Understanding Attention: A Review of Historical and Contemporary Literature - Prof. 1398, Apuntes de Filología Inglesa

¡Descarga Understanding Attention: A Review of Historical and Contemporary Literature - Prof. 1398 y más Apuntes en PDF de Filología Inglesa solo en Docsity!

Attention

Attention

From Theory to Practice

Edited by

Arthur F. Kramer Douglas Wiegmann Alex Kirlik

1 Oxford University Press, Inc., publishes works that further Oxford University’s objective of excellence in research, scholarship, and education. Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Copyright © 2007 by Arthur F. Kramer, Douglas Wiegmann, and Alex Kirlik Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Attention: from theory to practice / edited by Arthur Kramer, Douglas Wiegmann, and Alex Kirlik. p. cm. – (Series in human-technology interaction; v. 4) Includes bibliographical references and index. ISBN-10: 0-19-530572- ISBN-13: 978-0-19-530572-

1. Attention. 2. Human-machine systems. I. Kramer, Arthur F. II. Wiegmann, Douglas A. III. Kirlik, Alex. IV. Series. BF321.A833 2006 153.7
   33–dc22 2006010098

1 3 5 7 9 8 6 4 2

Printed in the United States of America on acid-free paper

description of early models of visual and auditory attention. He also discusses two real-world examples of modeling and understanding the role of attention beyond the laboratory, both in terms of the design and use of radar displays, and the design of navigation dis- plays and signals for train engineers. Finally, Boot, Kramer, and Becic provide a critical review of the lit- erature on attentional capture and guidance, and dis- cuss how laboratory research does (or does not) scale up to address real-world problems. They conclude by discussing two recent experiments that ask how best to alert operators to important changes on cluttered dynamic radar displays. Part II presents research on a variety of emerging topics in applied attention theory. Parasuraman and Greenwood note that many performance laws in applied psychology do not adequately capture individ- ual differences, and these authors introduce a neu- roergonomics approach to individual performance prediction based on molecular genetics and neuro- science. Their work breaks new ground in the effort to provide a neural and genetic basis for characterizing individual differences in various cognitive functions, including attention and memory. Lee similarly takes the study of attention into new realms by considering the role of affect in information processing. Lee observes that a rapidly growing body of empirical evi- dence now demonstrates that factors such as the emo- tional content of stimuli and responses to technology should no longer be ignored in the application of psy- chology to design. Vicente seeks to broaden our understanding of attention in yet another direction through field investigation. In his study of monitoring a nuclear power plant (NPP), Vicente demonstrates the rich diversity of information sources used to support attention allocation in an operational context, and the clever strategies and devices used by operators to com- pensate for their limited cognitive resources, conclud- ing that “it is simply not possible to monitor an NPP using attentional resources alone.” On a related note, Gray, Neth, and Schoelles argue that understanding attention and performance in interactive systems requires a detailed functional analysis of the external resources available to support performance in addi- tion to internal cognitive resources. Gray and his col- leagues conclude that it may be most fruitful to consider behavior to be adapted to a task environment comprised of a mix of these internal and external resources.

The understanding of driver distraction, which is closely related to failure of selective attention, has become an increasingly important topic given the rapid proliferation of cellular phones, global position- ing satellite navigation systems, in-vehicle entertain- ment and information management systems, and other automotive telematic devices. Indeed, a sub- stantial percentage of automobile accidents are now attributed to driver distraction of inattentiveness. Part III addresses the important issue of driver distraction. The Strayer and Drews chapter is unique in that it brings a number of converging methodologies to bear, including observational field studies, well-controlled simulator studies, and a psychophysiological study, in the examination of the human information processing costs associated with hands-free cell phone use during driving. The conclusions of the study, in terms of the nature and magnitude of performance and safety costs, are both theoretically and practically important, and suggest that recent legistration on cell phones and driving should be reexamined in light of the data. Fisher and Pollatsek examine the important issue of teenage drivers. Why do teen drivers have such a high accident rate and what can be done about it? Both of these questions are addressed by the studies described in the chapter. Changes in attentional processes across the life span have become an increasingly important topic, with both theoretical and practical implications, given the “aging” of most industrialized societies. Tsang provides a criti- cal review of models of attention and, more specifi- cally, of attentional control during the performance of multiple concurrent tasks in the context of aging. She then goes on to discuss the results of a number of stud- ies conducted in her laboratory during which con- verging operations are used to localize age-related costs in multitask processing within the context of Wickens’ multiple-resource model. Finally, Tsang describes the important influence of experience or expertise, in the presented case in the context of pilot- ing, as a moderator of age-related decline of atten- tional control. Like Tsang, Fisk and Rogers begin by providing a review of the literature on skill acquisition and maintenance, focusing on their important research on the development of automatic processing for young and older adults. They then discuss how such data can be used to design products to enhance the independence of older adults. In particular, they describe some very interesting research on aging and

vi PREFACE

independence in the context of the Aware Home at the Georgia Institute of Technology. Part V covers the implications of multiple-resource theory for interface design. In Chapter 13, Sarter pre- sents examples of the successful implementation of multimodal interfaces in support of concurrent task performance and information processing. The chap- ter also describes the additional benefit of distributing information across sensory channels, including redundancy, complementarity, and substitution. In Chapter 14, Theeuwes and colleagues summarize the results of their research on cross-modal interactions between sensory modalities and the implications for the design of multisensory displays. Their findings sup- port the multiple-resource theory’s assumption of inde- pendent resources for auditory and visual processing. However, their research also indicates that cross- modal interference can occur when central process- ing is necessary for information consolidation. Hence, the extent to which multisensory displays will have large advantages over unimodal displays may depend heavily on whether one or more of the modalities compete for limited-capacity central processes. Part VI focuses on attention and training, particu- larly as they apply to multitasking. Chapter 15 by Gopher focuses on “emphasis change,” which is a training protocol that requires individuals to change systematically their emphasis, efforts, and attention allocation policy on a major subcomponent of the performed task. Gopher also introduces a new con- cept of “task shell,” which is a mental model of the integrated structural and dynamic properties of a task. A task shell developed through emphasis change training can lead to greater sensitivity to changes in task difficulty and load, and to better adaptation to changes through attention reallocation. In Chapter 16, Dismukes and Nowinski also focus on a relatively

new but rapidly growing topic in cognitive psychol- ogy called prospective memory. Prospective memory is the process of recalling and performing an action that could not be executed at the time the original intention was formed. Intentions or goals are often deferred as a result of other concurrent tasks in the environment competing for attentional resources. Failures of prospective memory often occur because concurrent tasks win this competition for resources that guide the retrieval of memory items associated with each task. Dismukes and Nowinski provide real- world examples in aviation that illustrate the impact that prospective memory failures can have on flight safety and they provide some recommendations for training individuals to overcome prospective memory failures. The final chapter in this volume, written by Christopher Wickens, provides both a critical review and discussion of the topic of attention over the past 100 years and a prescription for the future. Several cautions and suggestions are offered for the future. First, that researchers become less focused on atten- tional paradigms and more focused on explaining important attentional phenomena. In the past, para- digms have been studied as an end in and of themselves rather than as a means to understanding important real-world attentional phenomena. Second, that the focus on mean effects has, to date, often precluded the study of extreme responses. Given that errors often arise from unusual or extreme events, it is important that they be more frequently examined during the study of human performance in complex systems. Finally, that computational models become more of a focus in the study of applied attention. Although our models will rarely provide the final answer, they serve to formalize our understanding and enable the testing of our hypotheses.

PREFACE vii

Contributors xi

Part I. Toward a Practical View of Attention: Theoretical and

Methodological Issues

1 Attention: From History to Application 3

Neville Moray

2 Attention and Its Allocation: Fragments of a Model 16

Thomas B. Sheridan

3 Capturing Attention in the Laboratory and the Real World 27

Walter R. Boot, Arthur F. Kramer, and Ensar Becic

4 Elaborations of the Multiple-Resource Theory of Attention 45

Peter A. Hancock, Tal Oron-Gilad, and James L. Szalma

Part II. Emerging Issues in Applied Attention Theory

5 Individual Differences in Attention and Working Memory:

A Molecular Genetic Approach 59

Raja Parasuraman and Pamela Greenwood

6 Affect, Attention, and Automation 73

John D. Lee

ix

Contents

7 Monitoring a Nuclear Power Plant 90

Kim J. Vicente

8 The Functional Task Environment 100

Wayne D. Gray, Hansjörg Neth, and Michael J. Schoelles

Part III. Attention and Driving

9 Multitasking in the Automobile 121

David L. Strayer and Frank A. Drews

10 Novice Driver Crashes: Failure to Divide Attention or Failure

to Recognize Risks 134

Donald L. Fisher and Alexander Pollatsek

Part IV. Attention and Aging

11 Attention Goes Home: Support for Aging Adults 157

Arthur D. Fisk and Wendy A. Rogers

12 The Dynamics of Attention and Aging 170

Pamela S. Tsang

Part V. Attention and Interface Design

13 Multiple-Resource Theory as a Basis for Multimodal Interface

Design: Success Stories, Qualifications, and Research Needs 187

Nadine Sarter

14 Cross-Modal Interactions between Sensory Modalities: Implications

for the Design of Multisensory Displays 196

Jan Theeuwes, Erik van der Burg, Christian N. L. Olivers, and

Adelbert Bronkhorst

Part VI. Attention and Training

15 Emphasis Change as a Training Protocol for High-Demand Tasks 209

Daniel Gopher

16 Prospective Memory, Concurrent Task Management, and Pilot Error 225

Key Dismukes and Jessica Nowinski

Part VII. Future Directions

17 Attention to Attention and Its Applications: A Concluding View 239

Christopher D. Wickens

Author Index 251

Subject Index 263

x CONTENTS

JESSICA NOWINSKI, National Aeronautics and Space Administration, Ames Research Center CHRISTIAN N. L. OLIVERS, Cognitive Psychology, Vrije Universiteit, The Netherlands TAL ORON-GILAD, MIT^2 Laboratories, University of Central Florida RAJA PARASURAMAN, Department of Psy- chology, George Mason University ALEXANDER POLLATSEK, Department of Psychology, University of Massachusetts at Amherst WENDY A. ROGERS, School of Psychology, Georgia Institute of Technology NADINE SARTER, Department of Industrial and Systems Engineering, University of Michigan MICHAEL J. SCHOELLES, Department of Cognitive Science, Rensselaer Polytechnic Institute THOMAS B. SHERIDAN, Massachusetts Institute of Technology and Volpe National Transportation Systems Center

DAVID L. STRAYER, Department of Psychology, University of Utah JAMES L. SZALMA, MIT^2 Laboratories, University of Central Florida JAN THEEUWES, Cognitive Psychology, Vrije Universiteit, The Netherlands PAMELA S. TSANG, Department of Psychology, Wright State University ERIK VAN DER BURG, Cognitive Psychology, Vrije Universiteit, The Netherlands KIM J. VICENTE, Department of Mechanical and Industrial Engineering, University of Toronto CHRISTOPHER D. WICKENS, Human Factors Division; and Department of Psychology, University of Illinois at Urbana-Champaign DOUGLAS WIEGMANN, Human Factors Division, University of Illinois at Urbana- Champaign

xii CONTRIBUTORS

Part I

Toward a Practical View of

Attention: Theoretical and

Methodological Issues

INTRODUCTION: A LITTLE HISTORY

Modern work on attention began with the develop- ment of Broadbent’s filter theory (Broadbent, 1953, 1958). Broadbent once pointed out that interest in attention had continued in Europe from the time of Titchener (1903), but hardly a single paper appeared in journals of the American Psychology Association after the malignant influence of Watson and the Ur-behaviorists took control of the discipline in the 1920s. Consider, for example, Stevens’s (1953) Handbook of Experimental Psychology. There are three pages on how animals may attend to only one part of the stimulus during learning, and several pages on the intensity of illumination and attention span—apart from that, nothing. Attention did not figure even in Fitts’s chapter on engineering psychology.^1 The renaissance of attention theory in the 1950s was truly radical.

AUDITORY ATTENTION: THE RENAISSANCE

Broadbent worked almost entirely on auditory attention. Much of his work was applied and, together with the famous “split-span” memory experiment (Broadbent, 1953), led him to formulate his filter theory: the single- channel model of selective attention. However, it did not lead to detailed quantitative predictions. Broadbent’s philosophy of experimental design was to eliminate a large class of alternative explanations by a single experiment, not to predict in detail what would happen in real time in real-world tasks. Almost no work captured the richness and, above all, the temporal dynamics of the real world, although Broadbent always emphasized the importance of applied research. It may be something of a surprise to Chris Wickens to be reminded that in my early days I tried to convince Broadbent that some parallel processing was possible. For example, in split-span experiments we have two

Chapter 1

Attention: From History to Application

Neville Moray

ears but only one mouth, so that output must be serial and sequential. When we gave listeners a stenogra- pher’s keyboard, on which several keys can be pressed simultaneously, we found almost parallel perform- ance. In addition, Broadbent’s original split-span sub- jects only had about 10 trials. We found that after 100 trials their performance on the classic split-span task greatly improved—a result that was confirmed by Geoff Underwood using a listener, one “Moray,” with years of practice at two-channel listening. He found almost perfect ability to listen to two messages at once (Underwood, 1974). We also found that small changes in pitch and loudness could be perceived in parallel (Moray, Fitter, Ostry, Favreau, & Nagy, 1976). Early modern research on attention was con- cerned with underlying mechanisms, rather than pre- dictions about situations unconstrained by laboratory experimental design. Typically, laboratory experi- ments are not dynamic. They consist of a series of sta- tistically independent trials of fixed duration, with the task specified by the experimenter. Participants play no part in determining what will happen next. Real tasks are dynamic, and people decide when attention will be paid to what, how long a “trial” (the word is hardly applicable) will last, and, by their interaction with the environment, alter the future that they will experience. In real tasks, the “experiment” is con- trolled by the participant as much as by the experi- menter (Rasmussen, Pejtersen, & Goodstein, 1995, pp. 219–224).

VISUAL ATTENTION: “THE EYES HAVE IT”

Because of the structure of the retina, with its small area of foveal vision, “real” visual attention is necessar- ily single channel at the level of gaze direction. Of course, one can pay attention to different parts of the retinal array, as Erikson and his group here at Illinois have shown. There are also earlier works, including a rather charming little paper by Babington-Smith (1961) that also shows this, and recently coaches have begun to teach players to pay attention to the periphery of vision in soccer and other sports. However, evolu- tion has provided us with a system of visual attention that is primarily mediated by switching fixation. I have heard it said that attention mediated by eye movements is “uninteresting,” presumably because moving the eyes does not seem to involve “deep” processes inside the head and hence does not seem

“truly cognitive.” This seems to me, a very strange atti- tude. Although movements of the head and eyes may be simple, great cognitive subtleties remain in the choice of what to look at, where to look for it, and when. The most important aspects of visual attention in the working environment are the strategy and tactics of gaze. I offer two examples of successful quantitative modeling on the assumption of single-channel visual attention.^2 In neither study are we concerned with the selection of one message and the rejection of others, but with the dynamic tactical distribution of attention over many sources in real time, all of which require attention. This kind of attention is akin to the “travel- ing salesman’s problem”: how to visit the maximum number of places with minimal travel (Dessouky, Moray, & Kijowski, 1995). The empirical study of eye movements has a surprisingly long history (Woodworth, 1938), but only recently has the technology improved to a level where it is (fairly) easy to collect and analyze data outside the laboratory. On the other hand, there is a plethora of mathematical models for how attention may direct eye movements (Moray, 1986). Here I want to show how one can develop an analytic model of attention and then verify it with empirical measures of eye movements.

Example 1: Visual Attention to Radar Displays: From Analytic Models to Behavior Probably the earliest quantitative model is that of Senders (1964, 1983), inspired by the empirical data of Fitts and his coworkers, who 50 years ago recorded pilots’ eye movements and estimated the probabili- ties of looking at different instruments and the transition probabilities among instruments (Jones, Milton, & Fitts, 1949, 1950). Senders proposed that the purpose of attention was to reduce uncertainty and thus support adaptive behavior in a dynamic envi- ronment. He used information theory’s Nyquist sam- pling theorem to model eye movement dynamics, predicting that observers would fixate quasi-random functions of time at twice the bandwidth of the sources, with the fixation duration dependent on the perceptual accuracy required. He found the predicted linear relations, although high frequencies were undersampled and low frequencies oversampled. He also discussed an early form of urgency model, in which the time until the next fixation depended not just on the bandwidth, but on how close to a constraint

4 ATTENTION: FROM THEORY TO PRACTICE

observation increases exponentially as the distance between the aircraft diminishes. (This is also known to be the case in bat echolocation.) Although the course and speed of the aircraft are known, as time passes and forgetting grows according to Equation 1.1, the area of uncertainty around the memory of each echo will grow, and these will eventually overlap. The region of overlap represents the probability that the aircraft are in collision, and because these areas are represented by a standard deviation, we can associate a probability with the area of overlap. We assume that there is a threshold of intolerability such that if the probability of both aircraft being in the same location exceeds this threshold, the observer will look at the echo about which the uncertainty is greatest. This can be general- ized to three or more aircraft located, and further by weighting the uncertainty thresholds by the relative importance of the aircraft. We further assume that as the distance between aircraft decreases, the threshold of tolerability decreases. Constraints on the maximum rate of changing fixation were added from the known distribution of eye movement times (Boff & Lincoln, 1988), modified by empirical data from a pilot study of eye movements in our operators. Equations repre- senting these dynamics were embodied as a computer program. The model was written in FORTRAN and run on an IBM mainframe. The output of the model is the sequence of fixations and the times at which these occurred. The model makes the following predictions:

1. There will not be more than about two fixations (acts of attention) per second.
2. The intervals between successive looks at an air- craft (the mean first passage time, or MFPT) will be long when aircraft are far apart and will shorten as they approach one another.
3. If there are more than two aircraft on the dis- play, and some of them are in the final stages of a close approach and others are far apart, atten- tion will be absorbed more and more by the for- mer, and in the limit no attention will be paid to distant aircraft as some aircraft become close enough for a potential collision.
4. Observations on features other than the aircraft will decrease (in the sense that the interval between them will increase) as aircraft close on one another.
5. Controllers will suffer from mental overload, and will lose control of the situation, when the number of aircraft being controlled results in MFPTs that are long enough to allow significant forgetting

as predicted by Equation 1.1. We believe this value to occur where the curves begin to accel- erate upward—that is, at about 6 seconds. (This value agrees quite well with the mean time for self-paced sampling found by Moray and colleagues [1973], and with the point in Senders’s [1964] experiment when oversampling begins.)

We were able to compare the predictions of the model with two sets of data: one for which we recorded eye movements from radar operators in a realistic mis- sion simulator and the other while they were control- ling real jet fighters, which played the roles of targets and friendlies. The data obtained both from the computer runs of the model and the records of eye movements from operators consist of long sequences of fixations of the general form

... F 1 -T 1 -T 2 -S-F 1 -t-F 1 -F 2...

where fixations F 1 -F 2 , are friendly aircraft nos. 1 and 2, T 1 -T 2 , are targets 1 and 2, S is a stranger, t is the tote board, and so on, together with the times at which each fixation would occur. Earlier work showed that it was sufficient to sample the eye movement records twice a second. (See Figure 1.1.) The most appropriate way to analyze this kind of data is to use Markov analysis. The sequence of obser- vations is cast into a table containing the frequencies of transitions between the different classes of obser- vations (states), and from these tables a table of state transition probabilities is derived. By appropriate mathematical transformations, several interesting statistics can be derived (Kemeny & Snell, 1960). If the transition probability matrix is raised to succes- sively higher powers, we obtain a table in which all the rows are identical. This is the limiting matrix, and the entry in a column is an estimate of the propor- tion of time spent in that state. The second is the table of MFPTs. If we look at any cell xrc at the intersection of row r and column c, its value is the mean number of samples that will occur from the last time the observer was in state r before state c is entered. For example, if we look at the intersection of F 1 and T 2 , the value in that cell estimates the mean number of states passed through (other things looked at) from the last time F 1 was fixated until the next time T 2 is fixated. If samples are taken every t seconds, then multiplying the MFPT by t coverts the MFPT to a measure of time. A table can also be obtained of variance of the MFPTs, from which the standard deviation of the MFPT can

6 ATTENTION: FROM THEORY TO PRACTICE

be derived. If data include self-transitions where for cells xrc r  c, we can also estimate the mean duration of fixation time for each variable. These are very powerful ways of looking at tempo- ral sequences. They capture the dynamics of real-time operations in a way that traditional means (which usu- ally reduce to the values of the limiting matrix) do not. Some examples are given from the work on radar operators in Figures 1.2 to 1.5. All the predictions were validated at least to a first approximation. The computer model predicted that the maximum number of aircraft that could be han- dled would be four, because with that number the MFPTs were approaching the point of significant for- getting, and if we added the range of values implied by the variance of the MFPT, many MFPTs would lead to serious forgetting. Our empirical studies supported

this prediction. With four aircraft performance not only began to deteriorate, but one operator flew one of his aircraft off the radar plot and never found it again. (Fortunately this was in a simulated mission!) It is also interesting with respect to prediction 3, that at the end of one sortie with real aircraft, the operator suddenly glanced at another part of the radar and said, “Oh my God!” He had so ignored everything but the two air- craft that were close together that he had failed to attend to another of his aircraft for many seconds and had allowed it to stray into a designated flight path for civil aircraft leaving the United Kingdom for the continent. We can conclude that a quantitative single-channel model of visual dynamic attention can predict significant details of the real-time behavior of humans performing real tasks to “a good engineering approximation.”

ATTENTION: FROM HISTORY TO APPLICATION 7

FIGURE 1.1. Graphical representation of eye fixation transition probabilities during a sortie using live aircraft. Given that the entity on the tail of an arrow is currently fixated, the probability is that the next fixation will be on the entity at the head of the arrow. F, friendly; S, stranger; T, target; t, tote board. C1, C2, and C3 are features on the console other than the screen.