COGNITIVE PSYCHOLOGY

To Christine with love (M.W.E.)

To Ruth with love all ways (M.K.)

The only means of strengthening ones intellect is to make up ones mind about nothingto let the mind be a thoroughfare for all thoughts. Not a select party.

(John Keats)

Cognitive Psychology A Students HandbookFourth Edition

Michael W. Eysenck (Royal Holloway, University of London, UK) Mark Keane (University College Dublin, Ireland)

HOVE AND NEW YORK

First published 2000 by Psychology Press Ltd 27 Church Road, Hove, East Sussex BN3 2FA www.psypress.co.uk Simultaneously published in the USA and Canada by Taylor & Francis Inc. 325 Chestnut Street, Philadelphia, PA 19106 Psychology Press is an imprint of the Taylor & Francis Group This edition published in the Taylor & Francis e-Library, 2005. To purchase your own copy of this or any of Taylor & Francis or Routledges collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk. Reprinted 2000, 2001 Reprinted 2002 (twice) and 2003 by Psychology Press 27 Church Road, Hove, East Sussex BN3 2FA 29 West 35th Street, New York, NY 10001 2000 by Psychology Press Ltd All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN 0-203-62630-3 Master e-book ISBN

ISBN 0-203-62636-2 (Adobe eReader Format) ISBN 0-86377-550-0 (hbk) ISBN 0-86377-551-9 (pbk) Cover design by Hybert Design, Waltham St Lawrence, Berkshire

Contents

Preface 1. Introduction Cognitive psychology as a science Cognitive science Cognitive neuropsychology Cognitive neuroscience Outline of this book Chapter summary Further reading 2. Visual perception: Basic processes Introduction Perceptual organisation Depth and size perception Colour perception Brain systems Chapter summary Further reading 3. Perception, movement, and action Introduction Constructivist theories Direct perception Theoretical integration Motion, perception, and action Visually guided action

xii 1 1 5 13 18 25 26 27 28 28 28 34 43 48 56 57 58 58 59 64 68 70 71

vi

Perception of object motion Chapter summary Further reading 4. Object recognition Introduction Pattern recognition Marrs computational theory Cognitive neuropsychology approach Cognitive science approach Face recognition Chapter summary Further reading 5. Attention and performance limitations Introduction Focused auditory attention Focused visual attention Divided attention Automatic processing Action slips Chapter summary Further reading 6. Memory: Structure and processes Introduction The structure of memory Working memory Memory processes Theories of forgetting Theories of recall and recognition Chapter summary Further reading

79 87 89 90 90 91 96 106 109 116 128 129 130 130 132 136 147 155 160 165 166 167 167 167 172 182 187 194 203 204

vii

7.

Theories of long-term memory Introduction Episodic and semantic memory Implicit memory Implicit learning Transfer appropriate processing Amnesia Theories of amnesia Chapter summary Further reading

205 205 205 208 211 213 216 223 234 235 236 236 238 245 249 256 261 263 264 265 266 266 267 270 271 276 282 287 293 298

8.

Everyday memory Introduction Autobiographical memory Memorable memories Eyewitness testimony Superior memory ability Prospective memory Evaluation of everyday memory research Chapter summary Further reading

9.

Knowledge: Propositions and images Introduction What is a representation? What is a proposition? Propositions: Objects and relations Schemata, frames, and scripts What is an image? Some evidence Propositions versus images Kosslyns computational model of imagery The neuropsychology of visual imagery

viii

Connectionist representations Chapter summary Further reading 10. Objects, concepts, and categories Introduction Evidence on categories and categorisation The defining-attribute view The prototype view The exemplar-based view Explanation-based views of concepts Conceptual combination Concepts and similarity Evaluating theories of categorisation Neurological evidence on concepts Chapter summary Further reading 11. Speech perception and reading Introduction Listening to speech Theories of word recognition Cognitive neuropsychology Basic reading processes Word identification Routes from print to sound Chapter summary Further reading 12. Language comprehension Introduction Sentence processing Capacity theory

299 304 305 306 306 307 313 317 320 322 325 326 331 332 333 334 335 335 336 340 345 348 352 357 365 367 368 368 368 376

ix

Discourse processing Story processing Chapter summary Further reading 13. Language production Introduction Speech as communication Speech production processes Theories of speech production Cognitive neuropsychology: Speech production Cognitive neuroscience: Speech production Writing: Basic processes Cognitive neuropsychology: Writing Speaking and writing compared Language and thought Chapter summary Further reading 14. Problem solving: Puzzles, insight, and expertise Introduction Early research: The Gestalt school Newell and Simons problem-space theory Evaluating research on puzzles Re-interpreting the Gestalt findings From puzzles to expertise Evaluation of expertise research Learning to be an expert Cognitive neuropsychology of thinking Chapter summary Further reading 15. Creativity and discovery

379 386 397 398 399 399 399 401 403 410 412 414 419 425 426 428 430 431 431 433 438 446 449 452 461 461 465 466 467 468

x

Introduction Genius and talent General approaches to creativity Discovery using mental models Discovery by analogy Scientific discovery by hypothesis testing Evaluating problem-solving research Chapter summary Further reading 16. Reasoning and deduction Introduction Theoretical approaches to reasoning How people reason with conditionals Abstract-rule theory Mental models theory Domain-specific rule theories Probabilistic theory Cognitive neuropsychology of reasoning Rationality and evaluation of theories Chapter summary Further reading 17. Judgement and decision making Introduction Judgement research Decision making How flawed are judgement and decision making? Chapter summary Further reading 18. Cognition and emotion Introduction

468 468 469 473 476 480 483 486 487 488 488 491 492 502 506 513 515 518 519 520 521 522 522 523 531 534 535 536 537 537

xi

Does affect require cognition? Theories of emotional processing Emotion and memory Emotion, attention, and perception Conclusions on emotional processing Chapter summary Further reading 19. Present and future Introduction Experimental cognitive psychology Cognitive neuropsychology Cognitive science Cognitive neuroscience Present and future directions Chapter summary Further reading Glossary References Author index Subject index

537 543 549 556 561 563 564 565 565 565 568 570 573 575 576 577 579 591 657 680

Preface

Cognitive psychology has changed in several excit- ing ways in the few years since the third edition of this textbook. Of all the changes, the most dramatic has been the huge increase in the number of studies making use of sophisticated techniques (e.g., PET scans) to investigate human cognition. During the 1990s, such studies probably increased tenfold, and are set to increase still further during the early years of the third millennium. As a result, we now have four major approaches to cognitive psychology: experimental cognitive psychology based mainly on laboratory experiments; cognit- ive neuropsychology, which points up the effects of brain damage on cognition; cognitive science, with its emphasis on computational modelling; and cognitive neuroscience, which uses a wide range of techniques to study brain functioning. It is a worthwhile (but challenging) business to try to integrate information from these four approaches, and that it is exactly what we have tried to do in this book. As before, our busy professional lives have made it essential for us to work hard to avoid chaos. For example, the first author wrote several parts of the book in China, and other parts were written in Mexico, Poland, Russia, Israel, and the United States. The second author followed Joyces ghost, writing parts of the book between Dublin and Trieste. I (Michael Eysenck) would like to express my profound gratitude to my wife Christine, to whom this book (in common with the previous edition) is appropriately dedicated. I am also very grateful to our three children (Fleur, William, and Juliet) for their tolerance and understanding, just as was the case with the previous edition of this book. How- ever, when I look back to the writing of the third edition of this textbook, it amazes me how much they have changed over the last five years. Since I (Mark Keane) first collaborated on Cognitive Psychology: A Students Handbook in 1990 my professional life has undergone considerable change, from a post-doc in psychology to Professor of Computer Science. My original motivation in writing this text was to influence the course of cognitive psychology as it was then developing, to encourage its extension in a computational direction. Looking back over the last 10 years, I am struck by the slowness of change in the introduction of these ideas. The standard psychology undergraduate degree does a very good job at giving students the tools for the empirical exploration of the mind. However, few courses give students the tools for the theoretical elaboration of the topic. In this respect, the discipline gets a could do better rather than an excellent on the mark sheet. We are very grateful to several people for reading an entire draft of this book, and for offering valuable advice on how it might be improved. They include Ruth Byrne, Liz Styles, Trevor Harley, and Robert Logie. We would also like to thank those who commented on various chapters: John Towse, Steve Anderson, James Hampton, Fernand Gobet, Evan Heit, Alan Parkin, David Over, Ken Manktelow, Ken Gilhooly, Peter Ayton, Clare Harries, George Mather, Mark Georgeson, Gerry Altmann, Nick Wade, Mick Power, David Hardman, John Richardson, Vicki Bruce, Gillian Cohen, and Jonathan St.B.T.Evans. Michael Eysenck and Mark Keane

1 Introduction

COGNITIVE PSYCHOLOGY AS A SCIENCE In the years leading up to the millennium, people made increased efforts to understand each other and their own inner, mental space. This concern was marked with a tidal wave of research in the field of cognitive psychology, and by the emergence of cognitive science as a unified programme for studying the mind. In the popular media, there are numerous books, films, and television programmes on the more accessible aspects of cognitive research. In scientific circles, cognitive psychology is currently a thriving area, dealing with a bewildering diversity of phenomena, including topics like attention, perception, learning, memory, language, emotion, concept formation, and thinking. In spite of its diversity, cognitive psychology is unified by a common approach based on an analogy between the mind and the digital computer; this is the information-processing approach. This approach is the dominant paradigm or theoretical orientation (Kuhn, 1970) within cognitive psychology, and has been for some decades. Historical roots of cognitive psychology The year 1956 was critical in the development of cognitive psychology. At a meeting at the Massachusetts Institute of Technology, Chomsky gave a paper on his theory of language, George Miller presented a paper on the magic number seven in short-term memory (Miller, 1956), and Newell and Simon discussed their very influential computational model called the General Problem Solver (discussed in Newell, Shaw, & Simon, 1958; see also Chapter 15). In addition, the first systematic attempt to consider concept formation from a cognitive perspective was reported (Bruner, Goodnow, & Austin, 1956). The field of Artificial Intelligence was also founded in 1956 at the Dartmouth Conference, which was attended by Chomsky, McCarthy, Minsky, Newell, Simon, and Miller (see Gardner, 1985). Thus, 1956 witnessed the birth of both cognitive psychology and cognitive science as major disciplines. Books devoted to aspects of cognitive psychology began to appear (e.g., Broadbent, 1958; Bruner et al., 1956). However, it took several years before the entire information-processing viewpoint reached undergraduate courses (Lachman, Lachman, & Butterfield, 1979; Lindsay & Norman, 1977). Information processing: Consensus Broadbent (1958) argued that much of cognition consists of a sequential series of processing stages. When a stimulus is presented, basic perceptual processes occur, followed by attentional processes that transfer some

2

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

of the products of the initial perceptual processing to a short-term memory store. Thereafter, rehearsal serves to maintain information in the short-term memory store, and some of the information is transferred to a long-term memory store. Atkinson and Shiffrin (1968; see also Chapter 6) put forward one of the most detailed theories of this type. This theoretical approach provided a simple framework for textbook writers. The stimulus input could be followed from the sense organs to its ultimate storage in long-term memory by successive chapters on perception, attention, short-term memory, and long-term memory The crucial limitation with this approach is its assumption that stimuli impinge on an inactive and unprepared organism. In fact, processing is often affected substantially by the individuals past experience, expectations, and so on. We can distinguish between bottom-up processing and top-down processing. Bottom-up or stimulusdriven processing is directly affected by stimulus input, whereas top-down or conceptually driven processing is affected by what the individual contributes (e.g., expectations determined by context and past experience). As an example of top-down processing, it is easier to read the word well in poor handwriting if it is presented in the sentence context, I hope you are quite___, than when it is presented on its own. The sequential stage model deals primarily with bottom-up or stimulus-driven processing, and its failure to consider top-down processing adequately is its greatest limitation. During the 1970s, theorists such as Neisser (1976) argued that nearly all cognitive activity consists of interactive bottom-up and top-down processes occurring together (see Chapter 4). Perception and remembering might seem to be exceptions, because perception depends heavily on the precise stimuli presented (and thus on bottom-up processing), and remembering depends crucially on stored information (and thus on top-down processing). However, perception is influenced by the perceivers expectations about to-be-presented stimuli (see Chapters 2, 3, and 4), and remembering is influenced by the precise environmental cues to memory that are available (see Chapter 6). By the end of the 1970s, most cognitive psychologists agreed that the information-processing paradigm was the best way to study human cognition (see Lachman et al., 1979): People are autonomous, intentional beings interacting with the external world. The mind through which they interact with the world is a general-purpose, symbol-processing system (symbols are patterns stored in long-term memory which designate or point to structures outside themselves; Simon & Kaplan, 1989, p. 13). Symbols are acted on by processes that transform them into other symbols that ultimately relate to things in the external world. The aim of psychological research is to specify the symbolic processes and representations underlying performance on all cognitive tasks. Cognitive processes take time, and predictions about reaction times can often be made. The mind is a limited-capacity processor having structural and resource limitations. The symbol system depends on a neurological substrate, but is not wholly constrained by it. Many of these ideas stemmed from the view that human cognition resembles the functioning of computers. As Herb Simon (1980, p. 45) expressed it, It might have been necessary a decade ago to argue for the commonality of the information processes that are employed by such disparate systems as computers and human nervous systems. The evidence for that commonality is now over-whelming. (See Simon, 1995, for an update of this view.) The information-processing framework is continually developing as information technology develops. The computational metaphor is always being extended as computer technology develops. In the 1950s and

1. INTRODUCTION

3

1960s, researchers mainly used the general properties of the computer to understand the mind (e.g., that it had a central processor and memory registers). Many different programming languages had been developed by the 1970s, leading to various aspects of computer software and languages being used (e.g., JohnsonLaird, 1977, on analogies to language understanding). After that, as massively parallel machines were developed, theorists returned to the notion that cognitive theories should be based on the parallel processing capabilities of the brain (Rumelhart, McClelland, & the PDP Research Group, 1986). Information processing: Diversity Cognitive science is a trans-disciplinary grouping of cognitive psychology, artificial intelligence, linguistics, philosophy, neuroscience, and anthropology. The common aim of these disciplines is the understanding of the mind. To simplify matters, we will focus mainly on the relationship between cognitive psychology and artificial intelligence. At the risk of oversimplification, we can identify four major approaches within cognitive psychology: Experimental cognitive psychology: it follows the experimental tradition of cognitive psychology, and involves no computational modelling. Cognitive science: it develops computational models to understand human cognition. Cognitive neuropsychology: it studies patterns of cognitive impairment shown by brain-damaged patients to provide valuable information about normal human cognition. Cognitive neuroscience: it uses several techniques for studying brain functioning (e.g., brain scans) to understand human cognition. There are various reasons why these distinctions are less neat and tidy in reality than we have implied. First, terms such as cognitive science and cognitive neuroscience are sometimes used in a broader and more inclusive way than we have done. Second, there has been a rapid increase in recent years in studies that combine elements of more than one approach. Third, some have argued that experimental cognitive psychologists and cognitive scientists are both endangered species, given the galloping expansion of cognitive neuropsychology and cognitive neuroscience. In this book, we will provide a synthesis of the insights emerging from all four approaches. The approach taken by experimental cognitive psychologists has been in existence for several decades, so we will focus mainly on the approaches of cognitive scientists, cognitive neuropsychologists, and cognitive neuroscientists in the following sections. Before doing so, however, we will consider some traditional ways of obtaining evidence about human cognition. Empirical methods In most of the research discussed in this book, cognitive processes and structures were inferred from participants behaviour (e.g., speed and/or accuracy of performance) obtained under well controlled conditions. This approach has proved to be very useful, and the data thus obtained have been used in the development and subsequent testing of most theories in cognitive psychology. However, there are two major potential problems with the use of such data: 1. Measures of the speed and accuracy of performance provide only indirect information about the internal processes and structures of central interest to cognitive psychologists.

4

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

2. Behavioural data are usually gathered in the artificial surroundings of the laboratory. The ways in which people behave in the laboratory may differ greatly from the ways they behave in everyday life (see Chapter 19). Cognitive psychologists do not rely solely on behavioural data to obtain useful information from their participants. An alternative way of studying cognitive processes is by making use of introspection, which is defined by the Oxford English Dictionary as examination or observation of ones own mental processes. Introspection depends on conscious experience, and each individuals conscious experience is personal and private. In spite of this, it is often assumed that introspection can provide useful evidence about some mental processes. Nisbett and Wilson (1977) argued that introspection is practically worthless, supporting their argument with examples. In one study, participants were presented with a display of five essentially identical pairs of stockings, and decided which pair was the best. After they had made their choice, they indicated why they had chosen that particular pair. Most participants chose the rightmost pair, and so their decisions were actually affected by relative spatial position. However, the participants strongly denied that spatial position had played any part in their decision, referring instead to slight differences in colour, texture, and so on among the pairs of stockings as having been important. Nisbett and Wilson (1977, p. 248) claimed that people are generally unaware of the processes influencing their behaviour: When people are asked to report how a particular stimulus influenced a particular response, they do so not by consulting a memory of the mediating process, but by applying or generating causal theories about the effects of that type of stimulus on that type of response. This view was supported by the discovery that an individuals introspections about what is determining his or her behaviour are often no more accurate than the guesses made by others. The limitations of introspective evidence are becoming increasingly clear. For example, consider research on implicit learning, which involves learning complex material without the ability to verbalise what has been learned. There is reasonable evidence for the existence of implicit learning (see Chapter 7). There is even stronger evidence for implicit memory, which involves memory in the absence of conscious recollection. Normal and brain-damaged individuals can exhibit excellent memory performance even when they show no relevant introspective evidence (see Chapter 7). Ericsson and Simon (1980, 1984) argued that Nisbett and Wilson (1977) had overstated the case against introspection. They proposed various criteria for distinguishing between valid and invalid uses of introspection: It is preferable to obtain introspective reports during the performance of a task rather than retrospectively, because of the fallibility of memory. Participants are more likely to produce accurate introspections when describing what they are attending to, or thinking about, than when required to interpret a situation or their own thought processes. People cannot usefully introspect about several kinds of processes (e.g., neuronal processes; recognition processes). Careful consideration of the studies that Nisbett and Wilson (1977) regarded as striking evidence of the worthlessness of introspection reveals that participants generally provided retrospective interpretations about information that had probably never been fully attended to. Thus, their findings are consistent with the proposed guidelines for the use of introspection (Crutcher, 1994; Ericsson & Simon, 1984).

1. INTRODUCTION

5

FIGURE 1.1 A flowchart of a bad theory about how we understand sentences.

In sum, introspection is sometimes useful, but there is no conscious awareness of many cognitive processes or their products. This point is illustrated by the phenomena of implicit learning and implicit memory, but numerous other examples of the limitations of introspection will be presented throughout this book. COGNITIVE SCIENCE Cognitive scientists develop computational models to understand human cognition. A decent computational model can show us that a given theory can be specified and allow us to predict behaviour in new situations. Mathematical models were used in experimental psychology long before the emergence of the informationprocessing paradigm (e.g., in IQ testing). These models can be used to make predictions, but often lack an explanatory component. For example, committing three traffic violations is a good predictor of whether a person is a bad risk for car insurance, but it is not clear why. One of the major benefits of the computational models developed in cognitive science is that they can provide both an explanatory and predictive basis for a phenomenon (e.g., Keane, Ledgeway, & Duff, 1994; Costello & Keane, 2000). We will focus on computational models in this section, because they are the hallmark of the cognitive science approach.

6

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

Computational modelling: From flowcharts to simulations In the past, many experimental cognitive psychologists stated their theories in vague verbal statements. This made it hard to decide whether the evidence fitted the theory. In contrast, cognitive scientists produce computer programs to represent cognitive theories with all the details made explicit. In the 1960s and 1970s, cognitive psychologists tended to use flowcharts rather than programs to characterise their theories. Computer scientists use flowcharts as a sort of plan or blue-print for a program, before they write the detailed code for it. Flowcharts are more specific than verbal descriptions, but can still be underspecified if not accompanied by a coded program. An example of a very inadequate flowchart is shown in Figure 1.1. This is a flowchart of a bad theory about how we understand sentences. It assumes that a sentence is encoded in some form and then stored. After that, a decision process (indicated by a diamond) determines if the sentence is too long. If it is too long, then it is broken up and we return to the encode stage to re-encode the sentence. If it is ambiguous, then its two senses are distinguished, and we return to the encode stage. If it is not ambiguous, then it is stored in long-term memory. After one sentence is stored, we return to the encode stage to consider the next sentence. In the days when cognitive psychologists only used flowcharts, sarcastic questions abounded, such as, What happens in the boxes? or What goes down the arrows?. Such comments point to genuine criticisms. We need to know what is meant by encode sentence, how long is too long, and how sentence ambiguity is tested. For example, after deciding that only a certain length of sentence is acceptable, it may turn out that it is impossible to decide whether the sentence portions are ambiguous without considering the entire sentence. Thus, the boxes may look all right at a superficial glance, but real contradictions may appear when their contents are specified. In similar fashion, exactly what goes down the arrows is critical. If one examines all the arrows converging on the encode sentence box, it is clear that more needs to be specified. There are four different kinds of thing entering this box: an encoded sentence from the environment; a sentence that has been broken up into bits by the split-sentence box; a sentence that has been broken up into several senses; and a command to consider the next sentence. Thus, the encode box has to perform several specific operations. In addition, it may have to record the fact that an item is either a sentence or a possible meaning of a sentence. Several other complex processes have to be specified within the encode box to handle these tasks, but the flowchart sadly fails to addresses these issues. The gaps in the flowchart show some similarities with those in the formula shown in Figure 1.2. Not all theories expressed as flowcharts possess the deficiencies of the one described here. However, implementing a theory as a program is a good method for checking that it contains no hidden assumptions or vague terms. In the previous example, this would involve specifying the form of the input sentences, the nature of the storage mechanisms, and the various decision processes (e.g., those about sentence length and ambiguity). These computer programs are written in artificial intelligence programming languages, usually LISP (Norvig, 1992) or PROLOG (Shoham, 1993). There are many issues surrounding the use of computer simulations and the ways in which they do and do not simulate cognitive processes (Cooper, Fox, Farrington, & Shallice, 1996; Costello & Keane, 2000; Palmer & Kimchi, 1986). Palmer and Kimchi (1986) argued that it should be possible to decompose a theory successively through a number of levels (from descriptive statement to flowchart to specific functions in a program) until one reaches a written program. In addition, they argued that it should be possible to draw a line at some level of decomposition, and say that everything above that line is psychologically plausible or meaningful, whereas everything below it is not. This issue of separating psychological aspects of the program from other aspects arises because there will always be parts of the program that have little to do

1. INTRODUCTION

7

FIGURE 1.2 The problem of being specific. Copyright 1977 by Sidney Harris in American Scientist Magazine. Reproduced with permission of the author.

with the psychological theory, but which are there simply because of the particular programming language being used and the machine on which the program is running. For example, in order to see what the program is doing, it is necessary to have print commands in the program which show the outputs of various stages in the computers screen. However, no-one would argue that such print commands form part of the psychological model. Cooper et al. (1996) argue that psychological theories should not beThree issues sorrounding computer simulation:

Is it possible to decompose a theory until one reaches the level of a written program? Is it possible to separate psychological aspects of a program f rom other aspects? Are there differences in reaction time between programs and human participants?

described using natural language at all, but that a formal specification language should be used. This would be a very precise language, like a logic, that would be directly executable as a program. Other issues arise about the relationship between the performance of the program and the performance of human participants (Costello & Keane, 2000). For example, it is seldom meaningful to relate the speed of the program doing a simulated task to the reaction time taken by human participants, because the processing times of programs are affected by psychologically irrelevant features. Programs run faster on more

8

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

powerful computers, or if the programs code is interpreted rather than compiled. However, the various materials that are presented to the program should result in differences in program operation time that correlate closely with differences in participants reaction times in processing the same materials. At the very least, the program should be able to reproduce the same outputs as participants when given the same inputs. Computational modelling techniques The general characteristics of computational models of cognition have been discussed at some length. It is now time to deal with some of the main types of computational model that have been used in recent years. Three main types are outlined briefly here: semantic networks; production systems; and connectionist networks.Semantic networks

Consider the problem of modelling what we know about the world (see Chapter 9). There is a long tradition from Aristotle and the British empiricist school of philosophers (Locke, Hume, Mill, Hartley, Bain) which proposes that all knowledge is in the form of associations. Three main principles of association have been proposed: Contiguity: two things become associated because they occurred together in time. Similarity: two things become associated because they are alike. Contrast: two things become associated because they are opposites. There is a whole class of cognitive models owing their origins to these ideas; they are called associative or semantic or declarative networks. Semantic networks have the following general characteristics: Concepts are represented by linked nodes that form a network. These links can be of various kinds; they can represent very general relations (e.g., is-associated-with or is-similar-to) or specific, simple relations like is-a (e.g., John is-a policeman), or more complete relations like play, hit, kick. The nodes themselves and the links among nodes can have various activation strengths representing the similarity of one concept to another. Thus, for example, a dog and a cat node may be connected by a link with an activation of 0.5, whereas a dog and a pencil may be connected by a link with a strength of 0.1. Learning takes the form of adding new links and nodes to the network or changing the activation values on the links between nodes. For example, in learning that two concepts are similar, the activation of a link between them may be increased. Various effects (e.g., memory effects) can be modelled by allowing activation to spread throughout the network from a given node or set of nodes. The way in which activation spreads through a network can be determined by a variety of factors For example, it can be affected by the number of links between a given node and the point of activation, or by the amount of time that has passed since the onset of activation. Part of a very simple network model is shown in Figure 1.3. It corresponds closely to the semantic network model proposed by Collins and Loftus (1975). Such models have been successful in accounting for a various findings. Semantic priming effects in which the word dog is recognised more readily if it is

1. INTRODUCTION

9

FIGURE 1.3 A schematic diagram of a simple semantic network with nodes for various concepts (i.e., dog, cat), and links between these nodes indicating the differential similarity of these concepts to each other.

preceded by the word cat (Meyer & Schvaneveldt, 1971) can be easily modelled using such networks (see Chapter 12). Ayers and Reder (1998) have used semantic networks to understand misinformation effects in eyewitness testimony (see Chapter 8). At their best, semantic networks are both flexible and elegant modelling schemes.Production systems

Another popular approach to modelling cognition involves production systems. These are made up of productions, where a production is an IF THEN rule. These rules can take many forms, but an example that is very useful in everyday life is, If the green man is lit up, then cross the road. In a typical production system model, there is a long-term memory that contains a large set of these IFTHEN rules. There is also a working memory (i.e., a system holding information that is currently being processed). If information from the environment that the green man is lit up reaches working memory, it will match the IF-part of the rule in long-term memory, and trigger the THEN-part of the rule (i.e., cross the road). Production systems have the following characteristics: They have numerous IFTHEN rules. They have a working memory containing information. The production system operates by matching the contents of working memory against the IF-parts of the rules and executing the THEN-parts. If some information in working memory matches the IF-part of many rules, there may be a conflictresolution strategy selecting one of these rules as the best one to be executed.

10

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

FIGURE 1.4 A schematic diagram of a simple production system.

Consider a very simple production system operating on lists of letters involving As and Bs (see Figure 1.4). The system has two rules: 1. IF a list in working memory has an A at the end THEN replace the A with AB. 2. IF a list in working memory has a B at the end THEN replace the B with an A. If we give this system different inputs in the form of different lists of letters, then different things happen. If we give it CCC, this will be stored in working memory but will remain unchanged, because it does not match either of the IF-parts of the two rules. If we give it A, then it will be notified by the rules after the A is stored in working memory. This A is a list of one item and as such it matches rule 1. Rule 1 has the effect of replacing the A with AB, so that when the THEN-part is executed, working memory will contain an AB. On the next cycle, AB does not match rule 1 but it does match rule 2. As a result, the B is replaced by an A, leaving an AA in working memory. The system will next produce AAB, then AAAB, and so on. Many aspects of cognition can be specified as sets of IFTHEN rules. For example, chess knowledge can readily be represented as a set of productions based on rules such as, If the Queen is threatened, then move the Queen to a safe square. In this way, peoples basic knowledge of chess can be modified as a collection of productions, and gaps in this knowledge as the absence of some productions. Newell and Simon (1972) first established the usefulness of production system models in characterising cognitive processes like problem solving and reasoning (see Chapter 14). However, these models have a wider applicability. Anderson (1993) has modelled human learning using production systems (see Chapter 14), and others have used them to model reinforcement behaviour in rats, and semantic memory (Holland et al., 1986).Connectionist networks

Connectionist networks, neural networks, or parallel distributed processing models as they are variously called, are relative newcomers to the computational modelling scene. All previous techniques were marked by the need to program explicitly all aspects of the model, and by their use of explicit symbols to represent concepts. Connectionist networks, on the other hand, can to some extent program themselves, in that they can learn to produce specific outputs when certain inputs are given to them. Furthermore, connectionist modellers often reject the use of explicit rules and symbols and use distributed representations, in which concepts are characterised as patterns of activation in the network (see Chapter 9). Early theoretical proposals about the feasibility of learning in neural-like networks were made by McCulloch and Pitts (1943) and by Hebb (1949). However, the first neural network models, called

1. INTRODUCTION

11

FIGURE 1.5 A multi-layered connectionist network with a layer of input units, a layer of internal representation units or hidden units, and a layer of output units. Input patterns can be encoded, if there are enough hidden units, in a form that allows the appropriate output pattern to be generated from a given input pattern. Reproduced with permission from David E. Rumelhart & James L.McClelland, Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 1), published by the MIT Press, 1986, the Massachusetts Institute of Technology.

Perceptrons, were shown to have several limitations (Minsky & Papert, 1988). By the late 1970s, hardware and software develpments in computing offered the possibility of constructing more complex networks overcoming many of these original limitations (e.g., Rumelhart, McClelland, & the PDP Research Group, 1986; McClelland, Rumelhart, & the PDP Research Group, 1986). Connectionist networks typically have the following characteristics (see Figure 1.5): The network consists of elementary or neuron-like units or nodes connected together so that a single unit has many links to other units. Units affect other units by exciting or inhibiting them. The unit usually takes the weighted sum of all of the input links, and produces a single output to another unit if the weighted sum exceeds some threshold value. The network as a whole is characterised by the properties of the units that make it up, by the way they are connected together, and by the rules used to change the strength of connections among units.

12

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

Networks can have different structures or layers; they can have a layer of input links, intermediate layers (of so-called hidden units), and a layer of output units. A representation of a concept can be stored in a distributed manner by a pattern of activation throughout the network. The same network can store many patterns without them necessarily interfering with each other if they are sufficiently distinct. An important learning rule used in networks is called backward propagation of errors (BackProp). In order to understand connectionist networks fully, let us consider how individual units act when activation impinges on them. Any given unit can be connected to several other units (see Figure 1.6). Each of these other units can send an excitatory or an inhibitory signal to the first unit. This unit generally takes a weighted sum of all these inputs. If this sum exceeds some threshold, it produces an output. Figure 1.6 shows a simple diagram of just such a unit, which takes the inputs from a number of other units and sums them to produce an output if a certain threshold is exceeded. These networks can model cognitive behaviour without recourse to the kinds of explicit rules found in production systems. They do this by storing patterns of activation in the network that associate various inputs with certain outputs. The models typically make use of several layers to deal with complex behaviour. One layer consists of input units that encode a stimulus as a pattern of activation in those units. Another layer is an output layer, which produces some response as a pattern of activation. When the network has learned to produce a particular response at the output layer following the presentation of a particular stimulus at the input layer, it can exhibit behaviour that looks as if it had learned a rule of the form IF such-and-such is the case THEN do so-and-so. However, no such rules exist explicitly in the model. Networks learn the association between different inputs and outputs by modifying the weights on the links between units in the net. In Figure 1.6, we see that the weight on the links to a unit, as well as the activation of other units, plays a crucial role in computing the response of that unit. Various learning rules modify these weights in systematic ways. When we apply such learning rules to a network, the weights on the links are modified until the net produces the required output patterns given certain input patterns. One such learning rule is called backward propagation of errors or BackProp. BackProp allows a network to learn to associate a particular input pattern with a given output pattern. At the start of the learning period, the network is set up with random weights on the links among the units. During the early stages of learning, after the input pattern has been presented, the output units often produce the incorrect pattern or response. BackProp compares the imperfect pattern with the known required response, noting the errors that occur. It then back-propagates activation through the network so that the weights between the units are adjusted to produce the required pattern. This process is repeated with a particular stimulus pattern until the network produces the required response pattern. Thus, the model can be made to learn the behaviour with which the cognitive scientist is concerned, rather than being explicitly programmed to do so. Networks have been used to produce very interesting results. Several examples will be discussed throughout the text (see, for examples, Chapters 2, 10, and 16), but one concrete example will be mentioned here. Sejnowski and Rosenberg (1987) produced a connectionist network called NETtalk, which takes an English text as its input and produces reasonable English speech output. Even though the network is trained on a limited set of words, it can pronounce the words from new text with about 90% accuracy. Thus, the network seems to have learned the rules of English pronunciation, but it has done so without having explicit rules that combine and encode sounds. Connectionist models such as NETtalk have great Wow! value, and are the subject of much research interest. Some researchers might object to our classification of connectionist networks as merely one among

1. INTRODUCTION

13

FIGURE 1.6 Diagram showing how the inputs from a number of units are combined to determine the overall input to unit-i. Unit-i has a threshold of 1; so if its net input exceeds 1 then it will respond with +1, but if the net input is less than 1 then it will respond with 1.

a number of modelling techniques. However, others have argued that connectionism represents an alternative to the information-processing paradigm (Smolensky, 1988; Smolensky, Legendre, & Miyata, 1993). Indeed, if one examines the fundamental tenets of the information-processing framework, then connectionist schemes violate one or two. For example, symbol manipulation of the sort found in production systems does not seem to occur in connectionist networks. We will return to the complex issues raised by connectionist networks later in the book. COGNITIVE NEUROPSYCHOLOGY Cognitive neuropsychology is concerned with the patterns of cognitive performance in brain-damaged patients. Those aspects of cognition that are intact or impaired are identified, with this information being of value for two main reasons. First, the cognitive performance of brain-damaged patients can often be explained by theories within cognitive psychology. Such theories specify the processes or mechanisms involved in normal cognitive functioning, and it should be possible in principle to account for many of the cognitive impairments of brain-damaged patients in terms of selective damage to some of those mechanisms. Second, it may be possible to use information from brain-damaged patients to reject theories proposed by cognitive psychologists, and to propose new theories of normal cognitive functioning. According to Ellis and Young (1988, p. 4), a major aim of cognitive neuropsychology:

14

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

is to draw conclusions about normal, intact cognitive processes from the patterns of impaired and intact capabilities seen in brain-injured patientsthe cognitive neuropsychologist wishes to be in a position to assert that observed patterns of symptoms could not occur if the normal, intact cognitive system were not organised in a certain way. The intention is that there should be bi-directional influences of cognitive psychology on cognitive neuropsychology, and of cognitive neuropsychology on cognitive psychology. Historically, the former influence was the greater one, but the latter has become more important. Before discussing the cognitive neuropsychological approach in more detail, we will discuss a concrete example of cognitive neuropsychology in operation. Atkinson and Shiffrin (1968) argued that there is an important distinction between a short-term memory store and a long-term memory store, and that information enters into the long-term store through rehearsal and other processing activities in the short-term store (see Chapter 6). Relevant evidence was obtained by Shallice and Warrington (1970). They studied a brain-damaged patient, KF, who seemed to have severely impaired short-term memory, but essentially intact long-term memory. The study of this patient served two important purposes. First, it provided evidence to support the theoretical distinction between two memory systems. Second, it pointed to a real deficiency in the theoretical model of Atkinson and Shiffrin (1968). If, as this model suggests, long-term learning and memory depend on the short-term memory system, then it is surprising that someone with a grossly deficient short-term memory system also has normal long-term memory. The case of KF shows very clearly the potential power of cognitive neuropsychology. The study of this one patient provided strong evidence that the dominant theory of memory at the end of the 1960s was seriously deficient. This is no mean achievement for a study on one patient! Cognitive neuropsychological evidence How do cognitive neuropsychologists set about the task of understanding how the cognitive system functions? A crucial goal is the discovery of dissociations, which occur when a patient performs normally on one task but is impaired on a second task. In the case of KF, a dissociation was found between performance on short-term memory tasks and on long-term memory tasks. Such evidence can be used to argue that normal individuals possess at least two separate memory systems. There is a potential problem in drawing sweeping conclusions from single dissociations. A patient may perform poorly on one task and well on a second task simply because the first task is more complex than the second, rather than because the first task involves specific skills that have been affected by brain damage. The solution to this problem is to look for double dissociations. A double dissociation between two tasks (1 and 2) is shown when one patient performs normally on task 1 and at an impaired level on task 2, and another patient performs normally on task 2 and at an impaired level on task 1. If a double dissociation can be shown, then the results cannot be explained in terms of one task being harder than the other. In the case of short-term and long-term memory, such a double dissocation has been shown. KF had impaired short-term memory but intact long-term memory, whereas amnesic patients have severely deficient long-term memory but intact short-term memory (see Chapter 7). These findings suggest there are two distinct memory systems which can suffer damage separately from each other. If brain damage were usually very limited in scope, and affected only a single cognitive process or mechanism, then cognitive neuropsychology would be a fairly simple enterprise. In fact, brain damage is often rather extensive, so that several cognitive systems are all impaired to a greater or lesser extent. This

1. INTRODUCTION

15

means that much ingenuity is needed to make sense of the tantalising glimpses of human cognition provided by brain-damaged patients. Theoretical assumptions Most cognitive neuropsychologists subscribe to the following assumptions (with the exception of the last one): The cognitive system exhibits modularity, i.e., there are several relatively independent cognitive processes or modules, each of which functions to some extent in isolation from the rest of the processing system; brain damage typically impairs only some of these modules. There is a meaningful relationship between the organisation of the physical brain and that of the mind; this assumption is known as isomorphism. Investigation of cognition in brain-damaged patients can tell us much about cognitive processes in normal individuals; this assumption is closely bound up with the other assumptions. Most patients can be categorised in terms of syndromes, each of which is based on co-occurring sets of symptoms. Syndromes The traditional approach within neuropsychology made much use of syndromes. It was claimed that certain sets of symptoms or impairments are usually found together, and each set of co-occurring symptoms was used to define a separate syndrome (e.g., amnesia; dyslexia). This syndrome-based approach allows us to impose some order on the numerous brain-damaged patients who have been studied by assigning them to a fairly small number of categories. It is also of use in identifying those areas of the brain mainly responsible for cognitive function such as language, because we can search for those parts of the brain damaged in all those patients having a given syndrome. In spite of its uses, the syndrome-based approach has substantial problems. It exaggerates the similarities among different patients allegedly suffering from the same syndrome. In addition, those symptoms or impairments said to form a syndrome may be found in the same patients solely because the underlying cognitive processes are anatomically adjacent. There have been attempts to propose more specific syndromes or categories based on our theoretical understanding of cognition. However, the discovery of new patients with unusual patterns of deficits, and the occurrence of theoretical advances, mean that the categorisation system is constantly changing. As Ellis (1987) pointed out, a syndrome thought at time t to be due to damage to a single unitary module is bound to have fractionated by time t+2 years into a host of awkward subtypes. How should cognitive neuropsychologists react to these problems? Some cognitive neuro-psychologists (e.g., Parkin, 1996) argue that it makes sense to carry out group studies in which patients with the same syndrome are considered together. He introduced what he called the significance implies homogeneity [uniformity] rule. According to this rule, if a group of subjects exhibits significant hetereogeneity [variability] then they will not be capable of generating statistically significant group differences (Parkin, 1996, p. 16). The potential problem with this rule is that a group of patients can show a significant effect even though a majority of the individual patients fail to show the effect. Ellis (1987) argued that cognitive neuropsychology should proceed on the basis of intensive single-case studies in which individual patients are studied on a wide range of tasks. An adequate theory of cognition should be as applicable to the individual case as to groups of individuals, and so single-case studies provide

16

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

a perfectly adequate test of cognitive theories. The great advantage of this approach is that there is no need to make simplifying assumptions about which patients do and do not belong to the same syndrome. Another argument for single-case studies is that it is often not possible to find a group of patients showing very similar cognitive deficits. As Shallice (1991, p. 432) pointed out, as finer and finer aspects of the cognitive architecture are investigated in attempts to infer normal function, neuropsychology will be forced to resort more and more to single-case studies. Ellis (1987) may have overstated the value of single-case studies. If our theoretical understanding of an area is rather limited, it may make sense to adopt the syndrome-based approach until the major theoretical issues have been clarified. Furthermore, many experimental cognitive psychologists disapprove of attaching great theoretical significance to findings from individuals who may not be representative even of braindamaged patients. As Shallice (1991, p. 433) argued: A selective impairment found in a particular task in some patient could just reflect: the patients idiosyncratic strategy, the greater difficulty of that task compared with the others, a premorbid lacuna [gap] in that patient, or the way a reorganised system but not the original normal system operates. A reasonable compromise position is to carry out a number of single-case studies. If a theoretically crucial dissociation is found in a single patient, then there are various ways of interpreting the data. However, if the same dissociation is obtained in a number of individual patients, it is less likely that all the patients had atypical cognitive systems prior to brain damage, or that they have all made use of similar compensatory strategies.Modularity

The whole enterprise of cognitive neuropsychology is based on the assumption that there are numerous modules or cognitive processors in the brain. These modules function relatively independently, so that damage to one module does not directly affect other modules. Modules are anatomically distinct, so that brain damage will often affect some modules while leaving others intact. Cognitive neuropsychology may help the discovery of these major building blocks of cognition. A double dissociation indicates that two tasks make use of different modules or cognitive processors, and so a series of double dissociations can be

Syndrome-based approach vs. single-case studies syndrome-based approach Advantages Provides a means of imposing order and categorising patients. Allows identification of cognitive functions of brain areas. Useful while major theoretical issues remain to be clarified. Disadvantages Oversimplification based on theoretical assumptions. Exaggeration of similarities among patients. Single-case studies Advantages Avoids oversimplifying assumptions, No need to find groups of patients with very similar cognitive deficits. Disadvantages Evidence lacks generalisability and can even be misleading.

used to provide a sketch-map of our modular cognitive system.

1. INTRODUCTION

17

The notion of modularity was emphasised by Fodor (1983), who identified the following distinguishing features of modules: Informational encapsulation: each module functions independently from the functioning of other modules. Domain specificity: each module can process only one kind of input (e.g., words; faces). Mandatory or compulsory operation: the functioning of a module is not under any form of voluntary control. Innateness: modules are inborn. Fodors ideas have been influential. However, many psychologists have criticised mandatory operation and innateness as criteria for modularity. Some modules may operate automatically, but there is little evidence to suggest that they all do. It is implausible to assume the innateness of modules underlying skills such as reading and writing, as these are skills that the human race has developed only comparatively recently. From the perspective of cognitive neuropsychologists, these criticisms do not pose any special problems. If the assumptions of information encapsulation and domain specificity remain tenable, then data from brain-damaged patients can continue to be used in the hunt for cognitive modules. This would still be the case even if it turned out that several modules or cognitive processors were neither mandatory nor innate. It is not only cognitive neuropsychologists who subscribe to the notion of modularity. Most experimental cognitive psychologists, cognitive scientists, and cognitive neuroscientists also believe in modularity. The four groups differ mainly in terms of their preferred methods for showing modularity.Isomorphism

Cognitive neuropsychologists assume there is a meaningful relationship between the way in which the brain is organised at a physical level and the way in which the mind and its cognitive modules are organised. This assumption has been called isomorphism, meaning that two things (e.g., brain and mind) have the same shape or form. Thus, it is expected that each module will have a different physical location within the brain. If this expectation is disconfirmed, then cognitive neuropsychology and cognitive neuroscience will become more complex enterprises. An assumption that is related to isomorphism is that there is localisation of function, meaning that any specific function or process occurs in a given location within the brain (Figure 1.7). The notion of localisation of function seems to be in conflict with the connectionist account, according to which a process (e.g., activation of a concept) can be distributed over a wide area of the brain. There is as yet no definitive evidence to support one view over the other.Evaluation

Are the various theoretical assumptions underlying cognitive neuropsychology correct? It is hard to tell. Modules do not actually exist, but are convenient theoretical devices used to clarify our understanding. Therefore, the issue of whether the theoretical assumptions are valuable or not is probably best resolved by considering the extent to which cognitive neuropsychology is successful in increasing our knowledge of cognition. In other words, the proof of the pudding is in the eating. Farah (1994) argued that the evidence does not support what she termed the locality assumption, according to which damage to one module has only local effects. According to Farah (1994, p. 101), The conclusion that the locality assumption may be false is a disheartening one. It undercuts much of the special appeal of neuropsychological architecture.

18

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

FIGURE 1.7 PET scans can be used to show localisation of function within the brain. This three-dimensional PET scan shows the metabolic activity within the brain during a hand exercise. The exercise involved moving the fingers of the right hand. The front of the brain is at the left. The most active area appears white; this is the motor cortex in the cerebral cortex where movement is coordinated. Photo credit: Montreal Neurological Institute/McGill University/CNRI/Science Photo Library.

One of the most serious problems with cognitive neuropsychology stems from the difficulty in carrying out group studies. This has led to the increasing use of single-case studies. Such studies are sometimes very revealing. However, they can provide misleading evidence if the patient had specific cognitive deficits prior to brain damage, or if he or she has developed unusual compensatory strategies to cope with the consequences of brain damage. COGNITIVE NEUROSCIENCE Some cognitive psychologists argue that we can understand cognition by relying on observations of peoples performance on cognitive tasks and ignoring the neurophysiological processes occurring within the brain. For example, Baddeley (1997, p. 7) expressed some scepticism about the relevance of neurophysiological processes to the development of psychological theories: A theory giving a successful account of the neurochemical basis of long-term memory would be unlikely to offer an equally elegant and economical account of the psychological characteristics of memory. While it may in principle one day be possible to map one theory onto the other, it will still be useful to have both a psychological and a physiological theoryNeurophysiology and neurochemistry are interesting and important areas, but at present they place relatively few constraints on psychological theories and models of human memory. Why was Baddeley doubtful that neurophysiological evidence could contribute much to psychological understanding? The main reason was that psychologists and neurophysiologists tend to focus on different levels of analysis. In the same way that a carpenter does not need to know that wood consists mainly of

1. INTRODUCTION

19

FIGURE 1.8 The spatial and temporal ranges of some techniques used to study brain functioning. Adapted from Churchland and Sejnowski (1991).

atoms moving around rapidly in space, so it is claimed that cognitive psychologists do not need to know the fine-grain neurophysiological workings of the brain. A different position was advocated by Churchland and Sejnowski (1991, p. 17), who suggested: It would be convenient if we could understand the nature of cognition without understanding the nature of the brain itself. Unfortunately, it is difficult, if not impossible, to theorise effectively on these matters in the absence of neurobiological constraints. The primary reason is that computational space is consummately vast, and there are many conceivable solutions to the problems of how a cognitive operation could be accomplished. Neurobiological data provide essential constraints on computational theories, and they consequently provide an efficient means for narrowing the search space. Equally important, the data are also richly suggestive in hints concerning what might really be going on. In line with these proposals, there are some psychological theories that are being fairly closely constrained by findings in the neurosciences (see Hummel & Holyoak, 1997, and Chapter 15). Neurophysiologists have provided several kinds of valuable information about the brains structure and functioning. In principle, it is possible to establish where in the brain certain cognitive processes occur, and when these processes occur. Such information can allow us to determine the order in which different parts of the brain become active when someone is performing a task. It also allows us to find out whether two tasks involve the same parts of the brain in the same way, or whether there are important differences. As we will see, this can be very important theoretically. The various techniques for studying brain functioning differ in their spatial and temporal resolution (Churchland & Sejnowski, 1991). Some techniques provide information about the single-cell level, whereas others tell us about activity over much larger groups of cells. In similar fashion, some techniques provide information about brain activity on a millisecond-by-millisecond basis (which corresponds to the timescale for thinking), whereas others indicate brain activity only over much longer time periods such as minutes or hours.

20

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

Some of the main techniques will be discussed to give the reader some idea of the weapons available to cognitive neuroscientists. The spatial and temporal resolutions of some of these techniques are shown in Figure 1.8. High spatial and temporal resolutions are advantageous if a very detailed account of brain functioning is required, but low spatial and temporal resolutions can be more useful if a more general view of brain activity is required. Single-unit recording Single-unit recording is a fine-grain technique developed over 40 years ago to permit the study of single neurons. A micro-electrode about one 10,000th of a millimetre in diameter is inserted into the brain of an animal to obtain a record of extracellular potentials. A stereotaxic apparatus is used to fix the animals position, and to provide the researcher with precise information about the location of the electrode in threedimensional space. Single-unit recording is a very sensitive technique, as electrical charges of as little as one-millionth of a volt can be detected. The best known application of this technique was by Hubel and Wiesel (1962, 1979). They used it with cats and monkeys to study the neurophysiology of basic visual processes. Hubel and Wiesel found there were simple and complex cells in the primary visual cortex, but there were many more complex cells. These two types of cells both respond maximally to straight-line stimuli in a particular orientation (see Chapter 4). The findings of Hubel and Wiesel were so clear-cut that they constrained several subsequent theories of visual perception, including that of Marr (1982; see Chapter 2).Evaluation

The single-unit recording technique has the great value that it provides detailed information about brain functioning at the neuronal level, and is thus more fine-grain than other techniques (see Figure 1.8). Another advantage is that information about neuronal activity can be obtained over a very wide range of time periods from small fractions of a second up to several hours or days. A major limitation is that it is an invasive technique, and so would be unpleasant to use with humans. Another limitation is that it can only provide information about activity at the neuronal level, and so other techniques are needed to assess the functioning of larger areas of the cortex. Event-related potentials (ERPs) The electroencephalogram (EEG) is based on recordings of electrical brain activity measured at the surface of the scalp. Very small changes in electrical activity within the brain are picked up by scalp electrodes. These changes can be shown on the screen of a cathode-ray tube by means of an oscilloscope. A key problem with the EEG is that there tends to be so much spontaneous or background brain activity that it obscures the impact of stimulus processing on the EEG recording. A solution to this problem is to present the same stimulus several times. After that, the segment of EEG following each stimulus is extracted and lined up with respect to the time of stimulus onset. These EEG segments are then simply averaged together to produce a single waveform. This method produces eventrelated potentials (ERPs) from EEG recordings, and allows us to distinguish genuine effects of stimulation from background brain activity. ERPs are particularly useful for assessing the timing of certain cognitive processes. For example, some attention theorists have argued that attended and unattended stimuli are processed differently at an early stage of processing, whereas others have claimed that they are both analysed fully in a similar way (see

1. INTRODUCTION

21

Chapter 5). Studies using ERPs have provided good evidence in favour of the former position. For example, Woldorff et al. (1993) found that ERPs were greater to attended than unattended auditory stimuli about 20 50 milliseconds after stimulus onset.Evaluation

ERPs provide more detailed information about the time course of brain activity than do most other techniques, and they have many medical applications (e.g., diagnosis of multiple sclerosis). However, ERPs do not indicate with any precision which regions of the brain are most involved in processing. This is due in part to the fact that the presence of skull and brain tissue distorts the electrical fields emerging from the brain. Furthermore, ERPs are mainly of value when the stimuli are simple and the task involves basic processes (e.g., target detection) occurring at a certain time after stimulus onset. As a result of these constraints (and the necessity of presenting the same stimulus several times) it would not be feasible to study most complex forms of cognition (e.g., problem solving; reasoning) with the use of ERPs. Positron emission tomography (PET) Of all the new methods, the one that has attracted the most media interest is positron emission tomography or the PET scan. The technique is based on the detection of positrons, which are atomic particles emitted by some radioactive substances. Radioactively labelled water is injected into the body, and rapidly gathers in the brains blood vessels. When part of the cortex becomes active, the labelled water moves rapidly to that place. A scanning device next measures the positrons emitted from the radioactive water. A computer then translates this information into pictures of the activity levels of different parts of the brain. It may sound dangerous to inject a radioactive substance into someone. However, only tiny amounts of radioactivity are involved. Raichle (1994b) has described the typical way in which PET has been used by cognitive neuroscientists. It is based on a subtractive logic. Brain activity is assessed during an experimental task, and is also assessed during some control or baseline condition (e.g., before the task is presented). The brain activity during the control condition is then subtracted from that during the experimental task. It is assumed that this allows us to identify those parts of the brain that are active only during the performance of the task. This technique has been used in several studies designed to locate the parts of the brain most involved in episodic memory, which is long-term memory involving conscious recollection of the past (see Chapter 7). There is more activity in the right prefrontal cortex when participants are trying to retrieve episodic memories than when they are trying to retrieve other kinds of memories (see Wheeler, Stuss, & Tulving, 1997, for a review).Evaluation

One of the major advantages of PET is that it has reasonable spatial resolution, in that any active area within the brain can be located to within about 3 or 4 millimetres. It is also a fairly versatile technique, in that it can be used to identify the brain areas involved in a wide range of different cognitive activities. PET has several limitations. First, the temporal resolution is very poor. PET scans indicate the total amount of activity in each region of the brain over a period of 60 seconds or longer, and so cannot reveal the rapid changes in brain activity accompanying most cognitive processes. Second, PET provides only an indirect measure of neural activity. As Anderson, Holliday, Singh, and Harding (1996, p. 423) pointed out, changes in regional cerebral blood flow, reflected by changes in the spatial distribution of intravenously administered positron emitted radioisotopes, are assumed to reflect changes in neural activity. This assumption may be more applicable at early stages of processing. Third, it is an invasive technique, because

22

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

FIGURE 1.9 MRI scan showing a brain tumour. The tumour appears in bright contrast to the surrounding brain tissue. Photo credit: Simon Fraser/Neuroradiology Department, Newcastle General Hospital/Science Photo Library.

participants have to be injected with radioactively labelled water. Fourth, it can be hard to interpret the findings from use of the subtraction technique. For example, it may seem plausible to assume that those parts of the brain active during retrieval of episodic memories but not other kinds of memories are directly involved in episodic memory retrieval. However, the participants may have been more motivated to retrieve such memories than other memories, and so some of the brain activity may reflect the involvement of motivational rather than memory systems. Magnetic resonance imaging (MRI and fMRI) What happens in magnetic resonance imaging (MRI) is that radio waves are used to excite atoms in the brain. This produces magnetic changes which are detected by an 11-ton magnet surrounding the patient. These changes are then interpreted by a computer and turned into a very precise three-dimensional picture. MRI scans (Figure 1.9) can be used to detect very small brain tumours. MRI scans can be obtained from numerous different angles. However, they only tell us about the structure of the brain rather than about its functions. The MRI technology has been applied to the measurement of brain activity to provide functional MRI (fMRI). Neural activity in the brain produces increased blood flow in the active areas, and there is oxygen and glucose within the blood. According to Raichle (1994a, p. 41), the amount of oxygen carried by haemoglobin (the molecule that transports oxygen) affects the magnetic properties of the haemoglobin MRI can detect the functionally induced changes in blood oxygenation in the human brain. The approach based on fMRI provides three-dimensional images of the brain with areas of high activity clearly indicated. It is more useful than PET, because it provides more precise spatial information, and shows changes over shorter periods of time. However, it shares with PET a reliance on the subtraction technique in which brain activity during a control task or situation is subtracted from brain activity during the experimental task. A study showing the usefulness of fMRI was reported by Tootell et al. (1995b). It involves the so-called waterfall illusion, in which lengthy viewing of a stimulus moving in one direction (e.g., a waterfall) is followed immediately by the illusion that stationary objects are moving in the opposite direction. There

1. INTRODUCTION

23

were two key findings. First, the gradual reduction in the size of the waterfall illusion over the first 60 seconds of observing the stationary stimulus was closely paralleled by the reduction in the area of activation observed in the fMRI. Second, most of the brain activity produced by the waterfall illusion was in V5, which is an area of the visual cortex known to be much involved in motion perception (see Chapter 2). Thus, the basic brain processes underlying the waterfall illusion are similar to those underlying normal motion perception.Evaluation

Raichle (1994a, p. 350) argued that fMRI has several advantages over other techniques: The technique has no known biological risk except for the occasional subject who suffers claustrophobia in the scanner (the entire body must be inserted into a relatively narrow tube). MRI provides both anatomical and functional information, which permits an accurate anatomical identification of the regions of activation in each subject. The spatial resolution is quite good, approaching the 12 millimetre range. One limitation with fMRI is that it provides only an indirect measure of neural activity. As Anderson et al. (1996, p. 423) pointed out, With fMRI, neural activity is reflected by changes in the relative concentrations of oxygenated and deoxygenated haemoglobin in the vicinity of the activity. Another limitation is that it has poor temporal resolution of the order of several seconds, so we cannot track the time course of cognitive processes. A final limitation is that it relies on the subtraction technique, and this may not accurately assess brain activity directly involved in the experimental task. Magneto-encephalography (MEG) In recent years, a new technique known as magneto-encephalography or MEG has been developed. It involves using a superconducting quantum interference device (SQUID), which measures the magnetic fields produced by electrical brain activity. The evidence suggests that it can be regarded as a direct measure of cortical neural activity (Anderson et al., 1996, p. 423). It provides very accurate measurement of brain activity, in part because the skull is virtually transparent to magnetic fields. Thus, magnetic fields are little distorted by intervening tissue, which is an advantage over the electrical activity assessed by the EEG. Anderson et al. used MEG in combination with MRI to study the properties of an area of the visual cortex known as V5 (see Chapter 2). They found with MEG that motion-contrast patterns produced large responses from V5, but that V5 did not seem to be responsive to colour. These data, in conjunction with previous findings from PET and fMRI studies, led Anderson et al. (1996, p. 429) to conclude that these findings provide strong support for the hypothesis that a major function of human V5 is the rapid detection of objects moving relative to their background. In addition, Anderson et al. obtained evidence that V5 was active approximately 20 milliseconds after V1 (the primary visual cortex) in response to motion-contrast patterns. This is more valuable information than simply establishing that V1 and V5 are both active during this task, because it helps to clarify the sequence in which different brain areas contribute towards visual processing.Evaluation

MEG possesses several valuable features. First, the magnetic signals reflect neural activity reasonably directly. In contrast, PET and fMRI signals reflect blood flow, which is assumed in turn to reflect neural

24

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

activity. Second, MEG supplies fairly detailed information at the millisecond level about the time course of cognitive processes. This matters because it makes it possible to work out the sequence of activation in different areas of the cortex.

Techniques used by cognitive neuroscientists Method Single-unit recording Information obtained over a wide range of time periods. ERPs Strengths Fine-grain detail. Only neuronal-level information is obtained. Detailed information about the time course of brain activity. Weaknesses Invasive.

PET

Active areas can be located to within 34 mm. Can identify a wide range of cognitive activities.

MRl and fMRI

MEG Gives detailed information about the time course of cognitive processes.

No known biological risk. Obtains accurate anatomical information. fMRl provides good information about timing. Provides a reasonably direct measure of neural activity, Does not give accurate information about brain areas active at a given time.

lack precision in identifying specific areas of the brain. Can only be used to study basic cognitive processes. Cannot reveal rapid changes in brain activity. Provides only an indirect measure of neural activity. Findings f rom a subtraction technique can be hard to interpret. Indirect measure of neural activity. Cannot track the time course of most cognitive processes. Irrelevant sources of magnetism may interfere with measurement.

There are some major technical problems associated with the use of MEG. The magnetic field generated by the brain when thinking is about 100 million times weaker than the Earths magnetic field, and a million times weaker than the magnetic fields around overhead power cables, and it is very hard to prevent irrelevant sources of magnetism from interfering with the measurement of brain activity. Superconductivity requires temperatures close to absolute zero, which means the SQUID has to be immersed in liquid helium at four degrees above the absolute zero of 273C. However, these technical problems have been largely (or entirely) resolved. The major remaining disadvantage is that MEG does not provide structural or anatomical information. As a result, it is necessary to obtain an MRI as well as MEG data in order to locate the active brain areas. Section summary All the techniques used by cognitive neuro-scientists possess strengths and weaknesses. Thus, it is often desirable to use a number of different techniques to study any given aspect of human cognition. If similar

1. INTRODUCTION

25

findings are obtained from two techniques, this is known as converging evidence. Such evidence is of special value, because it suggests that the techniques are not providing distorted information. For example, studies using PET, fMRI, and MEG (e.g., Anderson et al., 1996; Tootell et al., 1995a, b) all indicate clearly that area V5 is much involved in motion perception. It can also be of value to use two techniques differing in their particular strengths. For example, the ERP technique has good temporal resolution but poor spatial resolution, whereas the opposite is the case with fMRI. Their combined use offers the prospect of discovering the detailed time course and location of the processes involved in a cognitive task. The techniques used within cognitive neuro-science are most useful when applied to areas of the brain that are organised in functionally discrete ways (S.Anderson, personal communication). For example, as we have seen, there is evidence that area V5 forms such an area for motion perception. It is considerably less clear that higher-order cognitive functions are organised in a similarly neat and tidy fashion. As a result, the various techniques discussed in this section may prove less informative when applied to such functions. You may have got the impression that cognitive neuroscience consists mainly of various techniques for studying brain functioning. However, there is more than that to cognitive neuroscience. As Rugg (1997, p. 5) pointed out, The distinctiveness [of cognitive neuroscience] arises from a lack of commitment to a single level of explanation, and the resulting tendency for explanatory models to combine functional and physiological concepts. Various examples of this explanatory approach are considered during the course of this book. OUTLINE OF THIS BOOK One problem with writing a textbook of cognitive psychology is that virtually all the processes and structures of the cognitive system are interdependent. Consider, for example, the case of a student reading a book to prepare for an examination. The student is learning, but there are several other processes going on as well. Visual perception is involved in the intake of information from the printed page, and there is attention to the content of the book (although attention may be captured by irrelevant stimuli). In order for the student to profit from the book, he or she must possess considerable language skills, and must also have rich knowledge representations that are relevant to the material in the book. There may be an element of problem solving in the students attempts to relate what is in the book to the possibly conflicting information he or she has learned elsewhere. Furthermore, what the student learns will depend on his or her emotional state. Finally, the acid test of whether the learning has been effective and has produced long-term memory comes during the examination itself, when the material contained in the book must be retrieved. The words italicised in the previous paragraph indicate some of the main ingredients of human cognition, and form the basis of our coverage of cognitive psychology. In view of the interdependent functioning of all aspects of the cognitive system, there is an emphasis in this book on the ways in which each process (e.g., perception) depends on other processes and structures (e.g., attention; long-term memory; stored representations). This should aid the task of making sense of the complexities of the human cognitive system.

26

COGNITIVE PSYCHOLOGY: A STUDENTS HANDBOOK

CHAPTER SUMMARY

Cognitive psychology as a science. Cognitive psychology is unified by a common approach based on an analogy between the mind and the computer. This information-processing approach views the mind as a general-purpose, symbol-processing system of limited capacity. There are four main types of cognitive psychologists: experimental cognitive psychologists; cognitive scientists; cognitive neuropsychologists; and cognitive neuroscientists, who use various techniques to study brain functioning. Cognitive science. Cognitive scientists focus on computational models, in which theoretical assumptions have to be made explicit. These models are expressed in computer programs, which should produce the same outputs as people when given the same inputs. Three of the main types of computational model are semantic networks, production systems, and connectionist networks. Semantic networks consist of concepts, which are linked by various relations (e.g., is-similar-to). They are useful for modelling the structure of peoples conceptual knowledge. Production systems are made up of productions in the form of IFTHEN rules. Connectionist networks differ from previous approaches in that they can learn from experience, for example, through the backward propagation of errors. Such networks often have several structures or layers (e.g., input units; intermediate or hidden units; and output units). Concepts are stored in a distributed manner. Cognitive neuropsychology. Cognitive neuropsychologists assume that the cognitive system is modular, that there is isomorphism between the organisation of the physical brain and the mind, and that the study of brain-damaged patients can tell us much about normal human cognition. The notion of syndromes has lost popularity, because syndromes typically exaggerate the similarity

of the Symptoms shown by patients having allegedly the same condition. It can be hard to interpret the findings from brain-damaged patients for various reasons: patients may develop compensatory strategies after brain damage; the brain damage may affect several modules; patients may have had specific cognitive impairments before the brain damage. Cognitive neuroscience. Cognitive neuroscientists use various techniques for studying the brain, with these techniques varying to their spatial and temporal resolution. Important techniques include single-unit recording, event-related potentials, positron emission tomography, functional magnetic resonance imaging, and magneto-encephalography. Critics argue that neurophysiological findings am often at a different level of analysis from the one of most value to cognitive psychologists. In addition, such findings often fail to place significant constraints on psychological theorising.

1. INTRODUCTION

27

FURTHER READING Ellis, R., & Humphreys, G. (1999). Connectionist psychology; A text with readings. Hove, UK: Psychology Press. Connectionism has become very influential within cognitive science, and this approach is discussed very thoroughly in this book. Gazzaniga, M.S., Ivry, R.B., & Mangun, G.R. (1998). Cognitive neuroscience: The biology of the mind. New York: W.W.Norton & Co. This is a comprehensive book in which the relevance of the cognitive neuroscience approach to the major areas of cognitive psychology is considered in detail. McLeod, P., Plunkett, K., & Rolls, E.T. (1998). Introduction to