test phase were response and reaction time, and they are dealt with separately. The independent measure on the plots in this section is distance along the

Table 1: Factors in Experiment 1 Information

No Information

Keypress

Match to Label

Free Classification

No Keypress

Label

Mental Decision

B-A continuum, with B at one end and A the other, and the intermediate values being examples of B with (10 × distance) squares changed to make the stimulus more like A. Data was analysed firstly by comparison of the groups using ANOVA. A factorial design was also used, as illustrated below (Table 1). The conditions included in the experiment have tried to control for possible effects of making a response interacting with the presence of accurate information about category membership. Responses A single mixed design analysis of variance (ANOVA), with one within-subject variable (continuum position, 13 levels) and one between subjects variable (training condition, 5 levels) was performed on the mean number of B responses at each point on the continuum for each participant. This failed to reveal any significant difference between the groups F(4,55) = 2.25, p0.9. There was a significant main effect of position along the B-A continuum F(12,660) = 6.88, p 0.15. The mean responses for each level at each point along the continuum are plotted in Figure 2. Reaction Times Participants were not considered to be under any time pressure, so the reaction times provide a secondary performance measure. The mean reaction times for each group are plotted against the distance from B along the BA continuum in Figure 3. ANOVAs were performed on the mean reaction times for each participant. A single mixed design ANOVA, with one within subject variable (continuum position, 13 levels) and one between subjects variable (training, 5 levels) revealed a significant main effect of group, F(4,55) = 5.00, p 0.1. A mixed design ANOVA, with one within subject variable (continuum position, 13 levels) and two between subject variables (the presence or absence of consistent information about category membership and the require-

4

Label Feedback Match

Free Mental

3.5

Feedback Mental Match

3 2.5 2 1.5 1

B

2

4 6 8 10 Continuum Position

Figure 1: Response Data for All Groups in Experiment 1

A

B 1 2 3 4 5 6 7 8 9 10 11 A Continuum Position Figure 3: Reaction Times for All Groups in Experiment 1

Label No Label 3 2.5 2 1.5 1 B

2

4 6 8 10 A Continuum Position Figure 4: Effect of Label in Experiment 1 ment to make a keypress) was performed. This revealed a significant effect of information, F(1,44) = 13.67, p = 0.001, the expected main effect of continuum position, F(12,528) = 4.83, p < 0.001 and an interaction between these two factors F(12,528) = 2.77, p < 0.05. No other effects approached significance in this analysis, p > 0.25. The mean reaction times for each level at each point along the continuum are plotted in Figure 4.

Discussion Experiment 1 provides evidence for a number of novel results with respect to the effect of feedback on categorisation. Whilst there is no significant observable effect of training condition on response performance, there is a significant effect on the reaction time data collected. Initially surprising is the fact that groups trained with a completely errorless teaching signal (Labelled and Match to Label) recorded the slowest reaction times. Two other groups received no feedback (Free Classification and Mental Decision), and both are found to be significantly faster than the Labelled condition. The response curves show that this speed advantage cannot be due to speedaccuracy trade off, and so there must be some other explanation for the deficit in performance observed in groups which intuitively should be the best at the task. The main effect of labelling in the factorial analysis for the reaction time data confirms this finding, with the presence of a category label causing an increase in response time on generalisation. The effect of making a keypress can be seen in the response analysis, but can be regarded as relatively uninteresting as there is no sign of an interaction which might point to the task having been learnt better, causing one group to have a more step-like function. The most obvious reason for the speed advantage seen for those groups who do not have a consistent piece of information relating to category membership is that they are required to make a decision about category membership during the training phase. This seemingly essential part of training is not present when a label is provided, as the stimulus and its category name are presented simultaneously. In the Feedback condition, even though there is consistent category information, a decision must be made before this feedback can be received and integrated. In the

conditions where no feedback it present, a decision made internally about previous stimuli is the only information available when deciding to which group subsequent stimuli should be assigned. The similarity of the generalisation functions implies that all groups have learnt how to differentiate between the two categories to the same extent. However it may be that the application of this knowledge is mediated by a response mechanism which is yet to be set up by those groups which are not required to make a decision as to category membership during training. This leads to the difference seen between the relatively flat reaction time curves for the three conditions where decisions are required in training and the inverted U-shaped reaction time curves produced by those participants who are presented with the label during training. The inverted U-shaped curves are typical of those produced during generalisation along a continuum using these procedures (Jones, Wills & McLaren, 1998). The centre point of the continuum is no more like an A than a B, so any response made to these stimuli must be indeterminate, and hence produces a longer latency than those responses to items which are more like those in training. Despite the potentially simple explanations for the differences observed between the groups, it is still an interesting result to have the Free Classification and Feedback conditions indistinguishable from one another even with the supposed added advantage of corrective feedback. This may be due to some motivational factor, as participants in the Free Classification group are never told that they are wrong, however with feedback, participants may be relatively sure that the stimulus that they are seeing is an A, but in fact turns out to be a B, which may disrupt their representation of the conditions for category membership. The feedback that they get may be consistent within the framework of the experiment, but may be inconsistent internally, and this may be part of the reason for the lack of benefit for the Feedback condition. The reason the Label condition is so slow may be because it takes time to learn, and then to use, the response mapping. It cannot be entirely due to motor learning as although there is some advantage for the Matching to Label group over the Label group, there is still a difference between the Matching condition and the three where a decision was made during training. The results from the Mental Decision group also tend to discount this line of reasoning as they were not required to form a key mapping before the test phase but their responses are indistinguishable from the Feedback and Free Classification.

Experiment 2 Experiment 2 was designed in a factorial fashion to investigate the effect of both feedback and consistent labelTable 2: Factorial Design of Experiment 2 Feedback

No Feedback

Label

Match to Label (FB)

No Label

Corrective Feedback

Match to Label (NFB) Free Classification

ling of the stimuli. In this respect it was similar to Experiment 1, but attempted to control for the formation of response mappings during training. Experiment 2 forced all participants to adopt a key mapping during training, so that any effect on test would be due to the actual training rather than differences in procedure. All participants were forced to make their responses to the stimuli within two seconds of the stimulus disappearing using the same keys as before.

0.9 0.7 0.6 0.5 0.4

These were identical to those in Experiment 1.

0.3

Participants and Design

0.2

Forty-eight Cambridge University students took part in the experiment. All were aged between 18 and 25. The experiment was designed to test the effect of two factors when learning an artificial categorisation problem. These were the presence and absence of feedback on the responses that were made during training, and the presence or absence of a consistent category label during training The design was similar to Experiment 1, with the test phase being identical, and is shown in Table 2.

0.1

The procedure was similar to Experiment 1. The only differences were during training. All groups were presented with a stimulus and asked to respond within two seconds to that stimulus once it had disappeared. The training differed from Experiment 1 by presenting the label, if necessary, before the stimulus rather than concurrently. For those conditions without a label, a ‘#’ symbol was presented before each stimulus, whether it was nominally an A or a B, in place of the label to equate the training time. The appropriate label or ‘#’ was displayed for five seconds before the five second presentation of the stimulus. The instructions were identical to those from Experiment 1 apart from detailing the separate presentation of the label or ‘#’ and the stimulus and informing the participants of their two second time limit.

Results As with Experiment 1, there were two dependent variables, response selection and reaction time, and as a time limit had been imposed, the reaction times can be considered a more informative performance indicator. The data were analysed using an appropriate ANOVA. The performance of individual groups was analysed along with the effect of the factors built into the experiment. Responses A single mixed design analysis of variance (ANOVA), with one within-subject variable (continuum position, 13 levels) and one between subjects variable (training condition, 4 levels) was performed on the mean number of A responses at each point on the continuum for each participant. This failed to reveal any significant difference between the groups F(3,44) = 0.76, p>0.5 or any interaction between the groups and the position along the continuum

Match (NFB) Feedback

0.8

Stimuli and Apparatus

Procedure

Match Free

1

0 A 1

2

3

4 5 6 7 8 9 10 11 B Continuum Position

Figure 5: Response Data for All Groups in Experiment 2 F(36,528) = 0.65, p>0.5. There was a significant main effect of position along the A-B continuum F(12,528) = 55.73, p 0.15. Reaction Times The mean reaction times for each group are plotted against the distance from A along the A-B continuum in Figure 6. ANOVAs were performed on the mean reaction times for each participant. A single mixed design ANOVA, with one within subject variable (continuum position, 13 levels) and one between subjects variable (training, 4 levels) revealed a significant main effect of group, F(3,44) = 3.38, p = 0.026, and of position along the continuum, F(12,528) = 5.64, p < 0.001. There was no significant interaction between the two factors, F(36,528) = .90, p > 0.6. A Tukey HSD test performed on the group factor revealed that the Match to Label condition with feedback was found to be significantly slower than the Free Classification condition, p < 0.05. A mixed design ANOVA, with one within subject variable (continuum position, 13 levels) and two between subject variables (the presence or absence of category membership information and the presence of feedback) was performed. This revealed a significant effect of category information, F(1,44) = 4.54, p = 0.039, with the label conditions showing the longer mean

1.35

systems (Rumelhart & Zipser, 1986; Saksida, 1999) which are able to extract the necessary information from the stimuli encountered to form coherent categories through exposure to the stimuli alone. It may be the case that all that is needed is exposure to stimuli in order to extract information about them, and this raises the interesting question of what exactly feedback does if it does not always aid decision making.

1.25

References

1.15

Bersted, C.T., Brown, B.R. & Evans, S.H. (1969). Free sorting with stimuli in a multidimensional attribute space. Perception and Psychophysics, 6B, 409-413. Estes, W.K. (1994). Classification and Cognition. Oxford: Oxford University Press. Evans, S.H. & Arnoult, M.D. (1967). Schematic concept formation: Demonstration in a free sorting task. Psychonomic Science, 9(4), 221-222. Homa, D., & Cultice, J. (1984). Role of Feedback, Category Size, and Stimulus Distortion on the Acquisition and Utilization of Ill-Defined Categories. Journal of Experimental Psychology: Learning, Memory and Cognition, 10, 83-94. Jones, F.W., Wills, A.J. & McLaren, I.P.L. (1998). Perceptual Categorisation: Connectionist modelling and decision rules. Quarterly Journal of Experimental Psychology, 51(B), 33-58. Rosch, E., & Mervis, C.B. (1975). Family resemblance: Studies in the internal structure of categories. Cognitive Psychology, 7, 573-605. Rumelhart, D.E., & Zipser, D. (1986). Feature discovery by competitive learning. In D.E. Rumelhart, J.L. McClelland, & The PDP research group. Parallel Distributed Processing: Explorations in the microstructure of cognition. Cambridge, MA: MIT Press. Saksida, L.M.(1999). Effects of similarity and experience on discrimination learning: A nonassociative connectionist model of perceptual learning. Journal of Experimental Psychology: Animal Behavior Processes, 25, 308-323. Wills, A.J. & McLaren, I.P.L. (1998). Perceptual Learning and Free Classification. Quarterly Journal of Experimental Psychology, 51(B), 235-270.

1.65

Match Free

Match (NFB) Feedback

1.55 1.45

1.05 0.95 0.85 0.75 A 1 2 3 4 5 6 7 8 9 10 11 B Continuum Position Figure 6: Reaction Time Data for All Groups in Experiment 2 reaction times, and the expected main effect of continuum position, F(12,528) = 5.64, p < 0.001. No other effects were significant in this analysis, p > 0.09.

General Discussion The results from Experiment 2 are broadly in line with those obtained in Experiment 1. The inclusion of a consistent category label appears to have a detrimental effect when compared with the requirement to make an active decision about category membership. However, there appears to be a speed-accuracy trade off present in the results. Whilst the Match to Label with feedback (MFB) group are slowest, they also show the greatest difference between the ends of the generalisation gradients. Although this difference is not significant it would be difficult to conclude anything definite on the basis of this alone. However, taken with the results from Experiment 1, it seems clear that the fact that feedback is not present does not seem to have a detrimental effect on the performance shown by participants. Instead it seems that, in some cases, providing an entirely consistent label for the stimuli during training causes participants to perform worse. This is not what would be predicted from Homa & Cultice (1984), and is at odds with the results from Estes (1994) who showed that a condition analogous to the Label condition in Experiment 1 gave better performance on test than when participants were trained using a corrective feedback approach. It may be that the real advantage lies in being able to make active (i.e. self-generated) decisions during training rather than simply being exposed to the stimuli and the appropriate category information (Figure 4). Thus it may be the case that different processes are at work

Conclusion It seems likely that the most successful approach to modelling such data will come form simple self-organising

An Analogue of The Phillips Effect Mark Suret ([email protected]) Department of Experimental Psychology; Downing Street Cambridge, CB2 3EB. UK I.P.L. McLaren ([email protected]) Department of Experimental Psychology; Downing Street Cambridge, CB2 3EB. UK

Abstract Previous experimental work has demonstrated that human participants can easily detect a small change in a visual stimulus if no mask intervenes between the original stimulus and the changed version and the inter-stimulus interval is small (Phillips, 1974). Rensink, O’Regan & Clark (1997) have shown that if a mask is used then detecting the change is extremely difficult, no matter how small the ISI is made. This work attempts to establish whether familiarity with a stimulus has any effect on a participants ability to detect a small change in it using Rensink’s masking procedure with Phillips’ stimuli (checkerboards). Participants were required to make judgements as to whether two stimuli, which alternated with one another in presentation, were the same or different. Some participants attempted the task using just one checkerboard pattern which became increasingly familiar across sessions, others were given new, randomly generated checkerboards for each trial. In both conditions, any change (which would occur on 50% of trials) would only affect one square of the pattern. The results show a clear advantage for the participants dealing with familiar stimuli in detecting any change, and go some way towards explaining why this is so.

Introduction Phillips (1974) demonstrated how easy it was for participants to detect a change between two stimuli if they were presented one after the other without a gap in a single alternation. This is the Phillips Effect. He also investigated the consequences of inserting a grey mask and a blank screen between the two stimuli. The inclusion of an interstimulus interval adversely affected participants' performance, and the presence of a mask made performance even worse. Rensink et al. (1997) demonstrated that the brief inclusion of a grey mask between repeated presentations of two slightly different stimuli made any change extremely difficult to detect. This is the Rensink Effect. Their experiment used electronically altered images which allowed manipulation of the colour, position and presence of an object. Without the mask, spotting the difference becomes trivial, if the stimuli are positioned in the same place and then alternated. Current explanations of this phenomenon cite retinal transients (Klein, Kingstone & Pontefract, 1992) as the mechanism for detecting changes in this latter case, which would be unaffected by familiarity.

Some pilot work using a single participant indicated that the effect of familiarity with the stimuli was likely to be very significant. Hence introducing the notion of familiarity removes some of the difficulties intrinsic to the Rensink Effect and makes the task more similar in difficulty to the Phillips Effect. One drawback of this pilot experiment was that it contained familiar and random trials in each session so after a while the participant became able to tell which were the familiar trials, and this may have differentially affected the responses to each trial. Nevertheless, the results of the pilot experiment (Figure 1) allowed the prediction that the Familiar condition would lead to better detection of changes than the Random condition. It was also predicted that there would be an improvement in performance as the amount of time spent on the task increased. The main aim of this work was to take the Rensink Effect, and attempt to ameliorate it, by allowing participants to practice on the same stimulus all the time. This would require the use of a different type of stimulus, as the repeated use of a real life scene would be impossible to control properly, so checkerboards were used as the training stimuli. In this way, the participants were presented with essentially the same stimulus on every trial, but with the possibility of a change in one of the elements within the 100

90

Familiar Random

80

70

60

50 Session 1

Session 2

Session 3

Session 4

Figure 1: The Basic Familiarity Effect

Session 5

The experiment was run in a quiet room on an Apple Macintosh LCIII computer using a colour monitor. Participants responded to the stimuli by pressing one of two keys, either [x] or [.] on a QWERTY keyboard. Between blocks, participants were required to fill in a sheet to record their errors and reaction times before pressing the space bar to continue to the next block. The responses for each block were logged in separate data files. Figure 2: An example pair of chequerboard stimuli. checkerboard. An example pair is shown in Figure 2) with the difference being rather difficult to spot. Once a reliable difference had been found between the Familiar and Random groups, subsequent testing concentrated on the mechanism being used by participants to detect the changes.

Experiment The experiment was conducted with two groups, each with different sets of stimuli. The Random group was a control group, and received different, randomly generated checkerboards for each trial during the experiment. In the Familiar group, each participant was trained on a single checkerboard unique to that participant. The primary aim of the experiment was to determine whether familiarity with the stimulus affected the participants’ performance. Initially two participants were run in each condition followed by a preliminary analysis. It was found that a large difference between the groups had already been established, and the final four participants were all run in the familiar condition. This allowed manipulations in the familiar condition, namely test blocks which departed from the standard task. The test blocks used manipulations intended to disrupt the performance of the participants in the Familiar group in the hope that this would suggest a possible mechanism for the way the changes were being detected. Clearly an effect of familiarity or session needed to be established first, as there is no directly relevant and properly controlled research in this area.

Stimuli and Apparatus All the stimuli were randomly generated, two centimetre square checkerboards, with sixteen elements on a side giving a total of 256. Each base pattern stimulus had equal numbers of black and white squares, before any change was introduced. An example pair for the change condition is shown above (Figure 2), with the difference between the two checkerboards in the top right centre of the stimulus. Checkerboards were chosen as they are easy to manipulate for this type of experiment. Many different individual changes could be made whilst keeping the majority of the stimulus the same. In addition, the participants were unlikely to be familiar with the stimuli prior to the experiment Those participants assigned to the random condition were given a newly generated checkerboard on each trial, whereas those in the familiar condition were always presented with the same pattern, albeit with a change on half the trials.

Participants and Design In total eight Cambridge undergraduates took part in the study. Four were allocated to the initial phase to determine the possible existence of a familiar/novel distinction. Two participants were allocated to the novel condition and two to the familiar condition. The remaining four participants were all allocated to the familiar condition. The experiment consisted of a training phase for all participants and a test phase for those participants in the familiar group. The first four sessions were used for training for both groups, with the fifth session containing some test blocks for participants in the Familiar condition. All sessions for the random group were identical, as the tests given to the familiar group would have made no difference to a participant receiving a new checkerboard on every trial. Each one hour session consisted of ten blocks of stimuli, each containing 24 trials giving 240 trials in a session. Each block contained twelve trials where a change was present and twelve where there was no change between the two checkerboards. These trials were presented in a random order. Participants were asked to try to detect a change between the two checkerboards, and respond appropriately as to whether or not they thought that a difference was present. For each trial, two checkerboards were alternated with one another, separated by a randomly generated mask. These checkerboards were either the same, or differed by one element within the pattern, i.e. one element that was black in one checkerboard was white in the other. The trials were such that each checkerboard was displayed for 500 milliseconds and the mask for 100 milliseconds. Over one trial, each checkerboard could be presented ten times, giving nineteen changes in a trial if the checkerboards were different. After the final alternation, the trial ended and if no decision had been made by the participant, then they were timed out. During the fifth session, the participants in the familiar group were given test blocks in between familiar blocks, in an attempt to determine how they might be detecting the changes. These test blocks were ones containing random trials, such as those given to the participants assigned to the random group, and another type of block, labeled “C”. In these blocks, there was always a fixed, random one square difference from the original base pattern, on both checkerboards, whether the trial was one of change or no change. This fixed change was different on each trial within the block. On change trials, there was also an additional change made to one of the checkerboards. This manipulation ensured that some difference from the base pattern was no longer a cue for change, although there was still a single change present between the two checkerboards

on change trials. The idea behind this manipulation was to contrast any changes in performance on “C” trials with that obtained on Random trials. In the former case, the perturbation of the familiar pattern is minimal, in the latter case it is, in some sense maximal. The sequence of blocks was: Familiar, “C”, Familiar, Random, Familiar, “C”, Familiar, Random, Familiar, “C”. This gave three Familiar (as the first block is removed from any analysis), three “C” and two Random blocks to be used in the analysis for each participant from a session of ten blocks. The Familiar blocks were inserted between the test blocks to allow the participants an opportunity to re-establish baseline performance before the next test block was administered.

Procedure The participants were seated in front of the computer approximately 50 centimetres from the screen and asked to make themselves comfortable. They were then read the instructions concerning their task, and were then asked if they had any questions about the instructions they had just been given. The participants were asked to respond to a “change” trial with their left index finger, by pressing the [x] key, and to a “no change” trial by pressing the [.] key with their right index finger. The participant was then asked to press the space bar to begin, and follow the on screen instructions that occurred throughout the experiment. The experimenter waited in the room until the first few trials had been completed, to ensure that the participant fully understood what it was that they were meant to be doing before leaving the room. Each block was started by pressing the space bar. The trials consisted of the alternation of two checkerboards, with a random black and white dot pattern mask being presented between presentations of the checkerboards. These checkerboards were either the same or differed by one element. The checkerboards subtended a visual angle of approximately two degrees and were presented in the centre of the screen. The participants were given feedback on each trial, with the words “correct” or “error” being displayed on the screen. If an error was made, the computer also beeped. After each block of twenty-four trials, participants were required to record their errors and reaction times on a sheet provided for them in the room. This was primarily to get the participants to take a break between blocks. It also gave a readily available source of data that could be tallied with the

analysis on the computer. At the end of the session of ten blocks, the participants were given a short questionnaire to determine how motivated they were feeling during the session and what, if any strategy they were using. After the questionnaire had been completed the participants were thanked for their time and the next session was arranged. After the fifth session, a more thorough questionnaire was administered, and the participants were paid and thanked for their participation.

Results The basic familiarity effect is shown below in Figure 3. The Familiar and Random groups are denoted by F and R respectively. The graph shows that both groups improved at the task at roughly the same rate, but that performance in the Familiar group is better than that in the random group on all sessions by a roughly constant amount. The initial analysis focused on finding significant effects of both group and session. Three dependent variables have been determined for each session: overall accuracy; percentage of changes detected; percentage of correct no change trials. Each of the three variables used may indicate something different about the way that the participants may be performing their task. 100

Overall (F) Overall (R)

90 80

70 60

50 1

2

3

4

5

Session Figure 3: Between Groups Familiarity Effect.

Table 1: Within Session Comparisons between Random and Familiar Groups (All probabilities are one-tailed) Overall Percentage Correct Session 1 Session 2 Session 3 Session 4 Session 5

U=0, p=0.022 U=0, p=0.022 U=0, p=0.022 U=0, p=0.022 U=0.5, p=0.033

Percentage of No Change Trials Correct U=0, p=0.023 U=0, p=0.020 U=0, p=0.022 U=0, p=0.020 U=0, p=0.017

Percentage of Changes Detected U=0.5, p=0.032 U=1, p=0.046 U=0, p=0.022 U=0, p=0.023 U=1, p=0.048

Table 2: Between Session Comparisons Collapsed across Groups (All probabilities are one-tailed) Change between Sessions

Overall Percentage Correct

1 and 2 2 and 3 3 and 4 4 and 5

T=1, p=0.009 T=0, p=0.009 T=2.5, p=0.024 T=6, p=0.173

Only non-parametric tests were used to analyse the data, as there were both small and unequal numbers in the groups. A significance level (α) of p1, and >1, respectively.

Retention Test Data from the retention test were pooled over the 2 test sessio ns and were subjected to a repeated-measures ANOVA, with task component (one-to-one vs. one-tomany) and Delay (0, 2, 4, and 8 sec) as factors. Most critically, the ANOVA indicted that there was a significant Task Component x Delay interaction, F(3,21) = 4.37. There was also a significant effect of Delay, F(3,21) 44.01. The effect of Task Component was not quite significant, F(1,7) = 4.79. The retention data are presented in Figure 1.

Discussion According to traditional instrumental views of conditional discrimination learning (i.e., Hartl & Fantino, 1996), the probability of a comparison choice should be determined by the conditional probability associated with each comparison stimulus, given the sample, and, if the sample is unavailable or forgotten, with the probability of reinforcement associated with each comparison (independently of the sample). Thus, the choice a particular comparison (e.g., C1) should depend on both the number of sample-comparison pairings (e.g., S1-C1) that are followed by reinforcement, as well as the number of reinforcements associated with that comparison, independent of the sample (Wixted, 1993). In the present experiment, the conditional probability of reinforcement associated with each of the comparisons, Figure 1. Retention functions following training in which two and samples, S1 and S2, were associated with comparison stimuli, C1 C2, respectively and S1 and S3 were associated with comparisons C3 and C4, respectively. Thus, S2 and S3 were involved in one-to-one matching (OTO) with C2 and C4, while the third sample, S1, was associated with two comparison stimuli, C1 and C3 (one-to-many matching, OTM). In training and test, each comparison was associated with reinforcement on 50% of the trials and C1 and C2 always appeared together as did C3 and C4. given one of the samples, was equal. Furthermore, the probability of reinforcement associated with choice of either comparison was also equal. Thus, in the present experiment, given presentation of C1 and C2, the only relevant samplecomparison associations determining comparison choice should be S1-C1 and S2-C2. If so, delay-induced sample degradation should have had a symmetrical effect on comparison choice and the retention functions should have been parallel and overlapping.

In the present experiment, clearly divergent retention functions were found. These results require the modification of current theories of delayed conditional discrimination performance (e.g., White & Wixted, 1999) because pigeons choice behavior is influenced not only by the probability of reinforcement associated with responding to each of the comparison stimuli and to the conditional probabilities associated with choice of the comparison stimuli as a function of memory for the sample but also by the relative frequencies of the samples. When delays are introduced, as the delay increases, pigeons have an increasing tendency to select the comparison associated with the more frequently presented sample, even though th at sample was not presented more often than the alternative sample in the context of either comparison pair. It is as if, on trials when memory for the sample is poor, presentation of the comparisons causes the pigeons to consult their reference memory for the overall probability of sample presentatio n (independent of the comparison pair). Of broader interest, such use of reference memory in delayed matching may be a general phenomenon. However, the u s e of sample frequency independently of other more relevant measures may be apparent only with a design such as that used in the present research because in the more typical design, either hypothesis makes the same prediction. Alternatively, in the present experiment, although the pigeons had equal opportunity to acquire each of the four sample-comparison associations, the more frequent presentations of the S1 sample could have allowed it to be more efficiently coded, better maintained in memory, or more easily retrieved from memory. That is, at the time of comparison choice, when the S1 stimulus had been the sample, it may have been more accessible than the S2 or S3 stimuli were when they had been the sample. But if the difference in slope of the retention functions was attributable to differences in sample accessibility at the time the comparisons were presented, both the S1 and the S2/S3 functions should have approached 50% correct with increasing retention interval. Instead, the S1 retention function appea rs to have leveled off, while the S2/S3 retention function declines below chance at delays of 4 and 8 sec. Such retention functions suggest that rather than better retrieval of the S1 sample, the pigeons developed a comparison bias to choose the comparison associated with the more frequently presented sample. This comparison bias in pigeons is analogous to the base-

rate neglect shown by humans when they fail to consider sufficiently the base-rate probability of occurrence of an event.

References Gigerenzer, G. & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Freque n c y formats. Psychological Review, 102, 684-704. Goodie, A. S., & Fantino, E. (1995). An experimentally derived base-rate error in humans. Psychological Science, 6, 101-106. Goodie, A. S., & Fantino, E. (1996). Learning to comit or avoid the base-rate error. Nature, 380, 247-249. Grant, D. S. (1991). Symmetrical and asymmetrical coding of food and no-food samples in delayed matching in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 17, 186-193. Hartl, J. A., & Fantino, E. (1996). Choice as a function of reinforcement ratios in delayed matching to sample. Journal of the Experimental Analysis of Behavior, 66, 11-27. Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgement of representativeness. Cognitive Psychology, 3, 430-453. Skinner B. F. (1950). Are theories of learning necessary? Psychological Review, 57, 193-216. Tversky , A, & Kahneman, D. (1980). Causal schemas in jud gements under uncertainty. In M. Fishbein (Ed.), Progress in social psychology (pp. 49-72). Hillsdale, NJ: Erlbaum. White, K. G., & Wixted, J. T. (1999). Psychophysics of remembering. Journal of the Experimental Analysis of Behavior, 71, 91-113. Wixted, J. T. (1993). A signal detection analysis of memory for nonoccurrence in pig e o n s . Journal of Experimental Psychology: Animal Behavior Processes, 19, 400-411.

Author Notes This research was supported by National Institute of Mental Health Grants 55118 and 59194. Correspondence should be addressed to Thomas R. Zentall, Department of Psychology, University of Kent ucky, Lexington, KY 40506-0044. Email, [email protected]

Member Abstracts

Explanations of words and natural contexts: An experiment with children’s limericks Greg Aist ([email protected]) Project LISTEN, Carnegie Mellon University, 5000 Forbes Avenue Pittsburgh, PA 15213 USA

Introduction

Results and discussion

Project LISTEN’s Reading Tutor listens to children read aloud, and helps them learn to read. Here, we used the Reading Tutor to study whether adding childfriendly definitions to natural text would help children learn new words. We compared four conditions:

To explore the effects of explanations and limericks, we used logistic regression – modeling a binary outcome variable using several categorical factors as input. If a factor’s coefficient was significantly greater than zero, than that factor affected the outcome variable. Word familiarity. Seeing an explanation helped, at p < 0.001: coefficient 1.08 ± .32; 99.9% confidence interval (CI) .02, 2.15. Seeing the word in a limerick showed only a weak trend, at .50 ± .32; 90% CI -.02, 1.03. Word knowledge. The results for word knowledge were more nuanced: a (not significant) trend for explanations (.24 ± .31), but none for limericks (-.05 ± .31). However, younger students were about at chance:

1. No encounter 3. In a story alone

2. In a definition alone 4. In a story and a definition

This study took place at a July 2000 reading and math clinic at a low-income urban elementary school in Pittsburgh. Each student was scheduled to spend 30 minutes per weekday on the Reading Tutor.

Experiment design We used eight children’s limericks by Edward Lear (19th c.), with one target word each: dolorous, laconic, imprudent, innocuous, mendacious, oracular, irascible, or vexatious. The target word was always the second word in the last line. Our text selection controlled for genre, author, intended audience, (approximate) word frequency, part of speech, and general semantic class: There was an Old Man of Cape Horn, Who wished he had never been born; So he sat on a chair, Till he died of despair, That dolorous Man of Cape Horn.

We wrote target word definitions in a consistent style using ordinary language, following McKeown (1993). For example: “We can say someone is dolorous if they are mournful, or feel really bad.” To reduce variance from first- or last-item effects, we held constant the order of presentation of the limericks. Each student saw two target words per condition. Word-to-condition assignment was set for each Reading Tutor computer. One or two days later – depending on attendance – we gave each student a paper questionnaire with two items per word. “Have you ever seen the word dolorous before?” tested familiarity, and “If someone is dolorous they must be… angry; sad; tired; afraid.” tested word knowledge. To exclude memorization, the definitions and the test answers used different words. In all, 29 students who had just finished 2nd - 5th grades completed the experiment, for a total of 232 trials, 58 trials for each of 4 conditions.

2nd grade, 19/72 right (26%) 4th grade, 16/56 right (29%)

3rd grade, 18/72 right (25%) 5th grade, 10/32 right (31%)

For 4th and 5th graders, however, in a main-effects-only model, explanations helped (p < .10): .89 ± .52, with 90% CI .04, 1.74. There was not an effect for limericks, at -.13 ± .51. (Disaggregation by grade is exploratory.) Conclusions. Thus in terms of learning word meaning, only explanations seemed to help – and only for fourth and fifth graders. These effects are neither same-day recency nor simple memorization. Aist (Aist 2000 ch. 6) discusses further.

Acknowledgments This paper is based on work supported in part by the National Science Foundation under Grant Nos. REC-9720348 and REC-9979894, and by the first author’s Harvey and NSF Graduate Fellowships. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the National Science Foundation or the official policies, either expressed or implied, of the sponsors or of the United States Government. Dr. Jack Mostow directs Project LISTEN; http://www.cs.cmu.edu/~listen lists other team members.

References Aist, G. 2000. Helping Children Learn Vocabulary during Computer-Assisted Oral Reading. Ph.D. dissertation, Language Technologies Institute, CMU. McKeown, M. G. 1993. Creating effective definitions for young word learners. Reading Research Quarterly 28(1), 17-31.

Understanding death as the cessation of intentional action: A cross-cultural developmental study H. Clark Barrett ([email protected]) Center for Adaptive Behavior and Cognition Max Planck Institute for Human Development Lentzeallee 94, Berlin, Germany

Introduction Taken together, the developmental literatures on children’s understanding of intentional action and on children’s understanding of death present a sort of paradox: while the former literature shows that even very young children have an intuitive grasp of the goal-directed nature of behavior that characterizes animate, living things, the latter seems to show that children do not realize until a much later age that the capacity to act intentionally ceases irreversibly when an organism dies. From an adaptationist perspective, this is perhaps surprising, given the potential adaptive value of being able to judge the capacity or incapacity of an animal to act, and perhaps to do harm. In many states which are characterized merely by a lack of motion – sleep, for example – an animal may still wake up and attack. When dead, however, it cannot. It may thus be adaptive for young children to understand that death, as opposed to sleep and other states of temporary inaction, entails the permanent cessation of the capacity for intentional action. And, given that an intuitive grasp of the capacity for intentional action in animate living things exists at a very young age, it should be available to serve as a conceptual substrate for an early understanding of death.

Study design The present study was designed to probe for an understanding of death as the cessation of the capacity for intentional action by comparing judgments of 3 to 5 year old children about hypothetical dead and sleeping animals (human and nonhuman). Children were first asked questions about an awake animal, then about the same animal either asleep or dead. The target questions required children to judge whether the sleeping or dead animal would be capable of movement in general or in response to being touched, awareness of an external stimulus (someone moving nearby), and / or of having an emotion. The intentional action theory of death understanding predicts that children will exhibit the clearest understanding of death in response to questions involving movement and movement in response to stimulus, as opposed to questions that do not involve intentional action (e.g., awareness only). There was no predicted lower bound on the age of emergence of reasoning abilities probed here.

The study was conducted in two parts, both using the same interview procedures and questions. The first part of the study was conducted with 70 preschool and kindergarten children, age 3 to 5 years, in Berlin, Germany. The second part of the study was conducted with 70 Shuar children, age 3 to 5 years, in six small rural villages in the Amazon region of Ecuador. Because these populations vary widely both in exposure to various kinds of cultural and environmental input (e.g., television and films, direct contact with animals, personal experience with death), they were selected for comparison to determine whether the development of the kind of understanding of death probed here depends significantly on cultural inputs such as television, and / or aspects of personal experience such as contact with animals or firsthand experience with death. Parents in both populations were surveyed about relevant experiences of their children which might have influenced understanding of death, such as religious background, exposure to representations of death on television, and personal experience with death of animals or people.

Results Both Shuar and German children demonstrated an understanding of death as the cessation of the capacity for intentional action by the age of 4. By this age, the large majority of children clearly distinguish sleep from death in this regard. In addition, many of these children understand that the capacity for subjective experience independent of action ceases in death as well, though performance on these questions, as predicted, was not as high at this early age as performance on questions involving intentional action. The results of the study show that by age 4, children understand at least one aspect of death which is crucially important from an adaptive perspective: dead things can no longer act. Although this result stands in contrast to much of the developmental literature on death understanding, which suggest that children’s understanding of death at this age is poor, it is consistent with a view of cognitive development which holds that development has been shaped by a history of selection for adaptive reasoning and decision making abilities.

Working Memory Processes During Abductive Reasoning Martin Baumann ([email protected]) Josef F. Krems ([email protected]) Department of Psychology, Wilhelm-Raabe-Str. 43 Chemnitz University of Technology 09107 Chemnitz, Germany

Introduction Abductive reasoning is the process of finding a best explanation for a given set of observations. It is an essential feature of many real world tasks like medical diagnosis, discourse comprehension, and scientific discovery. Such problems often need the processing of an amount of information far beyond the capacity limits of working memory (WM). But on the other hand, working memory is expected to play a central role in human reasoning. On the basis of a computational model of abductive reasoning (Johnson & Krems, 2000) and of theories of text comprehension we propose a mechanism that reduces WM load during abductive reasoning. It suggests that only unexplained symptoms are kept in working memory with explained symptoms are transferred to long-term memory reducing WM load. From this model it follows that unexplained observations should be more available in a recognition or recall task during abductive reasoning than explained ones. We tested this prediction in three experiments each using a different memory task to test the availability of observations.

Experimental Studies The Experimental Task In all experiments a task (BBX) was used where participants had to discover the hidden state of a system through indirect observations. The observations were presented sequentially to the participants. Only the current observation was visible. In each trial, after a variable amount of observations, the participants had to perform a memory task testing the availability of a given observation. The major manipulation in all experiments was whether this observation was already explained at the time of the memory task or not. That is, whether the participant had received the neccessary additional information to explain the observation and actually generated a hypothesis explaining this observation.

Results and Discussion In the first experiment we used a recognition test as memory task to test the availability of the relevant observation. In the second experiment the recognition test was replaced with an implicit memory task. The mental availability of explained and unexplained observations were

tested here by presenting a probe hypothesis that had to be judged with regard to its compatibility with observations presented until then. The results of the first experiment showed that unexplained observations are recognised significantly faster than explained ones, consistent with model predictions. Regarding the recognition accuracy there was no significant effect of interval or explanation status. We also found that maintaining an unexplained observation in WM slows down the recognition and reduces the recognition accuracy for other observations. The second experiment showed contrary to the model’s predictions a tendency of explained observations being forgotten more often with increasing number of intervening observations than unexplained ones. This result suggested that observations are held actively in WM until they are explained. After an explanation was generated they are lost from WM. The result also indicates that explained observations are not integrated in a representation in long-term memory. This interpretation was confirmed in a third experiment showing that participants memory for explained and unexplained observations in an unexpected recall test after the interruption of the reasoning task was equally low.

General Discussion The results confirmed the hypothesis that unexplained observations are actively hold in WM during abductive reasoning until a causal explanation can be generated. Contrary to the predictions of the model these explained observations seem not to become integrated into a representation in long-term memory, but are simply forgotten. But this could be due to the structure of the reasoning task we used, which makes the construction of an integrated representation rather difficult. Therefore in future investigations we need to use a task providing a richer structure, more comparable to real world tasks like medical diagnosis.

References Johnson, T.R., & Krems, J.F. (2000). Use of Current Explanations in Multicausal Abductive Reasoning. Manuscript submitted for publication.

Organizing Features into Attribute Values Dorrit Billman ([email protected]) Carl Blunt ([email protected]) Jeff Lindsay ([email protected]) School of Psychology, Georgia Institute of Technology Atlanta, GA 30332 USA The problem of representation change is important for understanding and developing both natural and artificial intelligence. People form new chunks or perceptual units from experience (Schyns & Rodet, 1997) and build up new, continuous, dimensions, (Goldstone, Lippa, and Shiffrin, 2001). We investigate representation change reorganizing unrelated features into alternative values of the same attribute. For example, ‘fins’, ‘wings’, and ‘legs’ might initially be unrelated properties but people might learn to reorganize them as values of a new attribute, LIMB. Representing properties as attribute values may be importantly different from representing properties as a collection of uncoordinated features, for interpreting novel properties and for projecting inferences. We investigated perceptually-based attribute formation, in the context of learning about cell-like organisms. We use stimuli where an initial analysis into features is highly available, but there are alternative ways of organizing these features into attributes.

For initial exposure, subjects saw 3 blocks of six 1unit displays (one block of each type) and described what they saw. Subjects then viewed 18 displays with one to five units (“slides with lower magnification”) shown together and indicated how many organisms were on each slide, by circling and counting. We varied the order subjects saw the Separate, Attached, and Overlapping blocks (Table 1). We expected the first exposure block to be influential. Does exposure order influence the final interpretation of the elements, as values of a new limb-like attribute or as values of a new buddy-like attribute?

Results & Discussion Table 1 (col. 2) shows that displays had highest counts when subjects first saw the overlapping block, followed by the Attached and then Separated (item analysis F(3,68)=181.9). Interestingly, some subjects analyzed the small internal element as a separate organism, particularly in the Overlap First Conditions. Table 2 (col. 3) shows % of displays where subjects counted units all as 1 or all as 2 organisms; Overlap first subjects were less consistent (F(3,48)=4.8, p