Neuropsychological Impairments of Short-Term Memory

Neuropsychological impairments of short-term memory Neuropsychological impairments of short-term memory Edited by G...

Author: Giuseppe Vallar | Tim Shallice

44 downloads 706 Views 14MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Neuropsychological impairments of short-term memory

Neuropsychological impairments of short-term memory

Edited by

Giuseppe Vallar and Tim Shallice

The right of the University of Cambridge to print and sell all manner of books was granted by Henry VIII in 1534. The University has printed and published continuously since 1584.

Cambridge University Press Cambridge New York Port Chester Melbourne Sydney

CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521370882 © Cambridge University Press 1990 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 1990 This digitally printed version 2007 A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication data Neuropsychological impairments of short-term memory / edited by Giuseppe Vallar and Tim Shallice. p. cm. Based on papers presented at a conference held in Como, Italy, in Sept. 1987, supported by the Consiglio nazionale delle ricerche. Includes bibliographical references. ISBN 0-521-37088-4 1. Memory, Disorders of — Congresses. 2. Short-term memory— Congresses. 3. Clinical neuropsychology - Congresses. I. Vallar, Giuseppe. II. Shallice, Tim. III. Consiglio nazionale delle ricerche (Italy) [DNLM: 1. Memory Disorders - physiopathology - congresses. 2. Memory, Short-Term — physiology — congresses. 3. Psycholinguistics - congresses. WM 173.7 N4935 1987] RC394.M46N48 1990 616.8'4-dc20 DNLM/DLC for Library of Congress 90-25174 CIP ISBN 978-0-521-37088-2 hardback ISBN 978-0-521-04275-8 paperback

Contents

List of contributors Acknowledgments General introduction

I

THE FUNCTIONAL ARCHITECTURE OF AUDITORY-VERBAL (PHONOLOGICAL) SHORT-TERM MEMORY AND ITS NEURAL CORRELATES 1

The impairment of auditory-verbal short-term storage

page ix xiii 1

7 11

TIM SHALLICE AND GIUSEPPE VALLAR

2

The development of the concept of working memory: implications and contributions of neuropsychology

54

ALAN D. BADDELEY

3

Multiple phonological representations and verbal short-term memory

74

FRANCES J. FRIEDRICH

4

Electrophysiological measures of short-term memory ARNOLD STARR, GEOFFREY BARRETT, HILLEL PRATT, HENRY J. MICHALEWSKI, AND JULIE V. PATTERSON

94

vi II

Contents PHONOLOGICAL SHORT-TERM MEMORY AND OTHER LEVELS OF INFORMATION PROCESSING: STUDIES IN BRAIN-DAMAGED PATIENTS WITH DEFECTIVE PHONOLOGICAL MEMORY 111 5

Auditory and lexical information sources in immediate recall: evidence from a patient with deficit to the phonological short-term store

115

RITA SLOAN BERNDT AND CHARLOTTE C. MITCHUM

6

Neuropsychological evidence for lexical involvement in short-term memory

145

ELEANOR M. SAFFRAN AND NADINE MARTIN

7

Auditory-verbal span of apprehension: a phenomenon in search of a function?

167

ROSALEEN A. McCARTHY A N D ELIZABETH K. WARRINGTON

8

Short-term retention without short-term memory

187

BRIAN BUTTERWORTH, TIM SHALLICE, AND FRANCES L. WATSON

III

SHORT-TERM MEMORY STUDIES IN DIFFERENT POPULATIONS (CHILDREN, ELDERLY, AMNESICS) AND OF DIFFERENT SHORT-TERM MEMORY SYSTEMS 9

Developmental fractionation of working memory

215 221

GRAHAM J. HITCH

10

Adult age differences in working memory

247

FERGUS I. M. CRAIK, ROBIN G. MORRIS, AND MARY L. GICK

11

Lipreading, neuropsychology, and immediate memory

268

RUTH CAMPBELL

12

Memory without rehearsal

287

DAVID HOWARD AND SUE FRANKLIN

13

The extended present: evidence from time estimation by amnesics and normals MARCEL KINSBOURNE AND ROBERT E. HICKS

319

Contents IV PHONOLOGICAL SHORT-TERM MEMORY AND SENTENCE COMPREHENSION 14

Short-term memory and language comprehension: a critical review of the neuropsychological literature

vii 331

337

DAVID CAPLAN AND GLORIA S. WATERS

15

Neuropsychological evidence on the role of short-term memory in sentence processing

390

RANDI C. MARTIN

16

Short-term memory impairment and sentence processing: a case study

428

ELEANOR M. SAFFRAN AND NADINE MARTIN

17

Phonological processing and sentence comprehension: a neuropsychological case study

448

GIUSEPPE VALLAR, ANNA BASSO, AND GABRIELLA BOTTINI

18

Working memory and comprehension of spoken sentences: investigations of children with reading disorder

477

STEPHEN CRAIN, DONALD SHANKWEILER, PAUL MACARUSO, AND EVA BAR-SHALOM

Name index Subject index

509 517

Contributors

Alan D. Baddeley

Ruth Campbell

MRC Applied Psychology Unit

Department of Psychology

Cambridge

Goldsmiths College University of London

Eva Bar-Shalom Department of Linguistics University of Connecticut

David Caplan Neuropsychology Unit Massachusetts General Hospital

Geoffrey Barrett

Boston

The National Hospitals for Nervous Diseases London

Fergus I. M. Craik Department of Psychology University of Toronto

Anna Basso Istituto di Clinica Neurologica University of Milan

Stephen Crain Department of Linguistics University of Connecticut

Rita Sloan Berndt Department of Neurology University of Maryland School of Medicine Gabriella Bottini

Sue Franklin Department of Psychology University of York Frances J. Friedrich

Istituto di Clinica Neurologica


University of Milan

The University of Utah

Brian Butterworth

Mary L. Gick



University College London

University of Toronto

Contributors Robert E. Hicks University of North Carolina at Chapel Hill Graham J. Hitch Department of Psychology University of Manchester David Howard Department of Psychology Birkbeck College London Marcel Kinsbourne Eunice Kennedy Schriver Center Waltham, Massachusetts Paul Macaruso Department of Linguistics University of Connecticut Rosaleen A. McCarthy Department of Experimental Psychology University of Cambridge

Charlotte C. Mitchum Department of Neurology University of Maryland School of Medicine Robin G. Morris Department of Psychology University of Toronto Julie V. Patterson Department of Neurology University of California at Irvine Hillel Pratt Technion Haifa Eleanor M. Saffran Department of Neurology Temple University School of Medicine Tim Shallice MRC Applied Psychology Unit Cambridge

Nadine Martin Department of Neurology Temple University School of Medicine

Donald Shankweiler Department of Linguistics University of Connecticut

Randi C. Martin Department of Psychology Rice University

Arnold Starr Department of Neurology University of California at Irvine

Henry J. Michalewski Department of Neurology University of California at Irvine

Giuseppe Vallar Istituto di Clinica Neurologica University of Milan

Contributors Elizabeth K. Warrington Department of Psychology The National Hospitals for Nervous Diseases London Gloria S. Waters School of Human Communication Disorders McGill University Montreal

xi Frances L. Watson MRC Applied Psychology Unit Cambridge

Acknowledgments

This book comprises chapters based on papers presented at a conference held in Villa Olmo, Como, Italy, September 14-16, 1987. We wish to acknowledge the support from the Consiglio Nazionale delle Ricerche, which made the conference possible. We would also like to thank the GLAXO S.P.A., Verona, Italy, and in particular Dr. Giuseppe Coppola, for their support. The participants at the conference spent three productive days in the Duke's Hall of Villa Olmo and in the Italian-style gardens surrounding the villa. We are grateful to Prof. Giulio Casati and to the staff of the Centro di Cultura Scientifica "Alessandro Volta," and in particular to Dr. Chiara DeSantis, Signora Donatella Marchegiano, and Dr. Federico Canobbio-Codelli for their assistance in the organization of the meeting. We would like to thank Katharita Lamoza for her expert overseeing of the production of the book.

Xlll

General introduction

In recent years the single-case approach has grown greatly in popularity in neuropsychology. Some workers have even argued that no pretheoretical generalizations across patients can be justified (e.g., Caramazza, 1986), for, they argue, one cannot know that any two patients have functionally equivalent lesions. Yet it is equally argued by people who hold these general positions that our theoretical understanding of the neurological organization of cognitive function is rudimentary, so theoretically driven grouping of patients is to be avoided too (see Ellis, 1987). This is an unsatisfactory state of affairs. Any science needs a data base that has some depth. A tentative understanding of the range of empirical phenomena that occur in a domain should be available. How robust the phenomena are needs to be roughly known. A practical way out of the dilemma is to take putative syndromes - patients with a common cluster of difficulties that can plausibly be attributed to a common functional cause - and to present multiple mixed practical-theoretical investigations of a domain by different investigators. Even if the syndrome proves not to be a single functional entity, the studies should provide a solid basis for future research. The overlapping empirical observations on different patients will provide an adequate basis for future theoretical analyses. Competing theoretical perspectives will sharpen the perspective for future empirical investigations. Two pioneer books on the acquired dyslexias - Deep Dyslexia (Coltheart, Patterson, & Marshall, 1980) and Surface Dyslexia (Patterson, Marshall, & Coltheart, 1985) - illustrate the value of the approach. Neither syndrome has remained solidly accepted as a single functional entity, but the value of each book in defining its field is undoubted. The present book adopts a similar approach for a different syndrome - the "short-term memory (STM) syndrome" - a specific impairment in the performance of span tasks. Span - the repeating back of a string of well-learned verbal units (digits, letters, words) — has long been a task used by experimental psychologists. It has also been a frequent component of the test batteries of psychometricians (e.g., it is part of the

2


Wechsler Adult Intelligence Scale, WAIS). Through the speculations of Hebb (1949), Miller (1956), and Broadbent (1958) a much greater theoretical interest developed in the 1950s in the short-term storage of information, and this grew still further in the 1960s with the development of specific models of short-term storage such as those of Waugh and Norman (1965) and of Atkinson and Shiffrin (1968). As a consequence the properties of span and related tasks became much more intensively investigated. During this decade a neuropsychological syndrome of a specific deficit on span tasks was isolated with other aspects of cognitive and language function including word comprehension and production, apparently intact. The original investigators (Warrington & Shallice, 1969) interpreted the disorder as a specific deficit in an auditory-verbal short-term store using a model related - but not identical - to those advocated by experimental psychologists of the period. Later investigators (e.g., Saffran & Marin, 1975; Caramazza, Basili, Koller, & Berndt, 1981; Vallar & Baddeley, 1984a) interpreted the disorder in a related fashion, namely, in terms of an impairment to a phonological input store that was viewed as part of the language comprehension system. The properties of the hypothetical store fit well with what would be expected from damage to an input phonological buffer of the working memory or multiple store models that were developed in the 1970s. By the 1980s the disorder was seen as one of the syndromes that best realized the blending of neuropsychological and normal experimental investigations that characterize cognitive neuropsychology. The general concordance of views did not last. At two cognitive neuropsychology meetings - one held in Venice in September 1985 and the other at Bressanone, Italy, in January 1986 - a day was devoted to presentations on the STM syndrome and heated debates occurred. It became clear that the syndrome, which had seen only a small trickle of studies since the late 1960s, was being investigated by quite a large number of workers in different countries (the United Kingdom, Italy, the United States, and Canada). It was also very evident that those involved were strongly and almost bitterly divided over a number of issues. The view put forward by the protagonists of the syndrome at these meetings (e.g., Baddeley, Shallice, and Vallar) was that the classical view was essentially valid and that the syndrome was important because of the symbiotic relation that analyses of the syndrome had with the development of models of normal short-term memory functions. The multiple stores or working memory frameworks had arisen in the 1970s partially as a result of work on the STM syndrome, and in turn the frameworks were able to provide a detailed account of its properties. The correspondence provided one of the strongest pieces of evidence for the validity of the cognitive neuropsychology approach, especially when opinions were beginning to become divided about two of the other pioneer syndromes, deep and surface dyslexia (see Coltheart, Patterson, & Marshall, 1980; Patterson, Coltheart, & Marshall, 1985). Moreover it was argued that unlike those syndromes, the present disorder has a consistent and relatively


3

circumscribed localization; thus from a wider neuroscience perspective it was a plausible candidate for a pure syndrome. These views were challenged on a number of different levels. Most heat was generated over what the evolutionary function of a phonological buffer might be. It had been standard to view it as a component of the language comprehension process. All the earlier patients who had been described with a specific impairment of auditory-verbal span had had some comprehension difficulties. This did not manifest itself as a clinically obvious comprehension disorder, since the patients had no difficulty with most sentences that had simple syntactic structures. The most popular suggestions have been that the store has a backup function (Shallice & Warrington, 1970; Saffran & Marin, 1975) or is important for the comprehension of long and syntactically complex sentences (Vallar & Baddeley, 1984b). However, by the mid-1980s problems with these positions were appearing, and they were challenged at the two meetings by a number of participants (e.g., Butterworth, Caplan, and Howard). The critics used several lines of evidence. Caplan, Vanier, and Baker (1986) had pointed out that although the STM patients all had language comprehension problems, the nature of the problem seemed to differ among them. They therefore questioned how critical the STM impairment was as the cause of the comprehension difficulties. At roughly the same time Butterworth, Campbell, and Howard (1986) had described a subject, RE, with a mild developmental disorder of STM who had no apparent difficulties in language comprehension; this was held to be an unlikely combination if the evolutionary function of STM was to aid language comprehension. In addition, at the Venice meeting Saffran and Martin had reported that a fairly severe STM patient had no difficulty in judging whether sentences were grammatical or not even when the number of words separating the syntactically critical elements of the sentence was considerably more than his span. Thus it would seem that STM cannot be necessary for appropriate syntactic parsing to take place. A second line of argument was forcibly presented by Caplan at the Bressanone meeting. If, he argued, it were granted that the short-term storage processes impaired in the patients had a language comprehension function, then the appropriate conceptual framework within which to describe it should be derived from linguistic or psycholinguistic theorizing rather than memory theory. For him, a concept like the "phonological buffer" is derived from an inappropriate conceptual framework, namely, memory. It was historically somewhat ironic that this criticism was made, as a major reason in the early 1970s as to why the multiple stores approach had replaced the socalled modal model of the previous decade was that theorists like Morton (1970), Jarvella (1971), and Glanzer (1972) had argued that short-term storage of auditory-verbal information must be situated within the context of a model of language processing rather than within the purely memory framework of models like those of Waugh and Norman (1965) and Atkinson and Shiffrin (1968). The same

4


criticism was now being posed against an approach that was a descendant of the theories they put forward. A final reason that criticism of the standard position on the STM patients began to mount is that the box-and-arrow models of early information-processing theorists were becoming less popular. "Interactive activation" (McClelland & Rumelhart, 1981) models and "connectionist" (Hinton & Anderson, 1981) models were beginning to be preferred. A natural consequence of such models, although not a necessary one, is to view short-term storage as arising from continuing activation in the same structures that are used for processing. In line with this position, Allport (1984) had argued that despite their seemingly intact word perception the STM patients have subtle phonological processing problems and their short-term storage difficulties are a different theoretical perspectives. The book is divided into four parts. In part I the short-term memory aspects of the syndrome are reviewed and related to the literature on normal short-term memory ( om two contrasting perspectives (presented in different chapters) - that of the working memory model and that of the interactive activation approach. In part II a number of beforehand. A three-day workshop on the syndrome was organized, and it took place in Como, Italy, in September 1987. This book is based on the papers presented at the meeting. It adopts the approach discussed in the initial paragraphs, namely, to present studies on closely related topics and similar patients by a variety of authors with different theoretical perspectives. The book is divided into four parts. In part I the short-term memory aspects of the syndrome are reviewed and related to the literature on normal short-term memory from two contrasting perspectives (presented in different chapters) - that of the working memory model and that of the interactive activation approach. In part II a number of empirical investigations of the syndrome are described. These are mainly concerned with the relative contribution that disorders at different levels in the auditory language perception process play in the genesis of the STM syndrome. In part III the focus is widened. The approach of examining how another type of subject contrasts with the normal adult in the present task domain - short-term memory - is extended to other groups of subjects, ranging from the young child to the amnesic patient. The aim of making these other contrasts is to assess whether the way these subjects perform on short-term memory tasks is or is not well captured by conceptual approaches - in particular working memory/multiple store models - used to explain the STM patient-normal adult contrast. The wider the fit the more robust are the models. The final part presents papers that take a number of positions on the much debated relation between short-term memory processes and language comprehension. The divisions into parts should not be thought of as an absolute one. The themes described in this introduction are alluded to in many of the chapters. Certain contributors also deal extensively with topics in more than one of these four parts. Where this occurs, cross-referencing is made in the introductions to the relevant parts.


5

References Allport, D. A. (1984). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 313-325) Hillsdale, NJ: Erlbaum. Atkinson, R. G, & Shirrrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89-195). New York: Academic Press. Broadbent, D. E. (1958), Perception and communication. London: Pergamon. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A, 705-738. Caplan, D., Vanier, M., & Baker, C (1986). A case study of reproduction conduction aphasia: 2. Sentence comprehension. Cognitive Neuropsychology, 3, 129—146. Caramazza, A. (1986). On drawing inferences about the structure of normal cognitive systems from the analysis of patterns of impaired performance: The case for single-patient studies. Brain and Cognition, 5, 41—66. Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. J. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271. Coltheart, M., Patterson, K., & Marshall, J. C (Eds.) (1980). Deep dyslexia. London: Routledge & Kegan Paul. Ellis, A. W. (1987). Intimations of modularity, or, the modelarity of mind: Doing cognitive neuropsychology without syndromes. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language (pp. 397-408). London: Erlbaum. Glanzer, M. (1972). Storage mechanisms in recall. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 5, pp. 129-193). New York: Academic Press. Hebb, D. O. (1949). The organization of behavior. New York: Wiley. Hinton, G. E., & Anderson, J. A. (Eds.) (1981). Parallel models of associative memory. Hillsdale, NJ: Erlbaum. Jarvella, R. J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Behavior, 10, 409-416. McClelland, J. L., & Rumelhart, D. E. (1951). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375-407. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits of our capacity for processing information. Psychological Review, 63, 81—97. Morton, J. (1970). A functional model of memory. In D. A. Norman (Ed.), Models of human memory. New York: Academic Press. Peterson, K. E., Marshall, J. C, & Coltheart, M. (Eds.) (1985). Surface dyslexia. London: Erlbaum. Saffran. E.M., & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with a deficient auditory short-term memory. Brain and Language, 2, 420-433. Shallice, T., & Warrington, E. K. (1970). Independent functioning of the verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261-273. Vallar, G., & Baddeley, A. D. (1984a). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Baddeley, A. D. (1984b). Phonological short-term store, phonological processing and sentence comprehension. A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory verbal short-term memory. Brain, 92, 885-896. Waugh, N. C, & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89-104.

Part I. The functional architecture of auditory-verbal (phonological) short-term memory and its neural correlates

The first part of the book comprises four chapters that discuss two main issues: (a) the possible functional architecture of the system(s) involved in the short-term retention of verbal material (Shallice & Vallar, chapter 1; Baddeley, chapter 2; Friedrich, chapter 3); and (b) some neural correlates (chapter 1; Starr, Barrett Pratt, Michalewski, & Patterson, chapter 4). Two main approaches are suggested on the functional structure of verbal short-term memory. Shallice and Vallar and Baddeley take the view of verbal short-term memory as a multicomponent system, which includes a number of distinct processing and storage subcomponents. Shallice and Vallar review the neuropsychological and, more briefly, the normal evidence for this approach. They consider a number of specific phenomena, such as auditory-verbal memory span and the recency effect in free and serial recall, noting a convergence in the findings obtained from normal subjects and patients. They note that a number of aspects of the original observations (Warrington & Shallice, 1969) of a selective impairment on span tasks have been replicated in a considerable number of cases, which they review. The evidence from normal subjects supports the position that short-term memory effects from paradigms such as span reflect the operation of a buffer store where information is coded phonologically. The results obtained from the patients are very similar to what would be expected if such a component were severely damaged. On the basis of this type of argument, Shallice and Vallar suggest that the selective deficit of auditory-verbal span may be conceived as a functional syndrome, which may be traced back to the selective impairment of a specific component of verbal short-term memory. Baddeley also discusses both "normal" and neuropsychological evidence, but in his case the greater emphasis is on the results from normal subjects. He places the development of models of verbal short-term memory in their historical context and highlights a key difference that distinguishes the memory models most influential in the late 1960s (Waugh & Norman, 1965; Atkinson & Shiffrin, 1968) from the multicomponent views developed in the following years. While the earlier type of model contains a single short-term store, which is not modality-specific, the multicomponent approach

8

Part I

to short-term memory fractionates this system into a number of subcomponents (e.g., auditory-verbal and visual stores: see also Sperling, 1967). Baddeley discusses in detail the more recent developments of Baddeley and Hitch's (1974) working memory model and its possible involvement in a number of cognitive activities, such as aspects of sentence comprehension, learning to read, fluent reading, and long-term phonological learning. The multicomponent approach taken in the chapters by Baddeley and Shallice and Vallar has been widely used in the analysis of the functional deficit of patients with a defective auditory-verbal immediate memory span (e.g., Howard & Franklin, chapter 12; Vallar, Basso, & Bottini, chapter 17). It is also used by Caplan and Waters (chapter 14), in their review paper concerning the relationships between verbal short-term memory deficits and speech comprehension disorders; they classify span-impaired patients according to a taxonomy based on a multicomponent view of short-term memory. An alternative view of the possible functional architecture of short-term memory systems is discussed by Friedrich in chapter 3. She distinguishes different types of phonological representations (e.g., auditory and articulatory), which, in interaction with visual and semantic representations, are involved in immediate retention. Friedrich emphasizes the variety of connections and interactions between the different types of representations. She argues that the functional characteristics of immediate memory performance (e.g., phonological similarity and irrelevant speech effects) may reflect differences in the pattern of interaction among representations. Friedrich's view is reminiscent of Craik and Lockhart's (1972) levels of processing approach and is related to the more recent "interactive activation" and "connectionist" models of cognitive function. A similar position is adopted by other contributors. Berndt and Mitchum (chapter 5) discuss their neuropsychological data within the framework of a multicapacity system consisting of transient representations of various codes generated during language-processing tasks. Saffran and Martin (chapter 6) propose a multilevel interactive model, comprising phonological, lexical, and semantic representations, which contribute to short-term memory performance. Campbell (chapter 11) suggests that immediate memory performance is characterized by the reciprocal interaction of phonological input and output devices, where speech events are represented as a stable pattern of distributed activity across a subset of units in each system. This "interactive activation approach" is used by these contributors to account for the normal and neuropsychological findings, which Shallice and Vallar and Baddeley discuss within the framework of a multistore model. It is worth noting, however, that the "interactive" approach, although put forward as an alternative to the multistore models (e.g. Saffran, chapter 6) and attractive because of the many other domains to which it relates, does not, at least at present, have a better explanatory value; with regard to short-term memory patients, it does not, for instance, account for

Architecture of auditory-verbal short-term memory

9

experimental data inconsistent with the multistore approach. Indeed, one specific feature of the multistore approach is the distinction between processing and storage components, which is unspecified in the type of multilevel interactive models discussed in the previously mentioned chapters. The existence of patients with severe defects of auditory-verbal span who nevertheless have preserved speech perception (see review in Shallice and Vallar, chapter 1) is consistent with a multistore view of short-term memory, but it is less clear how an interactive activation approach would deal with this dissociation. Finally, a main feature of Friedrich's model, namely, the interaction among different representations, can be also present in multistore models (see, e.g., Shallice & Vallar, chapter 1, Figure 1.2; Butterworth, Shallice, & Watson, chapter 8). Indeed, it may well turn out that the two types of frameworks reflect different aspects of the total system rather than being direct competitors. The neural correlates of auditory-verbal short-term memory deficits are discussed in the chapters by Shallice and Vallar and Starr et al. On the basis of the traditional anatomoclinical correlation method (see Vallar & Perani, 1987, for a discussion), Shallice and Vallar argue that a lesion of a specific region of the left hemisphere, the inferior parietal lobule, is the neurological correlate of the functional syndrome of defective auditory-verbal short-term memory. This conclusion is primarily supported by evidence from positive cases, namely, individual patients with a selective deficit of auditory-verbal memory span. However, both group studies of patients selected on the basis of neurological criteria (i.e., the site of the lesion) and data from left hemisphere-damaged patients with a normal span provide additional support for this view. The neurological evidence then suggests that a specific region of the brain is involved in the immediate retention of phonologically coded verbal material. This parallels the behavioral data from both patients and normal subjects, which, at a functional level, suggest the existence of a discrete phonological short-term storage component. The investigations reported by Starr at al. (chapter 4) use a dynamic methodology, event-related potentials, which can serve as an indicator of neural processes during a variety of cognitive activities. In contrast to the review by Shallice and Vallar, their study is not concerned with correlations at a macroanatomical level, but attempts instead to discover neurophysiological correlates of the operation of functional components, such as visual and phonological processing and storage. Starr and coworkers recorded event-related potentials in immediate memory tasks in patients with selective deficits of auditory—verbal span. Their observation of abnormal potentials in the case of retrieved auditory-verbal stimuli, associated with a normal electrophysiological activity with visual and auditory nonverbal items, suggests a neural deficit confined to the retention of auditory—verbal material. More specifically, their patient's abnormality involves the late potentials, but not the early components. This may be taken as an indication that auditory-verbal processing is spared. This finding, which is

10

Part I

relevant to the comparative evaluation of level of processing versus multistore models of short-term memory, clearly illustrates the contribution that this methodology can offer to the understanding of the structure of the system. An additional example is provided by the observation of a neurophysiological correlate of a specific short-term memory phenomenon, the recency effect in immediate memory.

References Atkinson, R. C, & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence and J. Taylor Spence (Eds.) The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89-195). New York: Academic Press. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 8, pp. 47-89). New York: Academic Press. Craik, F. I. M , & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. journal of Verbal Learning and Verbal Behavior. 11. 671-684. Sperling, G. (1967). Successive approximations to a model for short-term memory. Ada Psychologica, 27. 285-292. Vallar, G., & Perani, D. (1987). The anatomy of spatial neglect in humans. In M. Jeannerod (Ed.), Neurophysiological and neuropsychological aspects of spatial neglect (pp. 235-258). Amsterdam: North Holland. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory-verbal short-term memory. Brain. 92. 885-896. Waugh, N. C, & Norman, D. A. (1965). Primary memory. Psychological Review, 72. 89-104.

1. The impairment of auditory-verbal short-term storage TIM SHALLICE AND GIUSEPPE VALLAR

1.1. Introduction Since the 1940s many theorists have suggested that information might be stored in the cognitive system in a different way over short time periods than for longer time periods. A number of possible mechanisms have been proposed - special or generalpurpose buffers, the continuing but temporary activation of the structures that have just processed an input, and the formation of temporary associations or temporary changes in association strength. Theoretical arguments for such mechanisms have been produced from a range of scientific fields from physiological psychology (e.g., Hebb, 1949), information-processing psychology (Broadbent, 1958), and symbol-processing artificial intelligence (Newell & Simon, 1972) through to connectionist neuroscience (e.g., Crick, 1984; Hinton & Plaut, 1987). Empirical support has come from a narrower set of approaches. Most of the relevant findings have come from experimental psychology (e.g., Waugh & Norman, 1965; Atkinson & Shiffrin, 1968; Glanzer, 1972; Baddeley & Hitch, 1974), but these have been subject to many criticisms (see, e.g., Craik & Lockhart, 1972; Crowder, 1982). From a neuropsychological perspective, if a short-term memory trace were being carried by certain of the aforementioned mechanisms, a selective deficit to that mechanism would be realized in a selective impairment in behaviour. This is clearly true for the most favoured possibility - damage to a specific short-term memory buffer — given that the buffer was not involved in all cognitive operations. It would also apply for a specific impairment of one of two association-weight-changing mechanisms. For instance, Hinton and Plaut (1987) have simulated a connectionist network using two types of weights - "slow" weights (which change slowly and carry the long-term learning) and "fast" weights (which change rapidly but decay fairly quickly to zero). The existence of fast weights allows (a) rapid temporary learning, (b) the creation of temporary binding between features, which can be thought of as a possible model of the We would like to thank David Caplan and Graham Hitch for very helpful comments on an earlier draft of this paper.

11

12

Shallice and Vallar

effects of attention (Feldman, 1982), (c) properly recursive processing, and (d) reduced interference when new learning occurs. How the slow weights change is influenced by the state of the fast weights. However, the mechanism by which the former ones change does not depend on the existence of fast weights, so if the fast weights are set permanently to zero, learning and retrieval can still take place. Most of the system functions will still be normal. Loss of the fast weights would, however, lead to problems for various types of operations such as recursion (see McClelland & Kanomoto, 1986) and increase interference effects in new learning. The one exception to the prediction of the existence of selective short-term memory impairments would appear to be when a short-term trace is carried, as suggested by Hebb (1949), in temporary activation of the processing structures and permanent changes require that period of temporary activation. In this case, damage that led to a more rapid loss of activation would also produce a difficulty in long-term learning. Any selective disorder of short-term memory (STM) presents a problem for such a model (Shallice & Warrington, 1970). In this chapter we will review published work on the first neuropsychological syndrome to be analysed as a selective disorder of short-term memory — that which has been characterized as a selective impairment either of auditory—verbal short-term storage or of the phonological buffer. There has been much criticism in recent years of any attempt to generalize across the findings obtained with separate patients for theoretical purposes (see Caramazza, 1986, and also Morton & Patterson, 1980; Ellis, 1987). However, to abandon totally the concept of "syndrome" in turn has severe disadvantages; the possibility of replication is given up as is the prospect of any effective link to other domains of explanation, for example, the neuroscience one. The approach we will adopt is to consider all patients who exhibit a particular basic pattern of performance. The basic pattern is in part historically motivated; it is that which was originally described (Warrington & Shallice, 1969) and has since been observed in a number of other patients. However, it consists of a highly selective disorder and is therefore a plausible candidate for a "pure" syndrome (see Shallice, 1988, for discussion). It also corresponds to what would result from a specific deficit of an auditory-verbal short-term store as characterized by a simple short-term memory model. We will use it to select the patients to be considered in this chapter. The performance of these patients will then be assessed from the perspective of more elaborate models. Consistency of performance across patients is treated as replication. The theoretical problem of a failure to find consistency will be discussed where it occurs. The overall validity of the syndrome approach will be considered again in section 1.5.

Impairment of auditory-verbal short-term storage

13

1.1.1. The basic pattern of performance The original observations were of severely impaired auditory-verbal span performance in the context of the relative sparing of other language and cognitive functions in a patient KF (Warrington & Shallice, 1969,1972; Shallice & Warrington, 1970). The basic pattern of performance had four components: 1. 2. 3. 4.

Selective deficit of span; A comparable performance level for all strings of unconnected auditory-verbal items; Evidence that the span deficit does not arise from impaired speech production; Intact auditory word perception.

This pattern of performance is of theoretical interest as it corresponds to what would result from a selective impairment to a short-term store (STS), to specifically short-term associations, or to a short-term weight-changing mechanism. The first property should arise given that the relevant mechanism is not generally required for language and/or cognitive operations. The second would follow if the system were used for all types of verbal material and also if there were not other systems that some but not all verbal material could utilize. The final two indicate that span is not impaired for the two most obvious potential causes other than a memory disorder, namely, from damage to speech perception or production. Since this pattern of performance was observed in KF a number of other patients have been described and held to have a related disorder. Their performance with respect to these four criteria is shown in Table 1.1 (a). For certain patients (EDE, RAN, NHA) information is not available to calculate span for material other than digits. However, in all three cases it is clear that span for material other than digits was impaired. For instance, both RAN and NHA repeated correctly less than half the words in three word triplets. Three tests are used to provide indirect evidence that the reduced span does not stem from an output speech problem - probe digit (Waugh & Norman, 1965), matching span, and span with pointing to digits as output. None of these tests requires speech output, so the criterion is satisfied if performance on them is at a comparable level as for normal digit span. For some patients no information is available on any of these tests. In all these cases, though, there is more direct evidence that the patient has an STM impairment; this is listed in the final column of Table 1.1 and refers to procedures to be discussed later in the chapter. The table also includes information on visual span and spontaneous speech, which will also be considered later. Certain other patients are represented in Table l.l(b). They have been held to have an analogous selective span deficit to the patients in Table 1.1 (a), but it was argued by the original investigators that the behaviour of these patients should not be attributed to an STS impairment; actually in certain cases their disorder does not correspond to the pattern described earlier.1 In addition, a subject with a developmental deficit of

Table 1.1. Extensively studied STM patients Test in which selectivity of span deficit demonstrated

Span (auditory) Digits

Letters

Words

KF 12

WAIS

2.3

1.8

2.3

3.0

JB 23

WAIS

3.4

2.5

2.5

>4.0

WH2

WAIS

2.9

2.5

2.4

>4.0

Dysphasic

IL4

Clinical (?)

2.9

1.7*

1.9*

> 5.0

Minimal expressive difficulty Fluent; word-finding difficulties and paragrammatism Fluent; virtually normal

Span

Spontaneous speech

Reduced span from an output speech problem

Auditory word ideni test

Direct

Tapping Test^/ Coughlan/ Warrington^y Tapping Test,/ Peabody^

Peterson x recency minimal Peterson x recency minimal Peterson x recency minimal

STM deficit

(a) STM patients

MC5

Boston

1.5*

—

1.9*

PV6

Milan Aphasia

3.1

1.6

2.5

4.5

EA7

Minnesota (?)

2.0(7)

2.0(7)

< 2.0

2.4

TP

Boston

3.5

—

1.9

3.5

> 2.5

Halting; wordfinding difficulties Fluent; virtually normal

Fluent; "occasional phonemic paraphasias" Fluent but hesitant

Probe digit x

Probe digit x

Matching span x

Tapping T e s t , /

Clinical^ (?) Probe digit x

Peabody^/

—

Pointing x

Milan Aphasia^/ Peabody^/

Pointing x

Minnesota^/

Peterson x recency minimal —

Pointing x

Boston^/

—

EDE9

TB10 RAN11

Boston 2.0 (NB: Comprehension also very poor) WAIS 2.0 WAIS 2(1)

Fluent with occasional paraphasias

2A —

2.1 —

2.0 > 3.0

NHA11

WAIS

3.0

—

> 2.0

—

ER13

Milan Aphasia

2.5

1.4

2.3

> 2.9

J T 14

Not selective ((WAIS)) WAIS

1.8

3.0

1.2*

3.5

Virtually normal Fluent but hesitant with some phonemic paraphasias Hesitant but syntactically correct Moderately nonfluent but not agrammatic Fluent with paraphasias

Matching x

Peabodyyj

Peabody y/

Recency minimal Peterson x

Peabody,/

Peterson x

Probe word x

Lexical Decision^/

Recency minimal (with digits)

—

Word-Picture Matching^/

Recency minimal

(b) Patients not clearly meeting the criteria

JO 1 4 15

Paris Battery

CA2 1 5

Paris Batterv

CA1

1.4* +

1.0*

1.1*

>3.9

1.8*

2.3*

>3.3

Anomic and paraphasic Relatively intact; some anomia Fluent with some distortions and repetitions Fluent but hesitant with phonemic paraphasias

Pointing x ((Matching^)) ((Pointing better))

?

Matching span x

Paris (?)

Matching span x

Paris (?)

((Peabody x ))

Table 1.1. (cont.) Test in which selectivity of span deficit demonstrated

Span (auditory) Digits

Letters

Words

CA3 15

Paris Battery

2.9 +

2.4*

2.5*

LS16

WAIS

1.9

1.9

1.6

RE17

WAIS

4

-

Span (visual)

>3.5

Direct evidence of STM deficit

Spontaneous speech

Reduced span does not stem from an output speech problem

Auditory word identification test

Fluent with phonemic paraphasias

Matching span x

Paris (?)

((Matching span better))

Tapping

Matching span x

Recency Many-sentence comprehension.^/ minimal

(c) Development STM subject 2.8

5?

Fluent, word finding difficulties, paraphasias Normal

test 7

*Performance with single items is not intact. ((...)) indicates that the test gives results at variance with that theoretically required on this patient on an STM-impairment position. CA (conduction aphasic) refers to Tzortzis and Albert's patients, x = Very poor performance; yj = Good performance. The measure of span is the summed probabilities of correct performance for all lengths of list, where possible. For some patients for whom the precise information is unavailable approximations are given. Sources:

1, Warrington and Shallice (1969); 2, Warrington et al. (1971); 3, Shallice and Butterworth (1977); 4, Saffran and Marin (1975); 5, Caramazza et al. (1981); 6, Basso et al. (1982); 7, Friedrich et al. (1984); 8, Saffran (1985), Saffran and Martin (chapter 6); 9, Berndt (1985, chapter 5); 10, Baddeley et al. (1987); 11, McCarthy and Warrington (1987); 12, Saffran and Martin (chapter 6); 13, Vallar, Basso and Bottini (chapter 17); 14, Kinsbourne (1972); 15, Tzortzis and Albert (1974); 16, Strub and Gardner (1974); 17, Campbell and Butterworth (1985).

Impairment of auditory—verbal short-term storage

17

auditory-verbal STM has been reported (Butterworth, Campbell, & Howard, 1986). Her performance is shown Table 1.1 (c).

1.1.2. Predictions from the 1970-vintage model: one or more short-term stores We will consider the functional explanation for the pattern of performance shown by these patients in two stages. First we will consider it in the context of a simple shortterm memory model. Then we will ask whether the changes that neuropsychological findings force on the model are the same as those that normal data require. Prior to 1970 many memory theorists believed that a short-term memory store existed that could be functionally distinguished from iconic stores and long-term memory stores (LTS) (e.g., Waugh & Norman, 1965; Murdock, 1967; Neisser, 1967; Atkinson & Shiffrin, 1968). We will call this the simple STS model. Could the behavioural pattern described in the previous section arise from an impairment to such a store? An essential prerequisite for the account to be worth considering is that storage required in span should be primarily the responsibility of a single memory system. There are a number of pointers from normal subjects that this is in fact the case. It was established in the 1960s that in the short-term retention of verbal material, whether the stimuli are presented auditorily or visually, the predominant error type is producing a word phonologically similar to the presented word (Conrad, 1964). Moreover, it was shown that the use of phonologically confusable items (e.g., rhyming letters) reduced span (e.g., Wickelgren, 1965a). Thus ofie store involved in span appeared to be phonologically based. If more than one store were involved, complex interactions between the variables that affect span would seem likely in normal subjects. For instance, if one store were affected by the phonological confusability of the stimuli and the other were not (e.g., a phonological and a central short-term store), how much the two were used would vary with rate of presentation and an interaction would be found. However, Baddeley (1968) obtained comparable phonological similarity decrement at all positions of a six-item list, and in an extensive study of the effect of a number of variables on span Sperling and Speelman (1970) found a general lack of interactions. They concluded that the main contribution did in fact come from a single store. These experiments all drew the stimuli for each trial from a small set - either digits, letters, or a small group of short words. This procedure may be contrasted with word span where each word is used on only one trial. For word span, Craik (1968a) argued that the STS readout "is augmented by one or two words retrieved from SM" ("secondary memory", i.e., long-term memory). In digit or letter span, which are as long or longer than word span, the amount retrieved from LTS must be considerably less

18

Shallice and Vallar

since the stimuli involved have little semantic content, and what semantic content they have is very similar from one trial to the next. Such digit or letter span tasks therefore approximate better to "single store" tasks.2 What form would an impairment to this single system take? One consequence would be rapid forgetting when effective retention depends on this system. An obvious experimental paradigm to test this is the Brown-Peterson procedure. It is sometimes assumed that the rapid decline in performance in Brown-Peterson tasks may not be reflecting the loss of information in STS because of the "proactive interference" effects that occur over the first few trials and that are attributable to LTS (e.g., Crowder, 1982). However, if recall can be from either LTS or STS, then for forgetting to occur it must not be possible to retrieve information from either store. Thus an LTS locus for the deterioration in performance over the first few trials is quite compatible with the Brown—Peterson decline within a trial at later stages of the experiment, reflecting loss of information in short-term storage. Four patients have been tested with recall for a single auditorily presented item following a brief period of distracting activity - 5 sec or less. Two patients (KF and PV) were perfect on immediate recall, and the other two (RAN and NHA) performed at 90% or better. They showed declines ranging from 20% to 70% when tested after the distracting activity (Warrington & Shallice, 1972; Basso, Spinnler, Vallar, & Zanobio, 1982; McCarthy & Warrington, 1987). This is a considerably greater decline than is found with a single item in normal subjects.3

Visual versus auditory span A standardly accepted characteristic of a store primarily responsible for span is that it is especially useful for auditorily presented material. Span is greater for auditory than for visual input - the "modality effect" (Conrad, 1964). Moreover, span performance for written input is affected by the phonological similarity of the material in the same way as span for auditory input (Conrad, 1964; Wickelgren, 1965a; Baddeley, 1966). By contrast, all but two of the STM patients considered earlier - TI and TB - had a larger visual than auditory span, and the visual superiority is often a large one. Moreover, the pattern of performance in those patients in whom it was examined was very different from that of normal subjects. For PV there was no effect of the phonological similarity between the letters in a list on the number recalled and no effect of preventing rehearsal by using so-called articulatory suppression, that is, continuous uttering of an irrelevant speech sound (Vallar & Baddeley, 1984a). The errors KF made with visual presentation were affected by visual, not phonological, similarity (Warrington & Shallice, 1972). The visual span superiority obtained with the patients presents a grave problem for one means of explaining normal behaviour, namely, that a single short-term store exists


19

that can receive input from both modalities together with a second smaller capacity store that can receive input only from the auditory modality; the most plausible candidate for this second store was the precategorical acoustic store (PAS) (Crowder & Morton, 1969). This type of system would not give rise to visual superiority if either or both of the stores were damaged. In fact, this account of the auditory span superiority shown by normal subjects also has difficulty explaining certain phenomena observed in normal subjects. Thus a simple explanation of the inverse modality effect obtained in the patients is that they use visual STS. Within cognitive psychology, the idea that a separate visual short-term store exists is far from novel (e.g., Margrain, 1967; Sperling, 1967). Moreover, there is now good evidence that even for verbal material a sizeable short-term store can be used for visual input by normal subjects that is distinct from the standard phonological one (e.g., Broadbent, Vines, & Broadbent, 1978; Salame & Baddeley, 1982). For normal subjects the auditory store is presumed to be the larger, but for the patients it can be assumed that their visual STS is larger in capacity than their impaired auditory STS. If this is the case, then the patients would presumably make no attempt to transfer information to their auditory store by subvocalizing as normal subjects have long been known to do (e.g., Conrad, 1964; Sperling, 1967). On this interpretation how is one to account for the two patients who fail to show a visual input superiority for span? Should they be treated as failing to replicate the original rinding? There are at least two possible explanations. Given that the auditory STS has a larger capacity than the visual STS in normal subjects, then mild deficits of the auditory STS may not lead to an inverse modality effect. Second, there is the possibility that the lesion has given rise to a "double deficit." In the light of Caramazza's (1986) argument that it is impossible to provide a proper replication on other patients of observations made in a single case study, is it not unprincipled to assume that where there is a failure to reproduce findings made on the originally studied patients the newer patients are either milder or have a double deficit while the earlier patients are held to be more pure? In fact, the conclusion is pragmatically justified. As there is an auditory superiority in normal span performance, then the a priori probability of obtaining visual superiority in a patient is low. Thus the observation of at least nine patients who are similar in other respects and for whom there is a visual span superiority indicates that the original finding can be attributed to an effect of the lesion; it does not arise from some specific characteristics of that patient that are not relevant for modelling, say, premorbidly atypical capacities. In other words, replication has been obtained. It does, however, remain an assumption that the lack of a visual span superiority in the other two patients - TI and TB - can be explained by the mildness of their basic syndrome or through a double deficit.4 The auditory STS has frequently been argued to be structured on the basis of temporal order. Phonological similarity, in particular, has most effect when the retrieval task involves order (Wickelgren, 1965a; Watkins, Watkins, & Crowder, 1974). Thus

20

Shallice and Vallar

Watkins et al. found that a phonological similarity decrement occurs in span only when scoring involves position as well as items correct. When normal subjects reach the limit of their span their errors are frequently ones of order (Ryan, 1969). A similar effect occurs in those STM patients in whom it has been tested, but for much shorter lists. On four-item lists the percentage of order errors made by KF, JB, and WH was 40%, 17%, and 75% respectively (when performance on digits and letters is combined) (Shallice & Warrington, 1977), and was 62>%, 55%, and 62% (digits only) for the three conduction aphasics of Tzortzis and Albert (1974). As will be discussed in section 1.2, STM does not just involve order. Again, though, performance fits with that which would be expected from an impaired auditory—verbal STS.

Auditory span: the verbal—nonverbal contrast

The issue of whether the auditory store impaired in the STM patients is speech specific has been less widely investigated. Two types of nonspeech auditory stimuli have been used — sequences of meaningful sounds and of taps. The performance of KF and JB on short-term recall of sets of three familiar sounds presented at a rate of one per 3 sec was assessed by Shallice and Warrington (1974). Articulatory suppression was used to prevent the normal control subjects with whom the patients were compared gaining an advantage by rehearsing the names of the sounds — the subjects counted rapidly to themselves. The large difference in performance between patients and controls that exists if three letters are used as stimuli is no longer present when familiar sounds are the stimuli. The conclusion was drawn that the store that is impaired is used only for speech input. The results using sequences of taps are less clear cut. Tzortzis and Albert (1974) found that their conduction aphasics were impaired in reproducing a sequence of taps. However, they did not use articulatory suppression, so their procedure made no allowance for the assistance normal subjects obtain in the retention of nonverbal material from verbal mediation and rehearsal. Indirect support that this is an important factor is available from an experiment carried out by Friedrich, Glenn, and Marin (1984) on EA. Using random sequences of two tones presented at a rate of two per second - too fast for effective subvocalizing - EA performed within the normal range in being able to reproduce sequences of six tones by tapping them out. Intact nonverbal auditory span has therefore been obtained in an STM patient for sequences of tones as well as of meaningful sounds. Converging evidence that the primary store for span is specific to speech can be obtained from a number of studies on normal subjects. The most direct analogue to the results of the patients comes from the comparison of short-term memory for words and for familiar sounds. A serial probe task, in which the subject must give the item in the stimulus list that occurred immediately following the probe, is more difficult for normal


21

subjects if the stimuli are familiar sounds than if they are words. This effect is found when free recall for two types of material is much the same (Rowe, 1974; Philipchalk & Rowe, 1971). Rowe therefore argued that in short-term memory for familiar sounds the short-term store employed for order-based verbal material is not used.

Auditory—verbal STS and LTS Early models of short-term memory presupposed that the short-term store held a temporary representation of the input while a more permanent trace was constructed (e.g., Waugh & Norman, 1965; Murdock, 1967). Various different lines of argument suggested that a separation existed between short-term and long-term stores (see Glanzer, 1972; Baddeley, 1976; Baddeley, this volume, chapter 2, for review). At the time the first STS patients were investigated, one aspect of their performance that presented a problem for this model was that auditory-verbal performance was intact on a number of long-term memory tasks. At least four patients (KF, JB, WH, and PV) have performed long-term memory tests normally, that is, on the Wechsler Memory Scale paired-associate learning, on learning 10 words given in repeated presentations, and the Warrington (1984) forced-choice recognition memory test. Of one patient, by contrast, it was held that long-term memory is patchy, particularly on recognition measures (Baddeley & Wilson, 1988). Again the dissociation between intact long-term and impaired short-term performance has been replicated, and Baddeley and Wilson point out that TB, who has the double impairment as well as a visual span impairment, is most plausibly treated as having associated deficits (see Shallice, 1979, for discussion of the lack of complementarity in inferences from dissociations and associations). If the auditory—verbal LTS of the STM patients is in fact spared, then on twocomponent memory tasks, where measures of STS and LTS can be separately isolated, the defect should be restricted to STS only. Within the classic LTS—STS framework free recall of unrelated auditorily presented words was the standard test that allowed this distinction to be made. It was held that the much better recall of the last few items compared to the earlier items that normal subjects show — the so-called recency effect — comes from these words still being represented in STS at the time of recall (e.g., Waugh & Norman, 1965; Glanzer & Cunitz, 1966). The patients who have been tested on free recall are KF, JB, WH, and PV (Shallice & Warrington, 1970; Warrington, Logue & Pratt, 1971; Vallar & Papagno, 1986). All the patients show a recency effect in immediate free recall of auditory lists limited to one item. In the earlier patients the studies of free recall did not employ normal controls; comparison of performance with that of the control subjects used in the amnesia study of Baddeley and Warrington (1970) indicated that the "secondary memory" component of the STS patient's performance was in the normal range (Shallice, 1979). Vallar and Papagno (1986) have now shown that PV, who has a greatly reduced recency effect with auditory input, has

22

Shallice and Vallar

an above average long-term component with auditory input when contrasted with normal controls. In those patients for whom it has been carried out (KF, JB, and WH), comparison of immediate and delayed recall using a capacity estimation procedure developed by Baddeley (1970) also showed that the STS patients had a normal longterm memory component. More recently the interpretation of free recall performance in terms of two components representing different memory stores has been widely attacked. In particular, the alternative suggestion has been made that the recency effect arises from a retrieval strategy based on the serial position of items or their temporal dating (Tulving, 1968; Bjork & Whitten, 1974; Baddeley & Hitch, 1977), Evidence from PV that supports the earlier position will be discussed in section 1.3.4. Overall, the syndrome fitted well with the late 1960s position on the existence of a separate short-term store. If the same short-term store is involved in span and in the recency component of free recall, then the existence of highly selective disorders in which both these two measures are grossly impaired is to be expected. The STM patients are like normal subjects in producing phonological and order errors when their span is exceeded. This suggests that the same system is in operation but has a much reduced capacity. That the impairment involves the store itself is supported by the very rapid Brown-Peterson forgetting found in the patients. One aspect of the patients' behaviour that did not fit the original models well is their preserved long-term memory performance. However, this could be easily dealt with by assuming that the STS and LTS used in auditory-verbal tasks are in parallel rather than in series (Shallice & Warrington, 1970). The contrast between the relative improvement most patients showed in visual span compared with auditory span, and the better performance normals show on auditory span, indicated that there are more than just one short-term store; earlier findings from normal subjects had also suggested that a visual STS existed as well as an auditory one. The evidence that the patients had specific problems only with auditory-verbal input and made phonological errors in repeating more than one item supported the standard view of the contents of the store, namely, that they are phonologically coded (e.g., Baddeley, 1968; Sperling & Speelman, 1970). This left the issue of what level of phonological representation is involved. Some theorists preferred a lexical level (e.g., Craik, 1968a) and others a prelexical one (e.g., Sperling & Speelman, 1970). It would fit with a word form or connectionist position if both levels are represented in the auditory—verbal STS. The earlier findings obtained on the patients do not, however, provide any relevant evidence on which alternative should be preferred.5

1.2. The relation of the syndrome to more classical ones The account of these patients in terms of damage to a specific short-term store was soon challenged. The most widely used framework for the aphasias was the


23

Wernicke-Lichtheim position, which had been resurrected as an overall theory by Geschwind (1965) and also provided the conceptual basis for the test battery of Goodglass and Kaplan (1972). An important syndrome within this framework is conduction aphasia, held to arise from a disconnection of the input speech systems from the output ones. On this framework the syndrome has two main aspects. The speech function that should be most impaired is repetition. A second aspect that fits less simply into the disconnection approach is that spontaneous speech should be fluent but contain phonemic paraphasias. Specific difficulties with span tasks would seem to fit fairly well with the Wernicke-Lichtheim position on conduction aphasia. Three specific accounts have been put forward. Kinsbourne (1972) developed an explanation for an impairment in span performance in the spirit of the classical Wernicke-Lichtheim account in terms of a disconnection between posterior and anterior speech systems. Tzortsis and Albert (1974) argued that the difficulty lay in the sequential programming of speech production, applying an explanation of conduction aphasia put forward by Dubois, Hecaen, Angelergues, Maufras Du Chatelier, and Marcie (1964). Strub and Gardner (1974) took a third position, one related to those of Goldstein and Kleist, namely, that the problem is a central aphasic disorder of the "processing, synthesis and ordering of phonemes." In Shallice and Warrington (1977) these accounts of the STM patients were considered as alternatives to the auditory-verbal STS account of the disorder. All three alternatives were rejected. The impaired transmission route theory cannot explain the very poor performance on probe digit tasks showed by KF, JB, and MC (Shallice & Warrington, 1970, 1977; Caramazza, Basili, Koller, & Berndt, 1981) where the only information that must be transmitted to the speech production system is that necessary to program "Yes" or "No". 6 In addition, grave problems for this interpretation are presented by the frequency of order errors in the span performance of the patients and, more particularly, by the more rapid decline patients show in retention over time in a Brown-Peterson task when compared with normal subjects. In the Brown-Peterson task neither normal subjects nor the patients can rehearse, so the difference in pattern cannot arise from a difference in the ability to carry this out. The recognition probe and the Brown-Peterson results present equal problems for the impaired-ordering explanation put forward by Tzortsis and Albert (1974). Order is an irrelevant factor in these tasks and also in free recall, on all of which the STM patients are impaired. This alternative also seems implausible. Strub and Gardner (1974) present two further arguments against the STM hypothesis. One is that span performance increased with a slower rate of presentation; the other is that the patients have difficulty in immediate recall of single nonsense syllables. Neither is in fact a problem for the hypothesis, given that the STM patients attempt to compensate for their inadequate STS by resorting much more to systems little used by normal subjects in span, and in particular to LTS (see Shallice &

24

Shallice and Vallar

Warrington, 1970; Saffran & Marin, 1975). Moreover, it is standardly accepted that a nonsense syllable takes up more of the capacity of the STS than does a word of equivalent structure (e.g., Heron & Craik, 1964). Allport (1984a, b) has recently presented further arguments for the position that the STM patients have an impairment of a central phonological code used in both speech perception and speech production. Of the patients he discusses, JB is the only one who has an STM deficit that has been analysed in detail. Three aspects of JB's performance suggested a central phonological code impairment. One, weak performance on a same-different task with consonant-vowel-consonant (CVG) nonwords (26% errors), could easily be attributed directly to an STS difficulty. A second effect was that in repetition of three words JB performed poorly when the words were of low frequency and abstract, making frequent phonemic paraphasias, which do not occur in her spontaneous speech. This, however, would also follow from the STM account as lowfrequency, abstract words cannot be retained in secondary memory; reliance on the impaired auditory—verbal STS for such words would lead to phonological approximations, which also occur in normal subjects when span is exceeded. With single lowfrequency abstract words JB does occasionally make reproduction errors. However, as pointed out by Vallar and Baddeley (1984b), it is not clear whether the words on which this occurs are within her speech vocabulary. One task - auditory lexical decision - where she performed less well than normal controls (10% vs. 3% errors) is less easily accounted for directly through an STS impairment. However, whether her performance is as far outside the normal range as it is with span is unclear.7 Three other patients produced normal performance on a test of input phonological processing similar to the one Allport used with JB. PV scored at 60/60 in same—different judgments of CV syllables, some of which differed by only a single distinctive feature (Vallar & Baddeley, 1984b). A second relevant patient is EDE (Berndt, 1985; Berndt & Mitchum, this volume, chapter 5), a right-handed patient who had a right hemisphere lesion, a span of two, and some comprehension difficulty for sentence material, but few individual word comprehension problems. Her performance was 90% or better on phoneme discrimination both in CV nonwords and in withinword identification (e.g., robe vs. rope, rose, road). It is not, however, clear that a disconnection explanation of the Kinsbourne type can be rejected in EDE, since matching and probe tests are not reported for her. Moreover, in EDE, who developed an aphasia after a right-sided stroke, the unusual laterality may increase the probability of such an account being relevant.8 Finally, TB (Baddeley, Vallar, & Wilson, 1987) was also perfect on making same-different judgments on CV and VC pairs. Overall, the idea that an impairment to a "central phonological code" would account for the syndrome is not strongly supported. If the specific span deficits cannot be explained by accounts derived from the


25

literature on conduction aphasia, how are the two concepts to be related? The most plausible view is that the classical syndrome of conduction aphasia as described empirically fractionates into a number of more basic functional syndromes. One — an impairment of STS — produces a specific inability to repeat a series of short highfrequency items, namely, span. A second disorder — to the phonological assembly system — would give rise to an inability to reproduce, either in spontaneous speech or by imitation, a single long infrequent word, even though span for short familiar items is intact (see Shallice & Warrington, 1977; Luria, 1977), A third, the classical disconnection itself (see McCarthy & Warrington, 1984, for a modern case), would involve reproduction more than spontaneous speech. In all, however, "repetition" as loosely defined would be impaired and errors would arise at the level of the ordering of phonemes.

1.3, Specific memory issues 1.3.1. The multiple store framework The STS approach of the 1960s began to be questioned for a variety of reasons in the early 1970s (e.g., Craik & Lockhart, 1972). It had become increasingly clear that shortterm storage was intimately linked to on-line processing and its relation to long-term storage seemed increasingly remote. In particular, auditory—verbal short-term storage began to be explicitly linked with language processing (Morton, 1970; Shallice & Warrington, 1970; Jarvella, 1971). In addition, it was becoming apparent that a variety of short-term storage systems might exist. One approach - that of levels of processing (Craik & Lockhart, 1972) - was to abandon specific stores and discrete processors in favour of a continuum of processes with storage as their side product. There has been little attempt to relate the disorders of STM patients to this type of theoretical framework. The other approach was to postulate a number of stores, each having a specific role and a specific place in the processing systems (e.g., Morton, 1970; Baddeley & Hitch, 1974). Models of this type became known as "working memory" models or "multiple store" models (see Monsell, 1985, and Baddeley, 1986, for reviews; for a more complex but related approach see Barnard, 1985). In this section we will assume a model of this general type (see Figures 1.1 and 1.2 for two representations of it) and make two further assumptions, for which the evidence has already been discussed: 1. The storage of the information used in span for a series of unrelated items is primarily the responsibility of a single phonologically based store. 2. The STM patients have an impairment to that store.

26

Shallice and Vallar

Visual Word-Form System

Phonological Word-Form System (Input)

Verbal Semantic and Syntactic Systems

Phonological Word-Form (Output) and Assembly Systems

Visual -Verbal STS

Orthographic -to-Phonological Recoding

Figure 1.1. The subprocesses involved in short-term storage of auditorily and visually presented verbal material placed within the context of the word-form approach to word recognition (see Shallice & McCarthy, 1985, for further discussion of this approach). This approach does not include any differentiation of the processes underlying the higher stages of speech production and the corresponding processes in rehearsal. The two wordform systems are involved in the processing of nonwords as well as words.

We will consider the neuropsychological evidence relevant to three contentious questions within the overall working memory/multiple store model approach and assess how the answers obtained relate to those derived from experimental findings on normal subjects. The issues are (a) whether the phonological buffer store is part of the input or the output system, (b) the relation between the systems used for implicit rehearsal and for speech production, and (c) the origin of the recency effect in free recall. In this section we also consider findings obtained with two experimental procedures widely used since the 1970s - articulatory suppression and the effect of varying word length. The evidence from these manipulations requires us to consider an additional interpretation of the STM patient data, namely, that their impairment is specific to the rehearsal process (see section 1.3.2).

Impairment of auditory-verbal

1

(Auditory Input)

short-term storage

27

(Written Input)

f

Phonological Analysis

Visu al Anal ysis Phonological Short-Term Store

J

Articulatory Rehearsal

i

Phonol ogical Recodi

Figure 1.2. Components involved in short-term retention of auditory and visual information (modified from Vallar & Cappa, 1987). The phonological similarity effect is held to occur in the phonological short-term store (B) and the word length effect in the articulatory rehearsal component (C), which is the component affected by suppression. The articulatory rehearsal unit corresponds to the part of the Phonological Assembly system used in rehearsal.

1.3.2. The locus of the phonological buffer store: input or output? Evidence from normal subjects

Since Conrad's (1964) observation that immediate memory is greater for sequences of items that are phonologically dissimilar rather than similar (the phonological similarity effect), there has been a consensus on the major role of phonological coding in verbal short-term memory. Even though the existence of a phonological code and its relevance to STM are generally accepted, its precise nature is a matter of considerable controversy. The phonological representation involved in short-term retention could have an auditory, articulatory, or more abstract form (see Wickelgren, 1969), and all such codes might contribute to span performance. In theoretical analyses the primary store responsible for span has been located in the speech input system (e.g., Green, 1973; Shallice, 1975) or in the speech output system (e.g., Morton, 1970; Baddeley & Hitch, 1974; Ellis, 1979). In normal subjects the role of articulatory and auditory codes in short-term memory has been investigated over the last 20 years by using articulatory suppression. There is

28

Shallice and Vallar

considerable evidence that suppression selectively interferes with articulatory coding without producing a general disruption of information processing.9 Suppression abolishes the phonological similarity effect when the material is presented visually (Murray, 1968; Peterson & Johnson, 1971; Baddeley, Lewis, & Vallar, 1984; Baddeley, this volume, chapter 2). In the case of auditory input, the persistence of the effect under suppression indicates that auditory stimuli have direct access to a nonarticulatory auditory store, consistent with Sperling's original suggestion (1967). For visual stimuli, two interpretations are possible. Visual stimuli could be recoded and conveyed to an auditory store by means of an articulatory rehearsal process (Sperling, 1967). Suppression, by disrupting rehearsal, would abolish the phonological similarity effect through preventing visual items from entering the phonological store. The effect would then reflect the operation of an auditory store, with both auditory and visual inputs. Alternatively, visually presented material could be recoded and stored in an articulatory buffer (see Levy, 1971), suppression would then abolish the phonological similarity effect thereby interfering with this articulatory storage component. On this view, auditory and articulatory codes both generate the effect, and do so for auditory and visual inputs, respectively. The interpretation that rehearsal feeds the (input) phonological store (Sperling, 1967; Vallar & Baddeley, 1984a) is, however, supported by the observation that the disruptive effects of unattended speech on immediate memory for visually presented material are abolished by articulatory suppression (see Baddeley, this volume, chapter 2). Unattended auditory stimuli, which directly feed the auditory store, compete with visual stimuli, which have indirect access to the system through the rehearsal process. Unattended speech would then disrupt memory performance, preventing the utilization of the auditory store by visually presented material. Under articulatory suppression, which blocks the rehearsal process, visual material cannot enter the auditory store. Accordingly, no disruptive effect of unattended speech occurs. Additional information concerning the role of articulatory coding in short-term memory comes from the study of another variable that affects immediate serial recall: the length of the individual memory items (see Baddeley, this volume). Memory span performance is greater for short words than for long, an effect that reflects the temporal duration of the items, rather than the number of component syllables. Articulatory suppression during presentation of the stimuli abolishes the word length effect with visual but not auditory input. However, when subjects suppress articulation throughout presentation and recall, the effect is disrupted in the case of auditory input too (Baddeley et al., 1984). This has led these authors to argue that the word length effect reflects the operation of an implicit articulatory component that is accessible to both input modalities. There is some evidence from serial and probe recall tasks that these disruptive effects of suppression are more pronounced with visual than with auditory input (Levy, 1971; Baddeley, Thomson, & Buchanan, 1975, Experiment 8). The observation that under certain conditions the word length effect is more


29

susceptible to the disruptive effect of suppression when input is visual has a parallel in the developmental data. The detrimental effect of word length on span performance may be observed in children as young as 4 - 6 years old in the case of spoken words: with pictures, however, the effect may be clearly observed only in 10-year-old children (Hitch & Halliday, 1983; Hitch, this volume, chapter 9). One possible explanation of this contrast, in accord with the position just described, is that auditorily presented material might have a privileged access to the articulatory rehearsal system. This fits with findings from dual-task experiments where it has been proposed that the auditory—articulatory route may be regarded as a privileged loop (see McLeod & Posner, 1984). The findings of Baddeley et al. (1984) seem in conflict with the attractive alternative that the capacity of the store itself is influenced by word length. The crucial finding here is the lack of a word length effect under suppression with auditory input and written recall; in this situation information has direct access to the phonological store. The visual input condition is less relevant, since in this latter case written recall under suppression is likely to make little use of information in the phonological buffer. To summarize, the normal data suggest that the main functional component involved in short-term retention of verbal material is a phonological (auditory) short-term store. Auditory stimuli have direct access to this system, whereas visual stimuli feed it through an articulatory rehearsal process, as originally suggested by Sperling (1963) (see also Vallar & Baddeley, 1984a). A final open issue concerns the operation of the articulatory rehearsal process. Given the storage function of the input phonological short-term store, rehearsal is unlikely to involve a system with a major capacity, which would be a duplicate of the input store. It may be viewed as a process that operates on the content of the phonological store, on one item at a time (e.g., Sperling & Speelman, 1970), with the rehearsal of short words being more rapid than that of long ones. Rehearsal may alternatively be conceived (see, e.g., Vallar & Cappa, 1987) as a time-limited loop (Baddeley et al., 1975) that recirculates the memory traces held in the phonological short-term store and contains more short words than long. The multistore approach discussed so far, which relates a number of immediate memory effects to the operation of discrete subcomponents in the memory process, has also provided new tools for investigating the pattern of impairment of patients with a selective deficit of auditory-verbal span. This evidence is reviewed in the following section.

Neuropsychological evidence

From a neuropsychological perspective, divergent views on the location of the damaged store in short-term memory patients emerged in the mid-1970s. Some authors (e.g., Warrington et al.,,1971; Saffran & Marin, 1975; Shallice, 1979) saw its role as primarily in speech comprehension. Others preferred an interpretation in terms of a

30

Shallice and Vallar

defective articulatory buffer (Baddeley et al., 1975; Ellis, 1979). This latter position, however, is not only at variance with the evidence for normal subjects just discussed, it also encounters a number of difficulties if one considers further aspects of the neuropsychological picture of these patients. As mentioned in section 1.1, nearly all of the published patients suffering from a putative short-term memory deficit show an auditory—visual dissociation, with a better immediate memory span performance when the stimuli are presented visually. Since in normal subjects an auditory advantage is typically shown, the neuropsychological finding is readily explained by the selective damage of an auditory input store. An interpretation in terms of an articulatorily based output store and its impairment cannot simply account for both phenomena, unless an additional, rather implausible, assumption is made. This could be that the two modalities of input feed two separate articulatory output buffer stores. A second argument stems from the analysis of the spontaneous speech of such patients. According to models assuming that an articulatory buffer is the main functional component involved in short-term memory tasks (Baddeley et al., 1975; Ellis, 1979), this store has a primary function within the speech production process, in holding already compiled sequences of intended speech at the phonological level. Accordingly, patients with a defective short-term memory should always show an associated impairment of speech production; in particular, they should show phonemic paraphasias or excessive pausing. Contrary to this prediction, patients do exist who have a grossly defective auditory memory span and normal spontaneous speech. This has been shown by a statistical analysis of pauses, rate of speech, and errors in spontaneous speech (case JB, Shallice & Butterworth, 1977) and by an analysis of articulation rate (case PV, Vallar & Baddeley, 1984a). From a clinical assessment, case IL (Saffran & Marin, 1975) would appear to be similar.

The STM patients' impairment: rehearsal process or input store?

An alternative hypothesis about the impairment of the patients needs to be considered, which, unlike the output buffer interpretation, incorporates the notion that an auditory input store is a main functional component involved in short-term retention. The findings concerning the effects of phonological similarity and word length in left hemisphere lesion patients with a selective deficit of immediate memory for auditorily presented verbal material appear prima facie consistent with the view that the functional locus of the deficit could be the rehearsal process (Component C in Figure 1.2) and not the input phonological store (Component B in Figure 1.2). Cases KF (Warrington & Shallice, 1972) and PV (Valler & Baddeley, 1984a) show the phonological similarity effect with auditory but not with visual input. In addition, in a span task for auditorily presented words PV does not show any effect of word length,


31

which wholly or in part reflects the activity of the rehearsal process. This pattern of results, which mimics the effects of suppression in normal subjects, raises the possibility that the rehearsal process is the functional locus of the deficit in these patients. This rehearsal hypothesis, however, not only encounters the difficulties of the output buffer interpretation discussed earlier, but also cannot easily account for the observation that JB and PV, in spite of a grossly defective auditory span, have a normal articulation rate. Since a direct relationship exists between span performance and articulation rate in normal subjects (e.g., Baddeley et al, 1975), an interpretation in terms of a primary rehearsal deficit has to make the additional assumption of a dissociation between overt (unimpaired) and covert (impaired) articulation. Another hypothesis remains to be considered. The locus of impairment could be a disconnection (see the two-way link connecting Components B and C in Figures 1.1 and 1.2) between an input phonological store and an output rehearsal process, which would itself be spared. This disconnection hypothesis, reminiscent of the classic Wernicke—Lichtheim interpretation of conduction aphasia, could account for the preserved speech production of patients such as JB and PV, since the output buffer is spared. The visual advantage in immediate memory tasks found in the majority of the reported patients would not be an insurmountable difficulty, since rehearsal would be available for visual, but not for auditory, material. This disconnection hypothesis is, however, the Kinsbourne hypothesis reviewed and rejected in section 1.2, unless the assumption discussed earlier is made of a separation between the output systems responsible for speech and for rehearsal. In addition, it predicts the presence of the effect of word length with visual input (and, possibly, of phonological similarity, since rehearsal operates on phonological representations). The effect should be absent, however, with auditory input, since the connection between the phonological store and reheasal is interrupted. The limited available data are not consistent with this prediction. In addition to the evidence from KF and PV discussed earlier, in the developmental case ER (Campbell & Butterworth, 1985) and in patient RE (Vallar, Basso, and Bottini, this volume, chapter 17) the word length effect is absent in both input modalities, and no effects of phonological similarity with visual input have been found. Furthermore, while these patients have an auditory span ranging from about 1 to 3 digits, span in normal subjects drops from about 7.9 to only about 5.7 digits with suppression throughout input presentation and recall (Baddeley & Lewis, 1984). The hypotheses of either (a) an isolated deficit of the articulatory rehearsal component or (b) a disconnection between rehearsal and the phonological store, which both predict comparable performance levels in patients and normals under suppression, cannot account for the much lower span of the patients. A final source of difficulty for such interpretations comes from normal evidence that indicates that the phonological short-term store is the main component involved in recency in immediate free recall of auditory lists, while rehearsal plays a comparatively

32

Shallice and Vallar

minor role (see section 1.3.3 and Vallar & Papagno, 1986). Accordingly, the observations that patients like KF and PV have a reduced or absent recency effect in the immediate free recall of supraspan lists of auditory verbal items, in addition to a defective memory span, cannot be accounted for by rehearsal or disconnection hypotheses. There appear to be two problems for the assumption that it is the phonological store and not the rehearsal process that is the locus of impairment. First, why do patients like KF or PV continue to show the phonological similarity effect with auditory stimuli? This may be explained in terms of direct access of auditory information to a phonological store with a pathologically limited capacity (Vallar & Baddeley, 1984a). If sequences of phonologically confusable items require comparatively more capacity than strings of distinct stimuli (Sperling & Speelman 1970), a reduction in capacity in the phonological store arising from brain damage will cause a decrement in performance without abolishing the phonological similarity effect. Second, why does PV not utilize the unimpaired rehearsal component as suggested by the absence of a word length effect. This failure to use rehearsal might reflect a strategy choice (Vallar & Baddeley, 1984a). Were a storage component defective, a process that makes use of this component would tend not to be adopted, even if the process itself were unimpaired. It is of little use to convey visual information to a defective phonological store. Similarly, relying on rehearsal to refresh phonological traces stored in a defective system might be highly ineffective. The strategy choice hypothesis has not been specifically tested. However, in studies of JB (Shallice, unpublished) no effect on performance was found when a 5- or 10-sec unfilled gap was interposed between presentation and recall. This finding supports the idea that rehearsal is possible, when it is useful. JB, moreover, claims to rehearse.

1.3.3. Rehearsal and speech production: the relation The arguments presented so far have assumed that the same output component: (a) is involved in short-term retention, in the conveying of visual material to the phonological store and the refreshing of the phonological trace, to prevent its decay; and (b) participates in speech production. For instance, in the model shown in Figure 1.1, rehearsal utilizes the phonological assembly system, which is an essential part of the speech production process. There is at present, however, little available empirical evidence to discriminate between the aforementioned unitary view and the hypothesis that the implicit speech used in rehearsal might utilize structures different from those involved in speech production. The model shown in Figure 1.2 does not assume that the same "output" processes are used in rehearsal and in speech production. The two hypotheses, however, make different neuropsychological predictions. On the fractionation view


33

patients might have preserved functioning of both the phonological store and the rehearsal process and yet show evidence of an output buffer deficit. Such patients might make a considerable number of phonemic errors or use only short phrase lengths. Such a dissociation would be incompatible with the unitary view. Are there patients who could be suitable candidates for investigating this putative dissociation? On the fractionation hypothesis one should look for patients with a normal auditory memory span at slow rates of presentation associated with phonemic errors and reduced phrase length in spontaneous speech. Possible candidates would be patients similar to those Damasio and Damasio (1980) observed with an auditory span of seven digits (at a rate of one item per second) who made frequent phonemic paraphasias in spontaneous speech.

The "articulatory" nature of the rehearsal process: evidence from anarthric patients

The rehearsal process has at times been regarded as "articulatory"10 because it is disrupted by articulatory suppression. The operation of rehearsal does not, however, appear to require any contribution from the peripheral musculature, such as kinesthetic feedback. Baddeley and Wilson (1985) and Vallar and Cappa (1987) reported preserved operation of the phonological short-term store/rehearsal process components in two anarthric nonaphasic patients. Vallar and Cappa's (1987) case GF has a brain-stem lesion sparing the cerebral cortex, which is also presumably preserved in Baddeley and Wilson's (1985) patient GB. Vallar and Cappa (1987) investigated short-term memory performance in a second anarthric patient, MDC, who suffers from bilateral cortical lesions involving the motor cortex. MDC has an auditory span of seven digits and shows the effects of both phonological similarity and word length with auditory input, but not with visual. This would be compatible with a failure in the transmission of information about visual items to a preserved rehearsal/phonological short-term store system. In addition, MDC has a defective performance with visual input in rhyming tasks, which require the phonological recoding of nonlexical visual items. This can be explained by making a distinction, related to the one used in the acquired dyslexia literature, between phonological recoding (E in Figures 1.1 and 1.2), the component that provides phonological conversion of visually analysed items, and rehearsal (B - • C - • B), which involves the feeding of the output of phonological recoding to the phonological store and the recirculation of information stored in this latter system. In MDC's case the failure of visual items to enter a presumably unimpaired rehearsal process would appear to stem from defective phonological recoding (see Vallar & Cappa, 1987). The distinction is consistent with findings on the effects of articulatory suppression in phonological recoding tasks, which have little, if any, memory load. Suppression appears to have comparatively minor detrimental effects, when compared with the disruption it produces in the phonological similarity and word length effects and in the

34

Shallice and Vallar

performance level in immediate memory span tasks. Rhyme judgments are affected by suppression, but this does not interfere with homophone judgments (see Vallar & Cappa, 1987, and Besner, 1987, for reviews; see also Besner & Davelaar, 1982). The preservation of homophone judgments indicates that phonological recoding (E) operates normally under suppression, which would, however, prevent its output from entering inner or outer speech. This is compatible with Unit C in Figure 1.1 being able to operate under suppression but its output to inner or outer speech being weakened. Rhyme judgments, in turn, may require additional processing (segmentation, deletion) involving a contribution from components further downstream. Finally, PV's pattern of performance in a range of tasks involving phonological processing of visually presented material (Vallar & Baddeley, 1984b) is in line with this distinction. PV's preserved ability to read both words and nonwords, together with a normal performance in a nonword-picture rhyming task, suggests a preserved phonological recoding process. The possibility remains to be considered that the operations performed according to Vallar and Cappa (1987) by two serial components, phonological recoding and rehearsal, reflect instead the activity of a single articulatory process, as previously suggested by Vallar and Baddeley (1984a). Consider the possibility that the short-term retention of visually presented material in a phonological format requires an involvement of the articulatory code greater than does reading nonwords or rhyme judgments on written items. This would explain why in normal subjects suppression reduces performance level and abolishes the effect of both phonological similarity and word length in the serial recall of visual items, but has comparatively minor effects on phonological tasks with a minimal memory load. Following this line of reasoning, in PV's case both the preserved performance on rhyme judgments and reading nonwords and the lack of phonological processing in immediate memory for visual materials might be accounted for by a partial defect of a unitary articulatory component. The unitary hypothesis cannot, however, easily explain the aforementioned findings (see section 1.3.2) that in normal subjects the disruptive effects of suppression are more pronounced when input is visual, as compared with auditory presentation, and their developmental parallel (see Hitch, this volume, chapter 9). A second source of difficulty is MDC's pattern of performance. She is defective in phonological judgments on written material, and in memory span tasks, rehearsal appears to be accessible to auditory but not visual input. On a unitary hypothesis one is then forced to make the additional assumption that auditory input has a privileged access to rehearsal, while visual items would require some preliminary processing stage. With this modification incorporated, however, the unitary view becomes very similar, if not indistinguishable, to Vallar and Cappa's (1987) model. This explicitly assumes that auditorily presented material has a direct access to rehearsal, whereas a preliminary stage, phonological recoding, is required in the case of visual items.


35

1.3.4. The recency effect in immediate recall As discussed earlier (section 1.1.2) patients with a selective deficit of auditory—verbal span do not show the standard recency effect in immediate free recall of a supraspan list comprising unrelated verbal items. Whereas in normal subjects recency involves the last four or five items, in patients in immediate free recall of auditorily presented lists it may be confined to the last stimulus (Shallice & Warrington, 1970; Warrington et al., 1971) or be totally absent (Basso et al, 1982). These findings are fully consistent with a shortterm memory interpretation of the recency effect, which was the dominant view among psychologists in the late 1960s and early 1970s (e.g., Glanzer, 1972). In the following years, however, the short-term memory interpretation of the recency effect was questioned on the basis of two sets of arguments. There is evidence that a number of variables, such as phonological similarity and word length, which affect immediate memory span, do not influence the recency effect (see Craik & Levy, 1970; Baddeley & Hitch, 1977). This may be interpreted as an indication that the shortterm store involved in memory span is not responsible for the recency effect (Baddeley & Hitch, 1974). Thus, the observation that these variables affect span performance, but do not affect recency in free recall (e.g., Craik & Levy, 1970; Glanzer, Koppenaal, & Nelson, 1972), prima facie suggests that the recency items are not held in a phonologically based short-term store. Two arguments, however, militate against this conclusion. Phonological similarity affects order but not item information (e.g., Watkins et al., 1974), and the latter only has to be retained in free recall paradigms. Accordingly, the absence of the effect does not necessarily indicate a nonphonological encoding of recency items. Moreover, positive evidence that recency items are indeed coded phonologically comes from the observation that misrecalls tend to be phonologically related to correct items only in the terminal positions of free recall lists (Craik, 1968b; Shallice, 1975). A second, and more serious, argument for a dissociation between recency and shortterm storage stems from the observation of "long-term" recency effects in delayed free recall, ranging from seconds and minutes (e.g., Bjork & Whitten, 1974; Watkins & Peynircioglu, 1983) to days or even weeks (Baddeley & Hitch, 1977). Similarly, longterm modality effects have been reported. It has been repeatedly observed in normal subjects that auditory input yields a better performance for the final four or five positions in immediate free recall (for reviews of the standard modality effect see Crowder, 1976; Watkins & Watkins, 1980). This auditory-visual difference, however, has been found to persist after a delay filled by an interpolated activity (e.g., Gardiner & Gregg, 1979). It appears rather unlikely that such long-term serial position and modality effects represent the output of a limited capacity, temporary short-term store. An alternative interpretation may account for these findings. Recency phenomena may reflect the utilization of retrieval cues, such as serial position or temporal dating of

36

Shallice and Vallar

the items in the input phase (Tulving, 1968) rather than output from a specific storage component. Tulving's (1968) suggestion has been subsequently elaborated in some detail, and accounts of short- and long-term recency effects in terms of temporal distinctiveness of the memory items, without any reference to specific short- or longterm storage components, have been suggested (see Bjork & Whitten, 1974; Glenberg, 1984; Glenberg & Swanson, 1986). Unlike the short-term store hypothesis, these retrieval interpretations of the recency and modality effects may account for their longterm persistence. They imply, however, that the defective recency of patients with a reduced memory span should be traced back to some inability to use retrieval cues. Thus, if it is maintained that the low span of these patients reflects the reduced capacity of the phonological store, an additional defect of retrieval strategies is to be assumed, in order to explain the co-occurring lack of recency. It is worth noting here that although such strategies may have a more general role in keeping track of events and maintaining orientation (Baddeley & Hitch, 1977), confusional states have not been reported in patients with a selective deficit of recency and span performance. Within a multistore framework, it remains possible, however, that recency and modality phenomena stem from the application of retrieval strategies to different storage components. The standard recency effect in immediate recall of auditorily presented items may represent the output of the phonological short-term store, whereas long-term recency may involve different retrieval cues and storage components. This interpretation, which explains the span and recency impairments in terms of a single-component deficit, has been recently tested in case PV by Vallar and Papagno (1986). In immediate free recall PV does not show any evidence of recency with auditory input, but with visual material her performance is within the normal range. The auditory-visual dissociation is in the opposite direction to the standard modality effect, but is of larger magnitude. This seems difficult to explain on a temporal distinctiveness theory of short-term memory phenomena (Glenberg & Swanson, 1986), assuming that in normal subjects auditory input gives rise to information that is many times more distinctive temporally than is visual input. The same effect is also present when the patient is instructed to recall the final items of the list first: With auditory input PV's recency remains grossly defective, being confined to the last serial position, as with Shallice and Warrington's (1970) finding with KF, whereas with visual presentation PV's recency performance is comparable to that of the control group. In this recall from end condition, PV has an output order comparable to that of control subjects with either input modality, suggesting that she is able to make an appropriate use of temporal cues. This conclusion is further corroborated by the observation that PV shows a normal long-term recency effect for recall of anagram solutions (Vallar, Papagno, & Baddeley, unpublished). These data are consistent with the view that in immediate recall recency may represent the output of a number of different short-term storage components. The


37

observation that for immediate free recall PV performs better when the input is visual argues that the phonological short-term store has a role in the generation of the modality effect. This is consistent with the way that the effect disappears in normal subjects when the items of the free recall list are phonologically similar (Watkins et al., 1974). The modality effect in immediate recall may reflect the direct and automatic access by auditory stimuli to a passive phonological store, without any additional recoding process. By contrast, visually presented verbal stimuli, which can also be held in a lower-capacity nonphonological, presumably visual, store (see Zhang & Simon, 1985), require the additional operation of phonological recoding and rehearsal as classically suggested by Conrad and Sperling. These conclusions are supported by the observation that recency for auditory items is not affected by a visual concurrent task but the reverse does not occur; namely, an auditory concurrent task interferes with recency for visual items (Anderson & Craik, 1974). Similarly, in the case of visual lists recency is abolished when either auditorily or visually presented tasks are interpolated between presentation and recall, but for auditory lists the disruptive effect is modality specific (Broadbent et al., 1978; Gathercole, Gregg, & Gardiner, 1983). This modalityindependent susceptibility of visual lists to concurrent and intervening tasks might reflect the greater amount of processing needed for short-term retention of verbal material, when presentation is visual. The phonological input store contributes to both immediate memory span and recency in free recall. This, however, is not the case for the rehearsal process, which, as suggested by normal data, appears to be comparatively less involved. In free recall the final items are rarely rehearsed (Rundus, 1971; Craik & Watkins, 1973; Shallice, 1975). So the word length effects that occur in immediate memory span tasks do not arise in the recency component of free recall (Craik, 1968a). The evidence discussed so far concerns the free recall paradigm. In the serial recall of supfaspan lists normal subjects show similar recency and modality effects (e.g., Watkins et al., 1974; Watkins & Watkins, 1980). As for free recall, the recency effect in serial recall for auditory lists is reduced more substantially when the interpolated task is presented in the same auditory modality, whereas with visual presentation recency is abolished independently of the modality of the intervening items (Watkins & Watkins, 1980). Finally, in serial recall the modality effect is significantly attenuated when phonologically similar lists are used (Watkins et al., 1974). Taken together, these normal data indicate that the phonological short-term store also contributes to retention of the recency items of a supraspan list, when a serial recall paradigm is used. Neuropsychological data are in line with this view. Patients IL (Saffran & Marin, 1975), PV (Basso et al., 1982), and EA (Friedrich et al., 1984) do not show any recency effect in the serial recall of auditory lists which exceed in length their abnormally low span. Unlike that of normal subjects, their performance level shows a progressive decline from the initial to the terminal positions of the list. With visual

38

Shallice and Vallar

input, IL has a better performance level overall and shows a recency effect confined to the terminal position, which is not detectably increased if he reads the stimuli aloud; this differs from the case of normal subjects (Crowder & Morton, 1969). The assumption that IL's phonological short-term store, to which auditory stimuli have direct access, is defective readily accounts for his inability to benefit from extra auditory input.

1.4. Anatomical correlates of selective deficits of auditory-verbal short-term memory Table 1.2 summarizes the available anatomical data of the reported patients who have a selective deficit of auditory-verbal short-term memory and also show a superior level of performance with visual presentation. It is apparent that such cases differ both in the aetiology of their cerebral disease (vascular, traumatic, neoplastic) and in the method of assessment of the lesion (postmortem or surgical verification, brain scan, CT scan). This lack of homogeneity (see Vallar & Perani, 1987, for a discussion of the issue), together with the limited number of relevant cases, is an unfavourable condition for a reliable anatomo-clinical correlation. However, from inspection of Table 1.2 it is apparent that in all patients the left parietal region is reported to be involved. A map of the lesion site and size has been provided in four patients. Three patients (KF, JB, WH: Warrington et al., 1971) have temporoparietal lesions, which superimpose in the supramarginal gyrus of the inferior parietal lobule (Warrington, 1979); in PV's case (Basso et al., 1982) CT scan shows a large lesion that involves the whole perisylvian region, including parts of the subcortical white matter underlying the supramarginal gyrus. LS (Strub & Gardner, 1974) has a parietooccipital lesion. In RAN's case (McCarthy & Warrington, 1987) parietal damage was found. For EA (Friedrich et al., 1984), who is presumably not fully right-handed, a detailed description of the left hemisphere injury has been provided: The lesion is reported to include the posterior temporal lobe (primary auditory cortex and Wernicke's area), the supramarginal and angular gyri of the inferior parietal lobule, and the superior parietal lobule. JT (Kinsbourne, 1972) and NHA (McCarthy & Warrington, 1987), who are not listed in Table 1.2 given the lack of precise data concerning the locus of their lesions, suffered from a left-sided head injury and a left middle cerebral artery aneurysm, respectively. Taken together, these data indicate the inferior parietal lobule as a crucial region for the function of the phonological (auditory—verbal) input short-term store. By contrast, JO (Kinsbourne, 1972), whose defective span performance may be at least in part traced back to output deficits (see Shallice & Warrington, 1977, and Table 1.1b), has a frontotemporal, presumably ischaemic, lesion assessed by brain scan; the neurological examination, which revealed a right hemiparesis without any sensory deficit, also suggests anterior damage. This major frontal involvement, which differs

Impairment of auditory-verbal

short-term storage

39

Table 1.2. Left hemisphere lesion site in ten patients with defective auditory—verbal shortterm memory Aetiology

Lesion site

Source

JB2

Head injury Meningioma

Postmortem Surgery

WH 2 LS3

CVA? Head injury

inf. P, O-T sup. half mid. T gyrus sup. T gyrus, inf. P T, sub-F, Inf. P P-O

IL4

CVA? CVA CVA CVA CVA CVA

post. P post. sup. T, inf. P F-T-P perisylvian post. T, sup. inf. P P T, P, insula

Patient 1

KF

MC PV6 EA7 RAN8 ER9

Brain scan Angiography Craniotomy Brain scan? CT scan CT scan CT scan CT scan CT scan

Note: F = frontal; T = temporal; P = parietal; O = occipital; mid. = middle; post. = posterior; SU p. = superior; inf. = inferior; CVA = cerebrovascular attacks. Sources: 1, Shallice and Warrington (1980a); 2, Warrington et al. (1971); 3, Strub and Gardner (1974); 4, Saffran and Marin (1975); 5, Caramazza et al. (1981) 6, Basso et al. (1982); 7, Friedrich et al. (1984); 8, McCarthy and Warrington (1987); 9, Vallar, Basso & Bottini, chapter 17.

from the prevailingly posterior damage found in the majority of the patients meeting the criteria listed in section 1.1.1, may represent the anatomical correlate of JO's output difficulties. The anatomical correlates of the two cases who do not show a visual superiority in immediate memory span differ from those of the patients listed in Table 1.2. No focal lesions were found in TB (Baddeley et al., 1987), who has additional long-term memory deficits, but a CT scan showed some temporal lobe atrophy, suggesting diffuse brain damage. The observation that he has bilateral and diffuse lesions, not confined to the left inferior parietal region, is consistent with the hypothesis that there may be an additional deficit to a visual store component. Case TI (Saffran & Martin, this volume, chapter 16) is also more complex. He is a right-handed man who suffered two successive strokes; his CT scan suggests left posterior temporoparietal and right inferior frontal ischaemic lesions. These limited data do not, of course, provide any information concerning the anatomical correlates of such a visual-verbal STM system.11 The conclusions drawn from individual case studies are in line with a recent anatomoclinical correlation group study by Risse, Rubens, and Jordan (1984), who had a series of 20 left hemisphere lesion aphasic patients suffering from an ischaemic infarction and examined 6 months postonset. The patients were subdivided into two groups according to the anterior-basal ganglia versus posterior site of the CT-assessed lesion. Auditory digit span was defective in patients with posterior lesions, whereas the performance level of patients with anterior damage was within the normal range. The

40

Shallice and Vallar

patients with anterior damage also have the better performance level on immediate memory span for sequences of visually presented digits.12 In more than two thirds of Risse et al/s patients with posterior damage the inferior parietal lobule (supramarginal and/or angular gyrus) and the posterior-superior temporal area are involved. It is worth noting that a trend towards an opposite dissociation was found for verbal long-term memory. Patients with anterior lesions do not show any detectable learning of a word list, whereas the learning curve of patients with posterior damage is similar to the control data, albeit at a lower performance level. Finally, Vallar, Papagno, and Cappa (1988) have recently reported a series of 11 left brain-damaged stroke patients with lexical—semantic processing deficits, but a preserved verbal short-term memory function (normal auditory digit span and recency effect in free recall). The anatomical correlates of this neuropsychological pattern of impairment are a range of cortico-subcortical and purely subcortical lesions, which consistently spare the left inferior parietal region. Auditory—verbal short-term memory may be viewed, as we have previously mentioned, as a phonological short-term storage component. Such a system is, in principle, likely to have close functional connections with other components at the same processing level, namely, phonological analysis and production systems. The question has to be considered, however, as to whether these putative components are functionally independent, and hence susceptible to selective impairment after brain damage, or a single code exists that mediates the phonological aspects of both verbal input analysis and storage and of speech production. Broadly consistent with the unitary view, patients with a putative impairment of the phonological short-term store may suffer from additional phonological deficits such as phonemic paraphasias in speech output and defective phonological analysis of auditory material (EA, Friedrich et al., 1984; MC, Caramazza, et al, 1981). Evidence for associations between the phonological aspects of perception and production is also provided by a group study carried out by Alajouanine, Lhermitte, Ledoux, Renaud, and Vignolo (1964). They took two groups of aphasic patients, differentiated as to whether phonemic or semantic paraphasias predominated in their spontaneous speech. Patients with phonemic paraphasias in spontaneous speech showed phonemic errors also in object naming and in repetition of words and nonwords, which was grossly defective. In addition to these phonological output disorders, these patients had an associated deficit of phonological analysis, as assessed by a word-nonword discrimination task, although oral comprehension of individual words and commands was preserved. Conversely, patients showing verbal paraphasias in spontaneous speech had a normal repetition of both words and nonwords and comparatively spared phonological discrimination abilities, in spite of their poor auditory comprehension.


41

However, as discussed in the earlier sections, dissociations are on record between defective auditory-verbal short-term memory and both preserved spontaneous speech and phonological discrimination abilities. Moreover, the three patients studied by Damasio and Damasio (1980) show a complementary dissociation of a normal digit span and frequent phonemic errors in spontaneous speech. The hypothesis of a central phonological code, which can readily account for the co-occurrence of both input and output phonological deficits, runs into difficulties when the comparatively rarer instances of dissociated deficits are considered. Associations and dissociations may be more satisfactorily interpreted in terms of anatomical contiguity of the cerebral structures involved in phonological coding. Patients clinically classified as conduction aphasics typically show damage of the posterior portion of the left perisylvian region, which comprises the posterior part of the superior temporal lobe and the inferior part of the supramarginal gyrus (see Benson et al, 1973; Green & Howes, 1977; Damasio & Damasio, 1980). The role of these left perisylvian areas in phonological processing has also been suggested by a recent CT-clinical correlation study by Cappa, Cavallotti, and Vignolo (1981), who found that the inferior parietal lobule and the posterior part of the superior temporal gyrus are frequently involved in fluent aphasics with predominantly phonemic errors on a naming task. Conversely, more posterior areas farther from the left sylvian fissure, such as the temporo-parieto-occipital junction, are most frequently damaged in patients with predominantly lexical errors. Additional support comes from studies using electrical stimulation mapping techniques: Ojemann (1983) found that errors in phoneme identification are evoked by stimulation of sites located in the frontal, temporal, and parietal left perisylvian cortex. The left posterior perisylvian region may be then regarded as a phonological processing system that can be functionally subdivided into input components, which may comprise phonological nonarticulatory analysis and short-term storage subsystems, and output components, possibly those involved in Butterworth's (1980) phonological assembly system. The anatomical contiguity of these sybsystems may account for the frequent observation of phonological deficits, which encompass input, storage, and output aspects, such as in the case of the classical conduction aphasia syndrome. Modularity within the phonological system accounts for the aforementioned dissociations, such as the repetition and reproduction forms of conduction aphasia (see Luria, 1977; Shallice & Warrington, 1977). Future research should explore whether these putative functional subcomponents of the phonological systems have discrete anatomical counterparts. Damasio and Damasio (1980) studied six conduction aphasic stroke patients, who have fluent speech rich in phonemic paraphasias, but a comparatively preserved auditory digit span, ranging between four and seven digits. The lesion sites most frequently involved are the insular

42

Shallice and Vallar

region and the superior temporal gyrus, with an extension to the supramarginal gyrus of the inferior parietal lobule; the angular gyrus is spared in all cases. Damasio and Damasio (1980) emphasize the role of the insular lesion, which, besides destroying the insular cortex itself, would damage corticocortical white matter pathways in the external capsule, disrupting the connections between the temporoparietal regions and the premotor frontal structures. An injury to this area could represent one anatomical counterpart of the output disorders of Damasio and Damasio's (1980) patients. On the other hand, in a number of patients with a grossly defective auditory digit span, ranging between 1.5 and 3 digits, the angular gyrus is involved (KF and JB, Warrington et al, 1971; EA, Friedrich et al., 1984). It is worth stressing, however, that the specific short-term memory impairment within the conduction aphasia complex does not yet have a clear anatomical counterpart. For instance, case PV, who appears to suffer from a very selective deficit of the phonological short-term store (analysis and output components both being unimpaired), has an extensive perisylvian lesion. Case EA (Friedrich et al., 1984), who shows a more widespread phonological deficit, has a posterior perisylvian lesion, which appears to spare the more anterior areas. This pattern is strikingly similar to the associations and dissociations between extrapersonal neglect, personal neglect, and anosognosia after injury to the right hemisphere. These disorders frequently co-occur and clear-cut functional double dissociations exist, but in both associated and dissociated cases the cortical correlate remains a lesion of the inferior parietal lobule (Bisiach, Vallar, Perani, Papagno, & Berti, 1986; Bisiach, Perani, Vallar, & Berti, 1986; Vallar & Perani, 1986). It would be very untimely to draw any definitive conclusion from these observations. The previously discussed patients differ in a number of important respects, such as the aetiology and assessment of the lesions and the type of psychological investigation, which ranges from a standard language examination to a detailed investigation of short-term memory function. Anatomo-clinical data of this sort, however, raise the possibility that different patterns of impairment of the phonological system may have partially separate anatomical correlates.

1.5. Conclusions In this chapter we have reviewed the literature on the short-term memory performance of patients who can be characterized theoretically as having an impairment to the input phonological buffer and operationally in terms of four criteria of which a selective deficit in span is the most fundamental. We have considered three groups of subjects (see Table 1.1). Group (a) is made up of those we judge to pass the four criteria. Group (b) has patients who fail on one or more of the criteria but whose impairments have been discussed in connection with one of the hypotheses that have been advanced to explain


43

the deficits of the first group of patients. Group (c) consists of one subject who exhibits the pattern of test performance similar to that of Group (a) but in whom the syndrome appears to be of developmental origin and the patient has a history of developmental dyslexia.13 We would argue that the patients in Groups (a) and (c) of Table 1.1, at least, are examples of a common functional syndrome that corresponds to damage to Subsystem B in Figures 1.1 and 1.2. Of course most of the patients in Group (a) also have lesions to other subsystems. In a number the speech production system is impaired. KF, for instance, has an orthographic-to-phonological recoding deficit, MC has problems reproducing function words, and EA and possibly JB have some phonological analysis problems. We would, however, view these as additional deficits that do not contribute to the basic pattern of performance outlined in section l.l. 14 The concept of a functional syndrome has come under considerable criticism recently (Caramazza, 1986; Ellis, 1987). Theories of normal function, it has been argued, can be tested only by what individual patients do. Similarity of performance across patients has been held not to be of basic importance. Thus Ellis argues that "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." Clearly the classical aphasia syndromes have fractionated (e.g., Schwartz, 1984, Badecker & Caramazza, 1985), and so have more modern ones such as deep dyslexia and surface dyslexia (e.g., Shallice & Warrington, 1980b; Patterson, Coltheart, & Marshall, 1985). Either different elements of the syndrome have been shown to arise from different functional loci or the overall pattern has been claimed to arise from more than one locus of damage. In the area of short-term memory studies we have reviewed there has, however, been much similarity of performance across patients on different tasks.15 There are certain exceptions. For instance TB and TI do not perform better on visual than on auditory span tasks. Should these exceptions be considered failures to replicate the relevant aspects of the original syndrome and so these aspects discarded as not functionally relevant? This possibility was considered earlier and rejected. In the context of a normal auditory superiority, for 9 of 11 patients to have a visual superiority would be most unlikely to arise for reasons unrelated to the lesion. More critically, two plausible explanations exist for the atypical pair of patients. In TB's case there is a strong neurological reason for suspecting a double deficit, namely, the presence of diffuse brain damage (see section 1.4). TFs situation is less clear. He had an additional stroke but its location (right inferior frontal lobe) is not a plausible location for a visual—verbal STS. However, his auditory-verbal span deficit was relatively mild. This will limit the possibility of an inverse modality effect being observed. The characterization of the patients as having a common functional deficit has led to replication of results being in practice possible, to a development of empirical study through transfer of ideas and procedures from investigations on earlier patients to those

44

Shallice and Vallar

on later ones, to a symbiotic deepening of the theoretical links made with normal memory theory and to the possibility of opening the issue of anatomical correspondences. In the second part of this chapter we argued that the impairments of the patients fit well with developments of the multiple store/working memory type of shortterm memory framework. Here information is available on a much smaller range of patients - PV and to a lesser extent KF and JB. The convergence between inferences based on these findings and from normal experiments is in our view impressive. It involves the separability of auditory and visual short-term storage of verbal and nonverbal material, of the input locus for the phonological buffer and for its role in recency. Other areas in which the correspondences are less solid as different interpretations are available relate to the effects of word length and to the role of rehearsal. In this discussion we have adopted a buffer framework. Whether in the longer term an alternative type of STM mechanism such as one of those discussed in section 1.1 will provide a better theoretical basis for characterizing the disorder is unclear. At present the interactive activation approach is popularly seen as a modern competitor to the supposedly out-of-date multiple store or working memory approaches (see other chapters in this volume, particularly Friedrich, chapter 3; Saffran & Martin, chapter 6; Campbell, chapter 11). In our opinion it is inappropriate to view the relation between the two approaches as a competitive one. This is, first, because the concepts are on very different levels of description of the cognitive system. Second, the interactive activation approach is compatible with a variety of procedures for the retention of information over short intervals of time. Thus, if one considers connectionist simulations along with older interactive activation models such as that of McClelland and Rumelhart (1981), short-term memory or serial order phenomena have already been simulated at least four different ways. As mentioned in the Introduction, some theorists (e.g., Hinton & Plaut, 1987) have assigned to each connection in a network a short-term weight that can be varied semi-independently of its long-term weight. In another context — that of the simulation of production system operations in a connectionist architecture - Touretzky and Hinton (1985) have incorporated a working memory that has an equivalent function in the simulation to that of its production system equivalent (see Newell & Simon, 1972) and can hold half a dozen or so separate elements at a time. A third possibility is that used by McClelland and Elman (1986). Their model essentially massively reduplicates the mechanisms required for perceptual processing for separate time slices up to the maximum temporal window that the system allows. Finally, Jordan (1986) has simulated the perception of serial order by adding additional "state units" that move into different modes for each succeeding discrete input; thus the effect of an input at time t1 from the onset of the string is differentiated from the effect of the same input at time t2 from the onset by the different contents of the state units.


45

The concept of a "phonological buffer store" in its turn can be used with at least three levels of specificity. At the most basic level it refers to (a) a structure separable from processing structures that holds information derived from phonological analysis systems over short intervals of time. More specifically there are the implications that (b) the information is held in the form of phonological representations and (c) the processing of subsequent inputs affects the contents of the buffer in a relatively simple way; classically this was through some combination of temporal decay and interference that could be associatively based (e.g., Wickelgren 1965b) or displacement based (e.g., Waugh & Norman, 1965). What is critical is that the operations of the processing system and the buffer are not inextricably interconnected. Finally, (d), the concept can be used in the sense of a unit having a fixed number of storage "slots". The fourth, most specific, usage has long been known to be inappropriate (see, e.g., Neisser, 1967).16 If we take the first three senses of the concept, then the Touretzky—Hinton working memory mechanism if applied to the present domain is compatible with all three and only the McClelland-Elman approach is compatible with none. The other two approaches are compatible with the first and second senses but not the third. However, it is to the first and second sense that neuropsychological evidence primarily speaks. Given the variety of types of implementation of short-term storage processes in connectionist simulations that are compatible with this first usage, we believe that for discussions of the overall functional architecture and how neuropsychological evidence relates to it, the terms phonological store and phonological buffer remain useful if not utilized in their most specific sense. Returning to the STM syndrome itself, if we are correct that it has survived better than the dyslexic ones, why should this be the case? Principally we would argue that if the cognitive architecture is modular, the initial characterization of a disorder in patients with a highly selective impairment reduces the danger of different aspects of the syndrome being derived from different functional impairments. This is, of course, the complement of the problem of additional deficits, which certain theorists now consider noncritical (Caramazza, 1986). In our view even if one cannot logically infer from a selective impairment to a functionally specific disorder (Sartori, 1988), pragmatically the use of such patients still seems the most appropriate methodology for cognitive neuropsychology.

Notes 1. JO (Kinsbourne, 1972), for instance, would fail Criterion 3. His span performance was considerably improved if he pointed to numbers instead of trying to repeat them aloud. It should be noted that JO also differs anatomically from the other patients (see section 1.4). RAN and NHA were originally described as classical "disconnection" conduction aphasics (McCarthy & Warrington, 1984). However, their patterns of impairment changed and were later argued to be of the STM deficit type.

46

Shallice and Vallar

2. Supraspan tasks using words (e.g., Watkins & Watkins, 1977, where lists roughly two times longer than span were used) are more clearly two-component tasks (see section 1.3.4). 3. Baddeley and Scott (1971) argued that the short-term memory component involved in carrying out the Brown-Peterson task only lasts 5 sec. The STM patients who have been tested show deficits on the task at longer intervals than this. Baddeley and Scott's conclusion was derived from single-trial Brown-Peterson experiments where measured performance tends to asymptote after 5 sec. Most of their conditions, however, involved a h've-or-sevendigit stimulus. For the three-digit stimulus - a length more typical of most Brown-Peterson experiments - the STS decline continued well after the 5-sec limit. Moreover, in an experiment of Conrad's (1967) using four consonants, phonological errors were still occurring significantly above chance at his longest (7 sec) distraction interval, even though their rate was considerably lower than after a 2-sec distraction interval. 4. TI is one of the milder patients (see Table 1.1). In TB's case there are grounds for assuming he has multiple deficits (see later in this section and section 1.4). 5. For further discussion on this point see Safrran and Martin, this volume, chapter 6. 6. One possibility that has not been investigated is that it is rehearsal during input that is critical and that the transmission route held to be impaired is required for rehearsal as well as repetition. Theoretically, to exclude this possibility, the probe digit task should be carried out with a rapid presentation rate, such as two digits per second, where rehearsal is not used by normal subjects. The patients should still show an impairment. This has yet to be done. However, the small reduction in auditory span occurring in normal subjects from articulatory suppression (Baddeley et al, 1975) suggests that rehearsal at input is not a major factor in normal span performance with one per second input (see also section 1.3.2). 7. Her errors were principally on the nonwords. Her comprehension of individual words is good. For further discussion of JB's perceptual analysis problems see Butterworth, Shallice, and Watson, this volume, chapter 8. 8. Kleist (1916) argued that conduction aphasia could arise if the systems underlying speech perception relateralized to the opposite hemisphere, but those concerned with speech production did not do so. Repetition would, it was held, then need to utilize inadequate speech production systems in the opposite hemisphere from those underlying spontaneous speech. The plausibility of an explanation of this sort is greater when the patient already has unusual laterality. 9. Independent sources of evidence suggest that suppression has specific disruptive effects, rather than producing a general interference with information processing. In a number of tasks requiring processing of visually presented material, performance level is affected by suppression but not by a concurrent activity such as tapping, which does not interfere with articulation (Baddeley, Eldridge, & Lewis, 1981). Second, the complex pattern of interaction between suppression, phonological similarity, word length, and unattended speech in immediate memory tasks also indicates a specific locus of interference (see Salame & Baddeley, 1982; Baddeley et al., 1984). Finally, at variance with the performance of normal subjects, in PV suppression does not affect performance level in a visual span task (Vallar & Baddeley, 1984a). This is also true of JB (Shallice, unpublished observations). Since PV and JB, as suggested by the absence of the phonological similarity effect with visual input, do not make use of phonological (acoustic and/or articulatory) components in the short-term retention of visual material, this observation is entirely consistent with the hypothesis that suppression has a specific locus of interference, rather than producing a more general disruptive effect (e.g., distraction). 10. The articulatory code of the rehearsal-output buffer components discussed here is likely to be phonemic (rather than phonetic) in nature, with specification of the precise phonetic forms occurring at later stages of the speech production process (see Ellis, 1979, and references therein).


47

11. Warrington and Rabin (1971), on the basis of a cerebral neuropsychological group study, argue that it also has a posterior left hemisphere localization. 12. In line with Risse et al.'s study (1984), Warrington and Rabin (1971) found that immediate memory for strings of visual stimuli presented simultaneously (digits, letters, and lines) is more defective in left brain-damaged patients with radiologically or surgically ascertained posterior lesions, as compared with left brain-damaged patients with anterior damage and right brain-damaged patients. 13. Developmental dyslexia is well known frequently to involve severe deficits on span tasks (see Rugel, 1974, and Crain, Shankweiler, Macaruso, & Bar-Shalom, this volume, chapter 18). 14. Others (e.g., Caplan & Waters, this volume, chapter 14) argue that phonological analysis problems rather than damage to a phonological buffer are responsible for the impaired span performance of patients like JB. However, in Butterworth, Shallice, and Watson (this volume, chapter 8) it is shown that in sentence processing JB has a specific problem retaining over short periods of time the results of phonological processing but can be unimpaired at semantic analysis and retention, which is based on this self-same phonological processing. To argue that her deficit lies in phonological analysis and not short-term storage is therefore not adequate. 15. This does not apply to the speech comprehension aspects (see Caplan & Waters, chapter 14). 16. For further evidence that presents problems for the notion, see Butterworth, Shallice, and Watson, chapter 8.

References Alajouanine, T., Lhermitte, F., Ledoux, D., Renaud, D., & Vignolo, L. A. (1964). Les composantes phonemiques et semantiques de la jargonaphasie. Revue Neurologique, 110, 5-20. Allport, D. A. (1984a). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes, (pp.

313-325), Hillsdale, NJ: Erlbaum. Allport, D. A. (1984b). Speech production and comprehension: One lexicon or two. In W. Prinz & A. F. Sanders (Eds.), Cognition and motor processes. Berlin: Springer. Anderson, C. B. M., & Craik, F. I. M. (1974). The effect of a concurrent task on recall from primary memory. Journal of Verbal Learning and Verbal Behavior, 13, 107—113.

Atkinson, R. C, & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The Psychology of learning and motivation: Advances in research and theory (Vol. 2). New York: Academic Press. Baddeley, A. D. (1966). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18,

362-365.

Baddeley, A. D. (1968). How does acoustic similarity influence short-term memory? Quarterly Journal of Experimental Psychology, 20, 249—264.

Baddeley, A. D. (1970). Estimating the short-term component in free recall. British Journal of Psychology, 61, 13-15. Baddeley, A. D. (1976). The psychology of memory. New York: Basic Books. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D., Eldridge, M., & Lewis, V. J. (1981). The role of subvocalisation in reading. Quarterly Journal of Experimental Psychology, 33, 4 3 9 - 4 5 4 .

Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47-89). New York: Academic Press. Baddeley, A. D., & Hitch, G. J. (1977). Recency re-examined. In S. Domic (Ed.), Attention and performance VI (pp. 647-667). Hillsdale, NJ: Erlbaum.

48

Shallice and Vallar

Baddeley, A. D., & Lewis, V. (1984). When does rapid presentation enhance digit span? Bulletin of the Psychonomic Society, 22, 403-405. Baddeley, A. D., Lewis, V., & Vallar, G. (1984) Exploring the articulatory loop. Quarterly Journal of Experimental Psychology, 36A, 233-252. Baddeley, A. D., & Scott, D. (1971). Short-term forgetting in the absence of proactive inhibition. Quarterly Journal of Experimental Psychology, 23, 275-283. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of shortterm memory. Journal of Verbal Learning and Verbal Behavior, 14, 575-589. Baddeley, A. D., Vallar, G., & Wilson, B. (1987). Sentence comprehension and phonological memory; Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and Performance XII. The Psychology of Reading (pp. 509-529). London: Erlbaum. Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and the distinction between long- and short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176-189 Baddeley, A. D., & Wilson, B. (1985). Phonological coding and short-term memory in patients without speech. Journal of Verbal Learning and Verbal Behavior, 24, 490—502. Baddeley, A. D., & Wilson, B. (1988). Comprehension and working memory: A single case neuropsychological study. Journal of Memory and Language, 27, 479-498. Badecker, W., & Caramazza, A. (1985). On considerations of method and theory governing the use of clinical categories in neurolinguistics and neuropsychology: The case against agrammatism. Cognition, 20, 97-125. Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to memory. In A. Ellis (Ed.), Progress in the psychology of language (Vol. 2). London: Erlbaum. Basso, A., Spinnler, H., Vallar, G., & Zanobio, M. E. (1982). Left hemisphere damage and selective impairment of auditory verbal short-term memory. A case study. Neuropsychologia, 20, 263-274. Benson, D. F., Sheremata, W. A., Bouchard, R., Segarra, J. M , Price, D., & Geschwind, N. (1973). Conduction aphasia. A clinico-pathological study. Archives of Neurology, 28, 339-346. Berndt, R. S. (1985). Working memory and sentence comprehension. Paper presented at the Second Venice Conference on Cognitive Neuropsychology, Venice. Besner, D. (1987). Phonology, lexical access in reading and articulatory suppression: A critical review. Quarterly Journal of Experimental Psychology, 39A, 467-478. Besner, D., & Davelaar, E. (1982). Basic processes in reading: Two phonological codes. Canadian Journal of Psychology, 36, 701-711. Bisiach, E., Perani, D., Vallar, G., & Berti, A. (1986). Unilateral neglect: Personal and extrapersonal. Neuropsychologia, 24, 749-758. Bisiach, E., Vallar, G., Perani, D., Papagno, G, & Berti, A. (1986). Unawareness of disease following lesions of the right hemisphere: Anosognosia for hemiplegia and anosognosia for hemianopia. Neuropsychologia, 24, 471-482. Bjork, R. A., & Whitten, W. B. (1974). Recency-sensitive retrieval processes in long-term free recall. Cognitive Psychology, 6, 173-189. Broadbent, D. E. (1958). Perception and communication. London: Pergamon. Broadbent, D. E., Vines, R., & Broadbent, M. (1978). Recency effects in memory, as a function of intervening items. Psychological Research, 40, 5—13. Butterworth, B. (1980). Some constraints on models of language production. In B. Butterworth (Ed.), Language production: Vol. 1. Speech and talk (pp. 423-459). New York: Academic Press. Butterworth, B. L., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A, 705—737. Campbell, R., & Butterworth, B. L. (1985). Phonological dyslexia and dysgraphia in a highly literate subject: A developmental case with associated deficits of phonemic processing and awareness. Quarterly Journal of Experimental Psychology, 37A, 435-475. Cappa, S. F., Cavallotti, G., & Vignolo, L. A. (1981). Phonemic and lexical errors in fluent aphasia: Correlation with lesion site. Neuropsychologia, 19, 171-177.


49

Caramazza, A. (1986). On drawing inferences about the structure of normal cognitive systems from the analysis of patterns of impaired performance: The case for single-patient studies. Brain and Cognition, 5, 41-66. Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271. Conrad, R. (1964). Acoustic confusion in immediate memory. British Journal of Psychology, 55, 75-84. Conrad, R. (1967). Interference or decay over short retention intervals. Journal of Verbal Learning and Verbal Behavior, 6, 49-54. Craik, F. I. M. (1968a). Two components in free recall. Journal of Verbal Learning and Verbal Behavior, 7, 996-1004. Craik, F. I. M. (1968b). Types of error in free recall. Psychonomic Science, 10, 353-354. Craik, F. I. M., & Levy, B. A. (1970). Semantic and acoustic information in primary memory. Journal of Experimental Psychology, 86, 77-S2. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684. Craik, F. I. M., & Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12, 599-607. Crick, F. (1984). Function of the thalamic reticular complex: The searchlight hypothesis. Proceedings of the National Academy of Sciences, 81, 4585-4590. Crowder, R. G. (1976). Principles of Learning and Memory. Hillsdale, NJ: Erlbaum. Crowder, R. G. (1982). The demise of short-term memory. Acta Psychologica, 50, 291-323. Crowder, R. G., & Morton, J. (1969). Precategorical acoustical storage (PAS). Perception and Psychophysics, 5, 365-373. Damasio, A. R., & Damasio, H. (1980). The anatomical basis of conduction aphasia. Brain, 103,

337-350. Dubois, J., Hecaen, H., Angelergues, R., Maufras Du Chatelier, A., & Marcie, P. (1964). Etude neurolinguistique de I'aphasie du conduction. Neuropsychologia, 2, 9-44. Ellis, A. W. (1979). Speech production and short-term memory. In J. Morton & J. C. Marshall (Eds.), Psycholinguistic series: Vol. 2. Structures and processes (pp. 157-187). London: Elek. Ellis, A. W. (1987). Intimations of modularity, or, the modelarity of mind: Doing cognitive neuropsychology without syndromes. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language (pp. 397-408). London: Erlbaum. Feldman, J. A. (1982). Dynamic connections in neural networks. Biological Cybernetics, 46, 27-39. Friedrich, F. J., Glenn, C. G., & Marin, O. S. M. (1984). Interruption of phonological coding in conduction aphasia. Brain and Language, 22, 266—291. Gardiner, J. M., & Gregg, V. H. (1979). When auditory memory is not overwritten. Journal of Verbal Learning and Verbal Behavior, 18, 705-719. Gathercole, S. E., Gregg, V. H., & Gardiner, J. M. (1983). Influences of delayed distraction on the modality effect in free recall. British Journal of Psychology, 74, 223-232. Geschwind, N. (1965). Disconnection syndromes in animals and man. Brain, 88, 237-294 and 585-644. Glanzer, M. (1972). Storage mechanisms in recall. In G. H. Bower (Ed.), The psychology of learning and motivation. Advances in research and theory (Vol. 5, pp. 129-193). New York: Academic Press. Glanzer, M., & Cunitz, A. R. (1966). Two storage mechanisms in free recall. Journal of Verbal Learning and Verbal Behavior, 5, 351-360. Glanzer, M., Koppenaal, L., & Nelson, R. (1972). Effects of relations between words on short-term storage and long-term storage. Journal of Verbal Learning and Verbal Behavior, 11, 403-416. Glenberg, A. M. (1984). A retrieval account of the long-term modality effect. Journal of Experimental Psychology, Learning, Memory and Cognition, 10, 16—31. Glenberg, A. M., & Swanson, N. G. (1986). A temporal distinctiveness theory of recency and

50

Shallice and Vallar

modality effects. Journal of Experimental Psychology, Learning, Memory and Cognition, 12, 3-15. Goodglass, H., & Kaplan, E. (1972). The assessment of aphasia and related disorders. Philadelphia: Lea & Febiger. Green, D. W. (1973). A psychological investigation into the memory and comprehension of sentences. Unpublished doctoral thesis. University of London. Green, E., & Howes, D. H. (1977). The nature of conduction aphasia: A study of anatomic and clinical features and of underlying mechanisms. In H. Whitaker & H. A. Whitaker (Eds.), Studies in neurolinguistics (Vol. 3). New York: Academic Press. Hebb, D. O. (1949). The organization of behavior. New York: Wiley. Heron, A., & Craik, F. I. M. (1964). Age differences in cumulative learning of meaningful and meaningless material. Scandinavian Journal of Psychology, 5, 209—217. Hinton, G. E., & Plaut, D. C. (1987). Using fast weights to deblur old memories. Paper presented at the Ninth Annual Conference of the Cognitive Science Society, Seattle. Hitch, G. J., & Halliday, M. S. (1983). Working memory in children. Philosophical Transactions of the Royal Society, London, B302, 325-340. Jarvella, R. J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Behavior, 10, 409-416. Jordan, M. I. (1986). Serial order: A parallel distributed approach. University of California San Diego Institute for Cognitive Science Report No. 8604. Kinsbourne, M. (1972). Behavioral analysis of the repetition deficit in conduction aphasia. Neurology, 22, 1126-1132. Kleist, K. (1916). Uber Leitungsaphasie und grammatische Storungen. Monatsschrift fur Psychiatrie und Neurologie, 40, 118-199. Levy, B. A. (1971). Role of articulation in auditory and visual short-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 123-132. Luria, A. R. (1977). Neuropsychological studies in aphasia. Amsterdam: Swets & Zeitlinger. McCarthy, R. A., & Warrington, E. K. (1984). A two-route model of speech production: Evidence from aphasia. Brain, 107, 463-485. McCarthy, R. A., & Warrington, E. K. (1987). Double dissociation of short-term memory for lists and sentences. Brain, 110, 1545-1563. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86. McClelland, J. L., & Kawamoto, A. H. (1986). Mechanisms of sentence processing: Assigning roles to constituents of sentences. In J. L. McClelland & D. Rumelhart (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 2, pp. 272-325). Cambridge, MA: MIT Press. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception. I. An account of the basic findings. Psychological Review, 88, 375407. McLeod, P. D., & Posner, M. I. (1984). Privileged loops from percept to act. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 55-66). London: Erlbaum. Margrain, S. A. (1967). Short-term memory as a function of input modality. Quarterly Journal of Experimental Psychology, 19, 109-114. Monsell, S. (1985). Repetition and the lexicon. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 12). London: Erlbaum. Morton, J. (1970). A functional model of memory. In D. A. Norman (Ed.), Models of human memory (pp. 203-254). New York: Academic Press. Morton, J., & Patterson, K. E. (1980). A new attempt at an interpretation, or, an attempt at a new interpretation. In M. Coltheart, K. E. Patterson, & J. C. Marshall (Eds.), Deep dyslexia (pp. 91-118) London: Routledge.


51

Murdock, B. B., Jr. (1967). Recent developments in short-term memory. British Journal of Psychology, 58, 421-433. Murray, D. J. (1968). Articulation and acoustic confusability in short-term memory. Journal of Experimental Psychology, 78, 679-684. Neisser, U. (1967). Cognitive psychology. New York: Appleton-Century-Crofts. Newell, A., & Simon, H. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Ojemann, G. A. (1983). Brain organization for language from the perspective of electrical stimulation mapping. Behavioral and Brain Sciences, 6, 189—230. Patterson, K. E., Coltheart, M , & Marshall, J. C. (1985). Surface dyslexia. London: Erlbaum. Peterson, L. R., & Johnson, S. T. (1971). Some effects of minimizing articulation on short-term retention. Journal of Verbal Learning and Verbal Behavior, 10, 346—354. Philipchalk, R., & Rowe, E. J. (1971). Sequential and nonsequential memory for verbal and nonverbal auditory stimuli. Journal of Experimental Psychology, 91, 341-343. Risse, G. L, Rubens, A. B., & Jordan, L. S. (1984). Disturbances of long-term memory in aphasic patients. Brain, 107, 605-617. Rowe, E. J. (1974). Ordered recall of sounds and words in short-term memory. Bulletin of the Psychonomic Society, 4, 559—561. Rugel, R. P. (1974). WISC subtest scores of disabled readers. Journal of Learning Disabilities, 7,

57-64. Rundus, D. (1971). Analysis of rehearsal processes in free recall. Journal of Experimental Psychology,

89, 63-77. Ryan, J. (1969). Grouping and short-term memory: Different means and patterns of grouping. Quarterly Journal of Experimental Psychology, 21, 137-147. Saffran, E. M v & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with deficient auditory short-term memory. Brain and Language, 2, 420—433. Salame, P., & Baddeley, A. (1982). Disruption of short-term memory by unattended speech: Implications for the structure of working memory. Journal of Verbal Learning and Verbal Behavior, 21, 150-164. Sartori, G. (1988). From neuropsychological data to theory and vice-versa. In G. Denes, C. Semenza, & P. Bisiacchi (Eds.), Perspectives in cognitive neuropsychology (pp. 59-73). London: Erlbaum. Schwartz, M. F. (1984). What the classical aphasia concepts don't do for us and why. Brain and Language, 21, 3-8. Shallice, T. (1975). On the contents of primary memory. In S. Dornic (Ed.), Attention and performance V (pp. 269-280). London: Academic Press. Shallice, T. (1979). Neuropsychological research and the fractionation of memory systems. In L.-G. Nilsson (Ed.), Perspectives on memory research (pp. 257-277). Hillsdale, NJ: Erlbaum. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729-735. Shallice, T., & McCarthy, R. (1985). Phonological reading: From patterns of impairment to possible procedures. In K. E. Patterson, M. Coltheart, & J. C. Marshall (Eds.), Surface dyslexia (pp. 361-397). London: Erlbaum. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261—273. Shallice, T., & Warrington, E. K. (1974). The dissociation between short-term retention of meaningful sounds and verbal material. Neuropsychologia, 12, 553-555. Shallice, T., & Warrington, E. K. (1977). Auditory-verbal short-term memory impairment and conduction aphasia. Brain and Language, 4, 479-491. Shallice, T., & Warrington, E. K. (1980a). K. F.: Post-mortem findings. In M. Coltheart, K. Patterson, & J. C. Marshall (Eds.) Deep dyslexia. London: Routledge.

52

Shallice and Vallar

Shallice, T., & Warrington, E. K. (1980b). Single and multiple component central dyslexic syndromes. In M. Coltheart, K. E. Patterson, & J. C. Marshall (Eds.), Deep dyslexia (pp. 119-145). London: Routledge. Sperling, G. (1963). A model for visual memory tasks. Human Factors, 5, 19—31. Sperling, G. (1967). Successive approximations to a model for short-term memory. Ada Psychologica, 27, 285-292. Sperling, G., & Speelman, R. G. (1970). Acoustic similarity and auditory short-term memory: Experiments and a model. In D. A. Norman (Ed.), Models of human memory (pp. 152—202). New York: Academic Press. Strub, R. L, & Gardner, H. (1974). The repetition defect in conduction aphasia: Mnestic or linguistic. Brain and Language, 1, 241-255. Touretzky, D., & Hinton, G, E. (1985). Symbols among the neurons: Details of a connectionist inference architecture. Proceedings of the Ninth International Joint Conference on Artificial Intelligence. Tulving, E. (1968). Theoretical issues in free recall. In T. R. Dixon & D. L. Horton (Eds.), Verbal behavior and general behavior theory. Englewood Cliffs, NJ: Prentice-Hall. Tzortzis, C, & Albert, M. L. (1974). Impairment of memory for sequences in conduction aphasia. Neuropsychologia, 12, 355-366. Vallar, G., & Baddeley, A. D. (1984a). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Baddeley, A. D. (1984b). Phonological short-term store: Phonological processing and sentence comprehenion. Cognitive Neuropsychology, 1, 121-141. Vallar, G., & Cappa, S. F. (1987). Articulation and verbal short-term memory: Evidence from anarthria. Cognitive Neuropsychology, 4, 55—78. Vallar, G., & Papagno, C. (1986). Phonological short-term store and the nature of the recency effect: Evidence from neuropsychology. Brain and Cognition, 5, 428-422. Vallar, G., Papagno, C, & Cappa, S. F. (1988). Latent dysphasia after left hemisphere lesions: A lexical-semantic and verbal memory deficit. Aphasiology, 2, 463-478. Vallar, G., & Perani, D. (1986). The anatomy of unilateral neglect after right-hemisphere stroke. A clinical/CT-scan correlation study in man. Neuropsychologia, 24, 609—622. Vallar, G., & Perani, D. (1987). The anatomy of spatial neglect in humans. In M. Jeannerod (Ed.), Neurophysiological and neuropsychological aspects of spatial neglect (pp. 235-258). Amsterdam: Elsevier Science Publishers. Warrington, E. K. (1979). Neuropsychological evidence for multiple memory systems. In Brain and Mind. Ciba Foundation Symposium 69 (new series) (pp. 153-166). Amsterdam: Excerpta Medica. Warrington, E. K. (1984). Recognition Memory Test. Windsor, Berks.: NEER-Nelson. Warrington, E. K., Logue, V., & Pratt, R. T. C. (1971). The anatomical localisation of selective impairment of auditory verbal short-term memory. Neuropsychologia, 9, 377-387. Warrington, E. K., & Rabin, P. (1971). Visual spaft of apprehension in patients with unilateral cerebral lesions. Quarterly Journal of Experimental Psychology, 23, 423-431. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory verbal short-term memory. Brain, 92, 885-896. Warrington, E. K., & Shallice, T. (1972). Neuropsychological evidence of visual storage in shortterm memory tasks. Quarterly Journal of Experimental Psychology, 24, 30-40. Watkins, M. J., & Peynircioglu, Z. F. (1983). Three recency effects at the same time. Journal of Verbal Learning and Verbal Behavior, 22, 375-384. Watkins, M. J., Watkins, O. C, & Crowder, R. G. (1974). The modality effect in free and serial recall as a function of phonological similarity. Journal of Verbal Learning and Verbal Behavior, 13, 430-447.


53

Watkins, O. G, & Watkins, M. J. (1977). Serial recall and the modality effect. Journal of Experimental Psychology: Human Learning and Memory, 3, 712—7IS. Watkins, O. G, & Watkins, M. J. (1980). The modality effect and echoic persistence. Journal of Experimental Psychology, General, 3, 251-278. Waugh, N. G, & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89-104. Wickelgren, W. A. (1965a). Short-term memory for phonemically similar lists. American Journal of Psychology, 78, 567-574. Wickelgren, W. A. (1965b). Short-term memory for repeated and nonrepeated items. Quarterly Journal of Experimental Psychology, 17, 14-25. Wickelgren, W. A. (1969). Auditory or articulatory coding in verbal short-term memory. Psychological Review, 76, 232-235. Zhang, G., & Simon, H. A. (1985). STM capacity for chinese words and idioms: Chunking and acoustical loop hypotheses. Memory and Cognition, 13, 193—201.

2. The development of the concept of working memory: implications and contributions of neuropsychology ALAN D. BADDELEY

2.1. Models of memory Research on short-term memory (STM) provides a particularly good example of the fruitful interaction of neuropsychology with techniques and theories developed in the study of normal memory. Since the majority of contributions to this volume will be concerned with data from patients, it was suggested that an overview of the field from the viewpoint of normal memory might be appropriate. This will be attempted, followed by a more detailed discussion of some of the issues that remain unresolved, and where further neuropsychological evidence might be particularly revealing.

2.1.1. How many kinds of memory? In his classic book The Organization of Behavior, Hebb proposed that memory comprised two separable systems, one based on temporary reverberating electrical activity, the other representing a more long-term change based on neural growth. Such a dichotomy became more widely supported in the late 1950s with the development of a range of techniques that appeared to indicate some kind of temporary storage where forgetting was rapid and was assumed to be based on trace decay (Broadbent, 1958; Brown, 1958; Peterson and Peterson, 1959). In the early 1960s, Melton (1963) argued that the assumption of a dichotomy was unnecessary and unparsimonious. He maintained that the phenomena attributed to short-term memory could better be conceptualized as reflecting the functioning of normal long-term memory (LTM) under conditions of brief presentation and minimal learning, with forgetting being based on the principles of interference theory. During the mid-1960s this led to a flurry of activity concerned with the question of whether it was necessary to assume separate long- and short-term memory systems. Evidence came from a number of sources, but the following three were perhaps the most prominent. 54

Development of the concept of working memory

55

(a) Two-component tasks. It was shown that a number of tasks appeared to be based on more than one component. The clearest example here is that of free recall in which a subject is presented with a list of unrelated items, and required to recall them in any order. On immediate recall, a clear recency effect occurs, with the last few items presented being particularly well recalled. If the subject is distracted for a few seconds and then allowed to recall, however, the recency effect disappears, whereas performance on earlier items is comparatively unaffected (Glanzer & Cunitz, 1966). One obvious interpretation of such results is to suggest that the most recent items are held in some temporary store, whereas the earlier items are registered in LTM. Further evidence for this comes from a range of studies showing that earlier items are influenced by a wide range of variables that influence long-term learning, such as word familiarity, rate of presentation, or number of rehearsals, whereas the recency effect is uninfluenced by these variables (Glanzer, 1972). A range of other tasks including minimal pairedassociate learning and the serial probe technique were also shown to have two separable components (see Baddeley, 1976, for a review). (b) Acoustic and semantic coding. Conrad (1964) showed that the immediate serial recall of visually presented consonant sequences gave rise to intrusion errors that were nonrandom. More specifically the errors were similar in sound or articulatory characteristics to the correct item, hence a subject was more likely to misremember B as V than to misremember it as something visually similar such as R. Conrad and Hull (1964) showed that sequences of consonants that were similar in sound (e.g., B T G V C P) were consistently harder to recall than phonologically dissimilar sequences (e.g., R W K Y Q N). Baddeley (1966a) contrasted phonological similarity with similarity of meaning, observing that immediate serial recall of a phonologically similar sequence of words such as man, cad, map, mad, cat was consistently harder than a dissimilar control list, whereas a sequence that was similar in meaning (e.g., huge, large, big, tall, long) created few problems in an immediate recall paradigm, suggesting that subjects in this task were coding the words on the basis of their phonological characteristics. In contrast, when the task was switched to the long-term serial learning of lists of 10 words presented over several trials, the pattern reversed, with similarity of meaning being important and phonological similarity having little or no effect (Baddeley, 1966b). A simple interpretation of these results was to suggest that the STM system relied on a phonological code, whereas LTM favoured semantic coding. Evidence that seemed to favour this view came from studies using the sequential probe technique (Kintsch & Buschke, 1969), and from memory for prose, where the literal surface structure of a sentence appeared to be held for a brief period of time, but was then lost, in contrast to semantically based information, which appeared to be much more durable (Sachs, 1967).

56

Baddeley

(c) Neuropsychological evidence. The third main cluster of evidence came from work on neuropsychological patients. It has been known for many years (e.g., Zangwill, 1946) that alcoholic Korsakoff patients with grossly impaired long-term learning capacity might have normal digit span. The evidence for normal immediate memory coupled with dense amnesia was demonstrated very clearly by the classic amnesic patient HM (Milner, 1966). Baddeley and Warrington (1970) tested amnesic patients on a range of tasks that were assumed differentially to reflect long- and short-term memory components, obtaining results that were broadly consistent with the assumed separation. Hence their patients showed normal performance on immediate memory span, the Peterson short-term forgetting task, and on recency in free recall, while showing grossly impaired performance on earlier free recall items, and on standard long-term learning tasks. Further neuropsychological evidence for the distinction was produced by Shallice and Warrington (1970), who studied a patient with the opposite pattern of deficits, namely, impaired performance on immediate memory for spoken sequences, on the Peterson task, and on recency in free recall, coupled with normal long-term learning.

2.1.2. The modal model By the late 1960s, the evidence seemed to be veering strongly in the direction of dichotomous theories of memory. The area was a lively one, with many competing models, often worked out in some mathematical detail, but all tending to have many features in common. The most influential of these was the model of Atkinson and Shiffrin (1968), which subsequently became known as the modal model. The modal model assumed three memory systems, a bank of sensory memories that operated in parallel, feeding information into a limited capacity short-term store that in turn fed information into and out of a long-term store. The short-term store within this model plays a crucial role, since it is necessary for both learning and retrieval. At a superficial level at least, the model was consistent with the pattern of evidence in favour of the STM-LTM distinction. The STM component of tasks such as free recall and immediate memory span were assumed to be based on the operation of the short-term store, whereas long-term learning depended on the long-term store. Phonological coding could be assumed to characterize much of the activity of the short-term storage system, with semantic factors dominating within long-term storage. Finally amnesic patients could be assumed to have a deficit in long-term storage and STM patients to have an impairment in the short-term store.

Problems with the modal model

By the early 1970s, the modal model was beginning to run into difficulties. One of the most prominent of these stemmed from data on STM patients. If such patients were


57

assumed to have an impairment in the capacity of their short-term store, and this was essential for long-term learning and retrieval, then such patients should have LTM problems. Shallice and Warrington (1970) found no evidence for this. Other problems began to crop up in the learning assumptions made by the modal model, namely, that the probability of long-term storage was a direct function of how long an item was maintained in STS. Craik and Watkins systematically varied time in STS in an incidental learning paradigm, and found no evidence for such a relationship (Craik & Watkins, 1973). Third, the standard interpretation of recency as reflecting residence in the STS system was challenged by Tzeng's (1973) demonstration of delayed recency effects under conditions where items should have been cleared from the STS, and other studies such as that of the Baddeley and Hitch (1974) showed clear recency in LTM. For example, the capacity of rugby players to recall games they had played showed a recency effect that obeyed the same general principles as recency in immediate free recall, but extended over a period of weeks. Interest in the modal model began to wane, with the amount of work on STM decreasing sharply in the 1970s. At a theoretical level, two developments occurred: the proposal by Craik and Lockhart (1972) of the levels-of-processing approach to memory, and the development of the concept of working memory by Baddeley and Hitch (1974).

2.1.3. Levels of processing Craik and Lockhart (1972) suggested that rather than treat long- and short-term memory as separate structures operating on the basis of different types of code, it would be more fruitful to interpret the differential durability of memory traces as a simple result of differential coding. They suggested that an incoming stimulus such as a printed word would be processed sequentially at a range of different levels, starting with a superficial visual encoding from which would be derived the phonological characteristics of the word, after which the word meaning would be extracted. Craik and Lockhart suggested that the durability of the memory trace was a direct function of the depth of processing involved. They cited clear evidence that when subjects in an incidental learning task were induced to process words in terms of their visual characteristics, subsequent recall was poorer than in the case of words processed phonologically. This in turn led to poorer recall and recognition than did semantic processing. Craik and Lockhart still assumed a dichotomous view of memory, with the successive levels of processing being dependent on the operation of a primary memory system. In this respect, levels of processing is a theory concerned with the relationship between manner of coding and long-term learning, leaving the details of the short-term or primary memory system as a separate issue. Nevertheless, levels of processing was seen

58

Baddeley

by many as an alternative to a dichotomous view of memory (e.g., Postman, 1975), a view that tended to further discourage research on STM.

2.1 A. Working memory A second alternative to the modal model was developed by Graham Hitch and myself as a result of an attempt to answer the question of what role was played by the shortterm store in general cognition. We tested the hypothesis that it operated as a working memory, a system for temporarily storing and manipulating information in the execution of complex cognitive tasks such as learning, reasoning, and comprehension. We argued that if STM were necessary for these, then it should be possible to disrupt a subject's performance on such tasks by absorbing STM capacity by means of a secondary task. We made the assumption that STM was limited in capacity, and was the limiting factor in performing the digit span task. We reasoned on the basis of this that if a subject were required to hold and rehearse a sequence of digits, then this should reduce to a minimum the capacity of STM remaining for learning, reasoning, or comprehending, and should dramatically disrupt performance. Another way of conceptualizing this procedure is that we were attempting to make our normal subjects functionally equivalent to STM patients; the patients had the STM system disrupted through brain damage, whereas our subjects had it disrupted by a demanding concurrent memory span task. Across a range of tasks, the results were broadly comparable. A concurrent digit span of six items led to an impairment in learning, reasoning, and comprehension that was clear but not nearly as dramatic as would be expected by a theory assuming a unitary limited-capacity STM system, such as was suggested by the modal model. Even more problematic for the modal model was the observation that concurrent digit span had no effect on the recency component in free recall. According to the modal model both recency and digit span should have been dependent on the same limitedcapacity system, and hence should have led to mutual massive interference. It had no effect. In response to this and other evidence, we proposed that the concept of a unitary STM system should be replaced by the concept of a multicomponent working memory. We proposed a model in which an attentional control system, the central executive, coordinated information from a number of subsidiary slave systems (Baddeley & Hitch, 1974). Two active subsystems were postulated - the articulatory loop, a system involved in the maintenance of speech-based information, and the visuospatial scratch pad or sketch pad, a system responsible for setting up and maintaining visuospatial images. The sketch pad was shown to be capable of setting up and maintaining temporary visuospatial representations, and to be disrupted by concurrent spatial activity such as


59

pursuit tracking. It was shown to be involved in the setting up and utilizing of visual imagery mnemonics such as those involved in pegword or location mnemonics. The system was not, however, responsible for the advantage enjoyed by concrete or imageable material in long-term learning, an effect that was interpreted in terms of the richer representation of imageable material in LTM (Baddeley, Grant, Wight, & Thomson, 1975; Baddeley & Lieberman, 1980). Neuropsychological evidence for a visuospatial sketch pad system is beginning to develop (e.g., Farah, 1988) but will not be further discussed here, since the emphasis of the present volume is on patients with short-term phonological rather than visuospatial deficits. For a more detailed discussion of this, see Baddeley (1986, chapter 6) and Farah (1988). 2.2. The articulatory loop The importance of speech coding within working memory was assumed to be based on the operation of the articulatory loop subsystem. This is assumed to comprise a phonological store within which memory traces will fade if not revived within 1-2 sec, supplemented by an articulatory control process. This serves two functions: First, it maintains memory traces within the store by means of subvocal rehearsal; and second, it allows visually presented items to be fed into the store, provided they are capable of being encoded phonologically and subvocalized. This combination of a phonological store and an articulatory control process was able to account for a rich pattern of results. These included the following. (a) The phonological similarity effect (Conrad, 1964; Baddeley, 1966a). Items that are phonologically similar are assumed to have similar and hence confusable codes within the phonological store. The greater the similarity, the greater the difficulty of trace discrimination at retrieval. (b) The word length effect. Baddeley, Thomson, and Buchanan (1975) showed that the immediate serial recall of word sequences decreased as the constituent words became longer. The crucial feature proved to be spoken duration and not number of syllables, since disyllabic words that have a long spoken duration such as Friday and harpoon led to consistently poorer serial recall than quickly spoken disyllabic words such as wicket and bishop. Furthermore, there was a correlation between the speed at which a subject articulates and his or her memory span, a result that will be explored in more detail in the chapter by Graham Hitch (this volume, chapter 9). In general, results in this area indicate that a subject's memory span is determined by the amount he or she can articulate in about 2 sec. (c) The unattended speech effect (Colle & Welsh, 1976; Salame & Baddeley, 1982). Immediate serial recall of visually presented digits is impaired when presentation

60

Baddeley

and/or recall are accompanied by spoken material the subject is free to ignore. Important characteristics of the unattended material appear to be phonological, with digits being disrupted by other monosyllabic words more than by disyllables, while semantic factors appear to be comparatively unimportant. Hence unattended digits are no more disruptive than nondigit words made up from the same phonemes (e.g., tun, woo instead of one, two); similarly, nonsense syllables, or words spoken in an unfamiliar foreign language, are just as disruptive as meaningful words. Finally, sound intensity does not appear to be an important variable, but vocal characteristics are, with noise having little or no influence on performance in contrast to speech, with unattended orchestral music having an intermediate effect (Colle, 1980; Salame & Baddeley, 1987). (A) Articulatory suppression. When vocal rehearsal is prevented by requiring the subject to articulate an irrelevant sound such as repeating the word the, then immediate serial recall is markedly impaired. Furthermore, with visual presentation, articulatory suppression removes the effect of phonological similarity, word length, and unattended speech. With auditory presentation, the phonological similarity effect is observed, although the effect of word length is not (Baddeley, Lewis, & Vallar, 1984). The simple articulatory loop model explains this pattern of results as follows: Phonological similarity impairs performance because the store is coded phonologically; hence similar items have codes that are less discriminable and more subject to error at retrieval. The word length effect occurs because the rehearsal mechanism is based on the time it takes to articulate the material; consequently, the rate at which the memory trace of a sequence of long words can be refreshed is lower than that for short words, leading to a lower ceiling on the maximum number of words that can be maintained through rehearsal. The unattended speech effect is assumed to occur because spoken material has obligatory access to the phonological store, causing the memory trace of the wanted items to be corrupted by the reading in of unattended, unwanted material. The semantic characteristics of such material is unimportant, since it is a phonological store, which does not encode semantic information. Articulatory suppression interacts with these various effects in somewhat different ways. In the case of phonological similarity, suppression will remove its effect when presentation is visual, since suppression interferes with the feeding of the visually presented material into the phonological store. In the case of auditory presentation, however, the material gains direct access to the phonological store without needing to rely on subvocal articulation; hence with auditory presentation, suppression does not remove the phonological similarity effect. A similar pattern holds in the case of the unattended speech effect. When the material to be remembered is presented visually, then suppressing articulation prevents the material from being registered in the phonological store; since the store is no longer of any assistance in performing the memory task, the fact that it is being corrupted by


61

unattended speech becomes irrelevant. On the other hand, if the material to be recalled were presented auditorily, then one would expect unattended speech to have an influence on performance regardless of whether or not the subject was suppressing articulation, as indeed it does (Hanley & Broadbent, 1987). The effect of articulatory suppression on the word length effect is, however, somewhat different. Since the influence of word length operates through the process of articulation itself, preventing subvocal rehearsal by suppression should remove the word length effect regardless of whether presentation is visual or auditory. Provided suppression occurs during both input and recall, then this result is indeed obtained (Baddeley et al., 1984). The assumption of a phonological store fed by an articulatory control process is therefore able to give a simple explanation of a relatively rich pattern of results. To what extent is it capable of explaining the memory performance of STM patients?

2.2.1. STM deficits and the articulatory loop At a general level, it was clear that the articulatory loop hypothesis was consistent with the findings of Shallice and Warrington (1970), in a way that was certainly not true of the modal model. In particular, the apparent paradox of impaired STM performance and normal LTM could be explained by assuming that the deficit was limited to one component of working memory, leaving other aspects, including the crucial central executive, unimpaired. Furthermore, data from the investigation of the unattended speech effect (Salame & Baddeley, 1982) argued for regarding the phonological component of the articulatory loop as an input store, a position close to that suggested by Shallice and Warrington. A more detailed exploration of the articulatory loop interpretation of an STM patient was, however, carried out by Vallar and myself (Vallar & Baddeley, 1984a). The patient, PV, is described elsewhere (see Shallice and Vallar this volume, chapter 1). She had a very pure STM deficit, with an auditory digit span of two items, coupled with normal long-term memory as measured by paired-associate learning or free recall of word lists, or by prose recall (Basso, Spinnler, Vallar, & Zanobio, 1982). We decided to explore the characteristics of PV's articulatory loop system using the standard variables of phonological similarity, word length, and articulatory suppression. Similarity proved to impair performance when presentation was auditory, but not when the material was presented visually. She showed no effect of word length, and her performance was unimpaired by concurrent articulatory suppression. PV's capacity to articulate rapidly appeared to be unimpaired, as measured by counting rate or speed of reciting the alphabet. We interpreted this pattern of results as consistent with the assumption of an impairment in the phonological input store, limiting its capacity to only two items.

62

Baddeley

When presentation was auditory, then the store would be used, as evidenced by the presence of a phonological similarity effect, but performance would be low. With visual presentation, an alternative system was presumably used, possibly based on some kind of visual coding. The optional strategy of recoding visual material phonologically was not adopted, as evidenced by the lack of a phonological similarity effect with visual presentation, and the absence of effects of either word length or articulatory suppression. We presume that PV does not use a recoding strategy, since it would merely serve to feed information into a defective store, a procedure that would be unlikely to enhance performance. On the basis of PV's normal speech output, and her unimpaired capacity to count and recite the alphabet at speed, we assume that her articulatory control process is unimpaired. It is of course conceivable that inner speech and overt speech involve quite separate systems, but in the absence of any clear evidence for this, we would regard such an interpretation as unjustified and unparsimonious. In conclusion, then, our results are consistent with the assumption that PV retains the use of the phonological store, although its capacity is substantially reduced. We obtained no evidence to suggest that her capacity for articulatory rehearsal was similarly impaired. Needless to say, it is likely that other patients may have deficits to other components of the system, giving rise to a somewhat different pattern of deficits. It would certainly be interesting to observe the performance of a wider range of STM patients on the tasks used to explore the articulatory loop. (See Shallice and Vallar, this volume, chapter 1, for a discussion of these issues.)

2.3. Dysarthria and the functioning of the articulatory loop The process of articulation is itself quite complex, presumably involving the setting up of motor programmes, their temporary storage, and subsequently their realization through commands to the articulators ultimately resulting in speech output. A study by Wilson and myself explored the role of the motor component in inner speech by analysing the memory performance of an anarthric patient (Baddeley & Wilson, 1985). The patient in question had lost the capacity to control the motor output of the speech system as a result of what was presumed to be a brain-stem lesion following a traffic accident. He showed no signs of aphasia, however, and could communicate grammatically and fluently by means of a keyboard device. We studied the functioning of his articulatory loop system, and found it to be apparently quite normal, with a digit span of six, clear effects of phonological similarity and word length, and an unimpaired capacity for making rhyme judgments on visually presented material. A broadly similar pattern of results has been shown by Vallar and Cappa (1987) and by Logie, Cubelli, Delia Sala, Alberoni, and Nichelli, (in press), who review this area and discuss some of the detailed variations between patients that have subsequently been studied.


63

An even more striking result was obtained by Bishop and Robson (1989), who studied the memory performance of congenially anarthric children. Again they appeared to have quite normal development of the articulatory loop system, a reasonable memory span, and clear evidence of both phonological similarity and word length effects. What are the implications of these results? First of all they indicate that the articulatory loop system does not depend on feedback from overt activity of the speech musculature for its operation. Nor is overt articulation necessary for developing the articulatory loop. Does this therefore mean that internal speech is quite unconnected to overt speech? I suspect that this is an overinterpretation of the data, which can equally well be interpreted on the assumption that the motor programme for setting up and running speech can operate at a very high level independently of its physical realization. Is it not implausible to assume that an articulatory loop would be set up without overt feedback? I think not, provided one makes the assumption of a built-in mechanism whereby, when a developing child hears a speech sound, there will be a tendency for a motor programme to be set up that in a normal child would allow that sound to be echoed back. Such an automatic repetition process could be an important component of learning to speak. There is of course evidence that in adults the repetition of a heard sound is a highly compatible response. Davis, Moray, and Treisman (1961) showed that such a repetition response is very rapid and unaffected by number of alternatives, while McLeod and Posner (1984) have also shown that such vocal repetition responses appear to have a special status, placing minimal load on a subject's concurrent processing capacity. The nature of the articulatory rehearsal process does, however, clearly require further explanation. Bishop and Robson (1989), for example, have suggested that the process of rehearsal may better be regarded as the scanning of a series of representations in long-term memory rather than the active running off of motor programmes. We are currently considering ways in which it might be possible to decide between these two options experimentally. Another important potential line of development is to study the memory performance of patients with less peripheral disruption in articulation. For example, the articulatory loop model would predict that dyspraxic patients would have impaired memory span due to the defective functioning of the articulatory control process. At a more general level, it would seem well worth exploring the implications of various types of aphasia for the functioning of the articulatory loop system. Initial work by Ostergaard and Meudell (1984) looks promising, and suggests that memory performance may in due course play a useful role in the diagnosis and analysis of aphasia.

64

Baddeley

2.4. Problems and growing points 2.4.1. Learning to read The concept of working memory is primarily concerned with the relationship between memory and cognitive performance. As such, the articulatory loop concept has been applied to the analysis of a range of cognitive skills. One of the more striking characteristics of children with specific reading disability is their tendency to have impaired memory span. This has led to the suggestion that an articulatory loop deficit may lie at the root of much developmental dyslexia. One possibility was that dyslexic children might simply not use the articulatory loop system in attempting to decode the printed word. Some evidence for this appeared to be provided by a range of studies that showed poor readers to be less influenced by the phonological similarity in immediate memory than were normal or good readers (e.g., Liberman, Mann, Shankweiler, & Werfelman, 1982). Unfortunately, this result has not always proved replicable, leading to a good deal of controversy as to whether or not poor readers are influenced by phonological similarity (see Baddeley, 1986, chapter 9, for a review). Data from Ellis, Baddeley, and Miles (see Baddeley, 1986, Table 9.2, p. 208) suggest that a sample of dyslexic boys were impaired in their overall memory span, but showed every evidence of using the articulatory loop system as indicated by the effects of phonological similarity, word length, and articulatory suppression. A possible resolution of this discrepancy was suggested by Hall, Wilson, Humphreys, Tinzmann, and Bowyer (1983), who proposed that the effect of phonological similarity might disappear when the memory span is grossly overloaded. Since dyslexics typically have a shorter span, and controls and dyslexics are typically tested at the same sequence length, this would account for the weakening of the phonological similarity effect. Evidence for the reduction of the effect of phonological similarity and indeed of phonological coding when span is grossly increased has been obtained in normal adult subjects (Salame & Baddeley, 1986). Finally Johnston (1982) has explored Hall et al/s interpretation directly and has demonstrated that sequence length is indeed a crucial variable, with dyslexic children showing clear effects of phonological similarity with relatively short sequences; the effect is lost, however, when the sequence length grossly exceeds their span. A second point of major controversy in this area concerns the question of the nature of the phonological deficit. One suggestion is that phonological awareness is the crucial factor (e.g., Bradley & Bryant, 1983), while an alternative possibility is that most measures of phonological awareness require the storage of the phonological information, hence making the tests an indirect measure of phonological storage. A related problem concerns the direction of causality. Does reading depend on phonological awareness, or does phonological awareness improve as a result of practice


65

in reading? Evidence for the latter comes from a series of studies using adults who are illiterate as a result of lack of the opportunity to learn to read. The evidence indicates that such subjects initially show poor performance on tests of phonological awareness, but improve as their reading improves (Morais, Alegria, & Content 1987). This suggests then that phonological awareness improves as a result of reading rather than the reverse. A broadly similar picture comes from a study by Ellis and Large (1987) in which they followed a group of young children through their first few years at school, measuring reading performance and performance on a range of tests of intelligence, phonological coding, memory, and phonological awareness. Ellis and Large find reasonably high correlations between phonological processing measures and reading, but in general observe that the reading pattern predicts subsequent phonological performance rather better than the reverse. It is almost certainly the case, therefore, that learning to read and performance on memory and phonological processing tasks are mutually supportive. It is also the case, however, that some children have great difficulty in starting to learn to read, and it seems likely that such children begin with some form of phonological deficit. Such a view is supported by a study carried out by Susan Gathercole and myself in which we explored the memory performance of a group of children who had been separated from their peers for special education, as having specific delayed language development, with otherwise normal intelligence (Gathercole & Baddeley, 1987). We observed the expected deficits in short-term verbal memory, and in particular found that the measure on which they showed the greatest impairment was a task requiring the repetition of nonwords, effectively an immediate memory span for unfamiliar phonological material. We are currently exploring the possibility that this measure will enable us to predict in advance which children are likely to have difficulties in learning to read. We have tested a sample of about 150 children when they began school, and are following them up at annual intervals, observing the development of memory performance and that of vocabulary and reading skills. A more detailed analysis of the performance of the delayed-language children indicated that their deficit was probably one of storage rather than speed of retrieval or articulation. The evidence that such children also have delayed vocabulary development suggests the interesting possibility that the articulatory loop system may be an important factor in the acquisition of vocabulary, a point that will be returned to later in discussing phonological long-term learning.

2.4.2. Phonological coding and fluent reading The issue of whether a phonological code plays an important role in the reading comprehension of fluent readers is an ancient one, extending back at least to the work of

66

Baddeley

Huey (1908). Evidence seems to suggest that articulatory suppression, for example, will impair performance on certain types of reading tasks, particularly those involving the detection of errors of word order (Baddeley, Eldridge, & Lewis, 1981), but that the impairment in comprehension under these conditions is far from massive (see Baddeley, 1986, chapter 8 for a review). A related problem concerns the question of whether articulatory suppression prevents the phonological encoding of the printed word. There appear to be conflicting results in this area with some (e.g., Baddeley & Lewis, 1981) finding little or no impairment in phonological processing, whereas others (e.g., Kleiman, 1975) claim substantial disruption of phonological processing. A recent article by Besner (1987) argues strongly that the data can be readily explained if a differentiation is made between those studies in which judgments of homophony were required (e.g., do the following sound the same, key-quayl) and those in which the item must be processed in some way, as, for example, in deciding whether two items rhyme or not, where the initial sounds must be deleted before the comparison is made. It appears to be the case that judgments of homophony are unimpaired by articulatory suppression but that suppression does interfere with rhyme judgments, or tasks that involve the storage and manipulation of phonological information (Besner, 1987).

2.4.3. Language comprehension in STM patients Exploration of the role of the articulatory loop in the reading of normal subjects has relied rather heavily on the single technique of articulatory suppression. Data from patients suffering from a deficit in the system clearly provide a potentially important source of supplementary information. Since this topic will no doubt be discussed at length elsewhere, I will not go into great detail, but would like briefly to mention some of our own data and also discuss an apparently conflicting source of information from a study by Butterworth, Campbell, and Howard (1986) of a subject with a developmental deficit in STM performance. Our own studies concern two patients, the previously described STM patient PV, and TB, a patient with a slightly less pure but more serious impairment in immediate memory performance. We have carried out two investigations into PV's language comprehension (Vallar & Baddeley, 1984b, 1987). Our results show that PV does have comprehension problems, but that these become clear only when material is selected that places a particularly heavy load on working memory. A simple increase in sentence length is not sufficient to create difficulties. Hence she can correctly verify both short sentences such as Slippers are sold in pairs, and verbose and lengthy equivalents such as It is commonly believed, and with some justification, that slippers fall into the category of objects that are normally sold in pairs. She does, however, show difficulty in processing sentences where comprehension requires the maintenance of literal information across several intervening words. Hence


67

she is able to verify accurately a short sentence such as Rivers are crossed by bridges, but not an equivalent sentence in a lengthier form such as It is fortunate that most rivers are able to be crossed by bridges that are strong enough for cars. Similarly, she is capable of

making correct judgments of anaphoric match or mismatch when the two components are reasonably close, but begins to make errors when a substantial number of words separate the anaphoric constituents. On the basis of PV's data one could draw either of two conclusions. The first possibility is that the articulatory loop system plays a role in comprehension, but is needed only for relatively demanding material. The second possibility is that the articulatory loop system is probably needed for virtually all material, but that the level of functioning of PV's phonological store is still sufficient to allow her to cope with most material. PV has a sentence span of about six words, and it is plausible to assume that a mnemonic "window" of six words may be enough to allow most material to be comprehended reasonably accurately. An opportunity of deciding between these alternatives was presented by a second case, TB, who has a sentence span of only three words. Like PV, he is intellectually relatively unimpaired, although, unlike PV, he shows an impairment in span when material is presented visually as well as auditorily. In addition, he has some long-term learning deficits, and for this reason we compared his comprehension performance with that of a pure amnesic patient who has an LTM deficit together with normal STM performance (Baddeley, Vallar, & Wilson, 1987). TB was capable of verifying simple sentences such as Slippers are sold in pairs but was quite unable to cope with the more verbose versions of such sentences. In general, his comprehension of spoken material was considerably more impaired than that of PV, with the probability of comprehension systematically increasing with sentence length. In an attempt to deconfound the effects of syntactic complexity and length, we compared his comprehension with auditory and visual presentation, arguing that the written version might provide memory support that will help offset his STM deficit. There was a consistent tendency for printed sentences to be verified more accurately than spoken ones, although this often involved very long latencies during which TB appeared to be hunting to and fro across the sentence, as if trying to solve a verbal jigsaw puzzle. A second attempt to separate the syntactic complexity from the memory overload hypothesis involved taking sentences that TB could understand and then lengthening them by adding supplementary adjectives and adverbs, adding syntactic constructions that we knew he was capable of comprehending. For example, The girl chases the horse was extended to The little girl vigorously chases the poor old horse. Under these conditions, performance dropped from 21/24 correct to a chance level of 6/24. We interpret these results as suggesting that the phonological store component of the articulatory loop acts as a kind of mnemonic window, holding word sequences from the sentence simultaneously, and allowing the subject to decode these into the

68

Baddeley

constituent meaning. We would argue that our results suggest that an impaired phonological store will reduce the subject's capacity to perform this type of analysis, and hence will interfere with language comprehension. Although others have drawn similar conclusions about the importance of short-term phonological storage in comprehension (e.g., Saffran & Marin, 1975; Caramazza, Basili, Koller, & Berndt, 1981) an opposite conclusion has been drawn by Butterworth et al. (1986) on the basis of the analysis of comprehension in a subject with a developmental impairment in immediate memory span. Despite having a reduced digit span, this subject was able to perform well on a range of language comprehension tests. Butterworth et al. conclude on the basis of this that STM does not play an important role in comprehension. There are a number of reasons for not accepting this conclusion. First of all, the case in question suffers from a developmental deficit, and by the time she was tested she had many years to learn to cope with the problem, either as a result of a possible adaptation at a neural level, or by developing alternative strategies. Evidence for the latter comes from two sources. First of all, her reading performance was that of a phonological dyslexic; she could read words, but had great difficulty in reading nonwords. This suggests that she has probably learned to read by mapping the visual pattern directly onto meaning, rather than by the normal process of phonological mediation, which presumably stems from the fact that her phonological system is in some way deficient. Evidence that her method of comprehension is also atypical comes from an experiment in which the Token Test was performed with or without articulatory suppression. Under nonsuppression conditions, the subject was not different from controls. When articulation was suppressed, however, controls showed an impairment in performance, whereas this subject did not, showing in fact significantly better performance than was shown by control subjects. A third problem concerns the magnitude of the memory deficit shown by this subject. For our hypothesis, the crucial measure would be sentence span. This is not quoted, although performance on a task that would seem to depend on sentence span indicates a remarkably substantial span, somewhere between 10 and 20 words. Overall then, the evidence seems to suggest that the phonological store does play an important role in the comprehension of spoken, and to a lesser extent of written, discourse. Our data suggest that a substantial impairment in the capacity of the store will lead to problems in comprehending material, particularly when this requires the integration of surface information across several intervening words or phrases.

2.4.4. The articulatory loop and long-term phonological learning As we saw earlier, impaired articulatory loop performance tends to be associated with developmental dyslexia. Why? One possibility is that the process of reading unfamiliar


69

words involves systematically decoding each letter, storing the resultant sound until the end of the word is reached, and then blending the sounds (Baddeley, 1979). I suspect that this is part of the reason, but it is probably not the whole story. It would not explain, for example, why developmental dyslexics tend to have a reduced vocabulary, or apparently have difficulty during the early stages of learning to read, before the development of word attack skills. Similarly it does not explain why such children tend to have impaired long-term learning, for example, of nonsense syllables (Torgeson, Rashotte, & Greenstein, 1985), or of multiplication tables (Miles & Ellis, 1981). This suggested the possibility that the deficit in STM patients might apply to long-term phonological learning, as well as to short-term phonological storage. We decided to explore this by looking at phonological learning in our pure STM patient PV (Baddeley, Papagno, & Vallar, 1988). We chose as our task the learning of vocabulary in an unfamiliar language, comparing PV's performance with that of a group of 14 subjects matched for age and educational background. Our first experiment checked PV's capacity for learning meaningful material by requiring her to master a list of pairs of meaningful words. Her learning performance on this task was quite equivalent to that of the controls. We then moved on to a task in which she learned to associate an unfamiliar word, based on the transliteration of a Russian word, with a familiar Italian word. She might, for example, learn to associate the unfamiliar word svieti with the familiar word rosa. Material was presented either rapidly, at a rate of 2 sec per pair, or at a slower rate of 5 sec per pair, and presentation was either auditory or visual. Under conditions of auditory presentation, the control subjects took about 10 trials to learn the list of eight pairs. By the end of the 10th trial, PV had not learned a single pair. With visual presentation, her performance was substantially better, but still significantly worse than the control group, indicating that visual coding was helpful, but was not sufficient to make up for the impairment in her phonological store. We regard this as a very important result for a number of reasons. First of all, it suggests that the phonological short-term store plays an important role in phonological learning, and hence presumably in the acquisition of a child's first language. It thus seems likely that the articulatory loop system evolved because of its role both in acquiring language and in supporting language comprehension. It presumably developed an even greater importance with the evolution of literacy, making the articulatory loop system an important precursor of learning to read using an alphabetic script. A second question concerns how the deficit in long-term phonological learning occurs. One possibility is that the phonological store holds the incoming material, hence allowing the long-term system better access to it. Another possibility, however, is that both short- and long-term phonological storage may be based on the operation of the same basic system. Recently, in exploring and developing parallel distributed models of memory, Hinton and Plaut (in press) have suggested that it may be particularly valuable

70

Baddeley

in devising a learning system to allow it to operate at two levels. One of these levels involves the gradual building up of long-term learning through a change in the relative weighting of connections between units. These involve a slow but stable process. In addition, Hinton suggests the advantage of having a system of fast weights, whereby connections can be set up more rapidly and allowed to dissipate more rapidly. It would be interesting to explore the possibility that long- and short-term phonological coding might perhaps represent a dual system of weights within a unitary phonological memory system, with STM patients showing a disruption of the underlying fast-weight system leading to an impairment in both STM and long-term learning.

2.5. Conclusion The last 30 years has seen a gradual development and evolution of our concept of shortterm memory, an evolution that has been strongly influenced by neuropsychological evidence from patients with STM deficits. One of the clearest themes to emerge has been the relationship between short-term memory and speech coding, and it is this aspect of working memory that has been developed most extensively and linked most clearly with evidence from neuropsychology. I believe we have started to approach a point at which it might be possible and valuable to come up with more detailed models of this component of working memory. In the meantime, it seems likely that other aspects of working memory, including the operation of the visuospatial sketch pad (Farah, 1988) and of the central executive (Baddeley, Logie, Bressi, Delia Sala, & Spinnler, 1986), will gain a similar advantage from the confrontation of models from the psychological laboratory with data from the neurological clinic.

References Atkinson, R. C, & ShifiFrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89-195). New York: Academic Press. Baddeley, A. D. (1966a). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 362—365. Baddeley, A. D. (1966b). The influence of acoustic and semantic similarity on long-term memory for word sequences. Quarterly Journal of Experimental Psychology, 18, 302-309. Baddeley, A. D. (1976). The psychology of memory. New York: Basic Books. Baddeley, A. D. (1979). Working memory and reading. In P. A. Kolers, M. E. Wrolstad, & H. Bouma, (Eds.), Processing visible language (pp. 355—370). New York: Plenum Press. Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press. Baddeley, A. D., Eldridge, M., & Lewis, V. J. (1981). The role of subvocalization in reading. Quarterly Journal of Experimental Psychology, 33, 439—454. Baddeley, A. D., Grant, W., Wight, E., & Thomson, N. (1975). Imagery and visual working memory. In P. M. A. Rabbitt & S. Dornic (Eds.), Attention and performance V (pp. 205-217). London: Academic Press.


71

Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. Bower (Ed.), Recent advances in learning and motivation (Vol. 8, pp. 47-90). New York: Academic Press. Baddeley, A. D., & Lewis, V. J. (1981). Inner active processes in reading: The inner voice, the inner ear and the inner eye. In A. M. Lesgold & C. A. Perfetti (Eds.), Interactive Processes in Reading (pp. 107-129). Hillsdale, NJ: Erlbaum. Baddeley, A. D., Lewis, V. J., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly Journal of Experimental Psychology, 36, 233-252. Baddeley, A. D., & Lieberman, K. (1980). Spatial working memory. In R. Nickerson (Ed.), Attention and performance VIII (pp. 521-539). Hillsdale, NJ: Erlbaum. Baddeley, A. D., Logie, R., Bressi, S., Delia Sala, S., & Spinnler, H. (1986). Dementia and working memory. Quarterly Journal of Experimental Psychology, 38A, 603-618. Baddeley, A. D., Papagno, G, & Vallar, G. 1988). When long-term learning depends on shortterm storage. Journal of Memory and Language, 27, 5S6-595. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of shortterm memory. Journal of Verbal Learning and Verbal Behavior, 14, 57'5-589. Baddeley, A. D., Vallar, G., & Wilson, B. A. (1987). Sentence comprehension and phonological memory: Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and performance XII: The psychology of reading (pp. 509-529). London: Erlbaum. Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and the distinction between long- and short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176—189. Baddeley, A. D., & Wilson, B. (1985). Phonological coding and short-term memory in patients without speech. Journal of Memory and Language, 24, 490-502. Basso, A., Spinnler, H., Vallar, G., & Zanobio, E. (1982). Left hemisphere damage and selected impairment of auditory verbal short-term memory: A case study. Neuropsychologia, 20, 263-274. Besner, D. (1987). Phonology, lexical access in reading, and articulatory suppression: A critical review. Quarterly Journal of Experimental Psychology, 39A, 467—478. Bishop, D. V. M , & Robson, J. (1989). Unimpaired short-term memory and rhyme judgement in congenitally speechless individuals: Implications for the notion of "articulatory coding." Quarterly Journal of Experimental Psychology, 41 A, 123-140. Bradley, L., & Bryant, P. E. (1983). Categorising sounds and learning to read: A causal connection. Nature, 301, 419-421. Broadbent, D. E. (1958). Perception and communication. London: Pergamon Press. Brown, J. (1958). Some tests of the decay theory of immediate memory. Quarterly Journal of Experimental Psychology, 10, 12-21. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A, 705-738. Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235—271. Colle, H. A. (1980). Auditory encoding in visual short-term recall: Effects of noise intensity and spatial location. Journal of Verbal Learning and Verbal Behavior, 19, 722-735. Colle, H. A., & Welsh, A. (1976). Acoustic masking in primary memory. Journal of Verbal Learning and Verbal Behavior, 15, 17-32. Conrad, R. (1964). Acoustic confusion in immediate memory. British Journal of Psychology, 55, 75-84. Conrad, R., & Hull, A. J. (1964). Information, acoustic confusion and memory span. British Journal of Psychology, 55, 429-432. Craik, F. I. M , & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684. Craik, F. I. M., & Watkins, M. J. (1973). The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12, 599-607.

11

Baddeley

Davis, R., Moray, N., & Treisman, A. (1961). Imitative responses and rate of gain of information. Quarterly Journal of Experimental Psychology, 13, 78-90. Ellis, N., & Large, B. (1987). The development of reading: As you seek you shall find. British Journal of Psychology, 78, 1-28. Farah, M. (1988). Is visual memory really visual? Psychological Review, 95, 307-317. Gathercole, S. E., & Baddeley, A. D. (1987). The processes underlying segmental analysis. Cahiers de Psychologie Cognitive, 7, 462-464. Glanzer, M. (1972). Storage mechanisms in recall. In G. H. Bower (Ed.), The Psychology of learning and motivation: Advances in research and theory (Vol. 5). New York: Academic Press. Glanzer, M , & Cunitz, A. R. (1966). Two storage mechanisms in free recall. Journal of Verbal Learning and Verbal Behavior, 5, 351—360. Hall, J. W., Wilson, K. P., Humphreys, M. S., Tinzmann, M. B. & Bowyer, P. M. (1983). Phonemic similarity effects in good vs poor readers. Memory and Cognition, 11, 520-527. Hanley, J. R., & Broadbent, C. (1987). The effect of unattended speech on serial recall following auditory presentation. British Journal of Psychology, '78, 287-298. Hebb, D. O. (1949). Organization of Behavior. New York: Wiley. Hinton, G. E., & Plaut, D. C. (In press). Using fast weights to deblur old memories. To appear in Proceedings of the Ninth Annual Conference of the Cognitive Science Society, Seattle, Washington, 1987. Huey, E. B. (1908). The psychology and pedagogy of reading. New York: Macmillan. Johnston, R. (1982). Phonological coding in dyslexic readers. British Journal of Psychology, 73, 455-460. Kintsch, W., & Buschke, H. (1969). Homophones and synonyms in short-term memory. Journal of Experimental Psychology, 80, 403-407. Kleiman, G. M. (1975). Speech recoding in reading. Journal of Verbal Learning and Verbal Behavior, 24, 323-339. Liberman, I. Y., Mann, V. A., Shankweiler, D., & Werfelman, M. (1982). Children's memory for recurring linguistic and nonliguistic material in relation to reading ability. Cortex, 18,

367-375. Logie, R. H., Cubelli, R., Delia Sala, S., Alberoni, M., & Nichelli, P. (In press). In J. Crawford & D. M. Parker (Eds.), Developments in clinical and experimental neuropsychology. New York: Plenum Press. McLeod, P., & Posner, M. I. (1984) Privileged loops from percept to act. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 55-66). London: Erlbaum. Melton, A. W. (1963). Implications of short-term memory for a general theory of memory. Journal of Verbal Learning and Verbal Behavior, 2, 1-21. Miles, T. R., & Ellis, N. C. (1981). A lexical encoding difficulty II: Clinical observations. In G. Th. Pavlidis & T. R. Miles (Eds.), Dyslexia research and its applications to education (pp. 217—244). Chichester: Wiley. Milner, B. (1966). Amnesia following operation on the temporal lobes. In C W. M. Whitty and O. L. Zangwill (Eds.), Amnesia (pp. 109-133). London: Butterworths. Morais, J., Alegria, J., & Content, A. (1987). The relationships between segmental analysis and alphabetic literacy: An interactive view. Cahiers de Psychologie Cognitive, 7, 415—438. Ostergaard, A. L., & Meudell, P. R. (1984). Immediate memory span: Recognition memory for subspan series of words, and serial position effects in recognition memory for supraspan series of verbal and nonverbal items in Broca's and Wernicke's aphasia. Brain and language, 22, 1-13. Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193-198. Postman, L. (1975). Verbal learning and memory. Annual Review of Psychology, 26, 291-335. Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception and Psychophysics, 2, 437-442.


73

Saffran, E. M , & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with deficient auditory short-term memory. Brain and Language, 2, 420—433. Salame, P., & Baddeley, A. D. (1982). Disruption of short-term memory by unattended speech: Implications for the structure of working memory. Journal of Verbal Learning and Verbal Behavior, 21, 150-164. Salame, P., & Baddeley, A. D. (1986). Phonological factors in STM: Similarity and the unattended speech effect. Bulletin of the Psychonomic Society, 24, 263-265. Salame, P., & Baddeley, A. D. (1987). Noise, unattended speech and short-term memory. Ergonomics, 30, 1185-1194. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261—273. Torgeson, J. K., Rashotte, C. A., & Greenstein, J. J. (1985). Listening comprehension in learning disabled children who perform poorly on memory span tasks. Unpublished manuscript, Florida State University (Experiments, 5, 6, and 7). Tzeng, O. J. L. (1973). Positive recency effect in delayed free recall. Journal of Verbal Learning and Verbal Behavior, 12, 436-439. Vallar, G., & Baddeley, A. D. (1984a). Fractionation of working memory. Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Baddeley, A. D. (1984b). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Vallar, G., & Baddeley, A. D. (1987). Phonological short-term store and sentence processing. Cognitive Neuropsychology, 417-438. Vallar, G., & Cappa, S. F. (1987). Articulation and verbal short-term memory: Evidence from anarthria. Cognitive Neuropsychology, 4, 55-7'8. Zangwill, O. L. (1946). Some qualitative observations on verbal memory in cases of cerebral lesion. British Journal of Psychology, 37, 8-19.

3. Multiple phonological representations and verbal short-term memory FRANCES J. FRIEDRICH

3.1. Introduction The importance of phonological coding to immediate memory performance has been apparent for many years, starting with Conrad's (1964) important demonstration of phonological errors in a memory task for visually presented letters. The specific characteristics of the phonological code have been the subject of debate, however. For example, Besner has argued (Besner, Davies, & Daniels, 1981; Besner & Davelaar, 1982) that the phonological representations underlying reading and short-term memory (STM) tasks are dissociable and that there are at least two kinds of phonological representations. A number of other distinctions among speech-based codes and processes have been described as well, including a distinction between a sensory "echoic" and a more abstract phonological representation (e.g., Crowder, 1978), between "auditory" and "phonetic" codes used in speech perception (e.g., Pisoni, 1973), between "assembled" and "addressed" phonological processes in reading (e.g., Patterson, 1982), and between a phonological store and an articulatory loop in working memory (e.g., Vallar & Baddeley, 1984b; Baddeley, 1986). The neuropsychological literature certainly seems to suggest that multiple representations are available for use in immediate memory tasks. Indeed, much of the recent literature on STM impairments has been interpreted in the context of a model of working memory that includes a phonological store and an articulatory rehearsal process that are separable (e.g., Shallice & Butterworth, 1977; Vallar & Baddeley, 1984a, b; Baddeley, 1986; Vallar & Cappa, 1987). It remains unclear how many different types of representations are available, what the relationships between the different types of representations are, and how multiple, simultaneously active representations might contribute to immediate memory performance. However, with the advent of multilevel and interactive models dealing with various aspects of language processing (e.g., Preparation of this chapter was supported in part by Biomedical Grant 5.32078 and URC Grant 215-161 from the University of Utah. I am grateful to Tim Shallice, Alan Baddeley, and an anonymous reviewer for their helpful comments on an earlier version of this chapter. 74

Multiple phonological representations

75

McClelland & Rumelhart, 1981; Rumelhart & McClelland, 1982; Shallice & McCarthy, 1985; McClelland & Elman, 1986), a general framework is emerging that may make it possible to accommodate a variety of related but distinct speech-based representations. In addition, such a framework would seem to hold the promise of clarifying the relationships between speech, reading, and memory processes. A logical first step in this process is to identify what types of representations are available; the first section of this chapter will provide an overview of evidence from both the normal and neuropsychological literature of separable speech-based representations, especially as they relate to immediate memory performance. A framework for considering how these representations might interact is then sketched. The strength of the connections between different types of representations is a feature of particular importance in defining the constraints within an interactive framework and, it will be argued here, in understanding the role that "control" processes play in immediate memory tasks.

3.2. Phonological representations The concept of a phonological code has been widely used in the memory literature and refers in a general sense to internalized speech-based representations. The word phonological is often explicitly neutral with regard to the specific characteristics of the representation; for example, Besner and Davelaar (1982, p. 702) use it in this way: "The term 'phonological' is a neutral one used so as to avoid specifying the exact form of the code; no claim is being made as to whether this code should be considered to be acoustic, articulatory, auditory imagery, or 'abstract-cognitive' (cf. Wickelgren, 1969)." However, discussions of the nature of the phonological representation have clustered around three main distinctions. A major distinction has focused on whether the phonological code reflects auditory or articulatory features; with respect to the possibility of multiple speech-based representations, the question becomes whether distinct auditory and articulatory representations can be identified. Second, within the speech perception literature in particular, a distinction has been made between different types of auditorily based representations. The separation of pre- and postcategorical speech codes draws something of a line between a sensory (acoustic) representation and a more abstract code that is not affected by acoustic variables. Finally, there may be multiple articulatory representations that differ in terms of the size and lexical status of the unit, and evidence from the reading literature suggesting a distinction between preand postlexical codes will be considered. 3.2.1. Auditory and articulatory codes Traditionally, the prerecency portion of the immediate memory span has been thought to reflect a phonological representation that is accessible to both visual and auditory

76

Friedrich

information, although the exact nature of the code, in terms of whether it has auditory or articulatory characteristics, has been the subject of debate in the past (e.g., see Crowder, 1976). Wickelgren (1969) raised the possibility of an "abstract" code that is neither acoustic nor articulatory, but correlated with both; Crowder (1978) concluded that the data more strongly supported an articulatory representation, although a clear resolution of the issue was not possible at that point. More recently, there has been substantial evidence that separable phonological and articulatory representations contribute to immediate memory performance. There are a number of effects in the normal literature that cannot be accounted for on the basis of articulatory representations alone (for reviews see Shallice & Vallar, this volume, chapter 1; Baddeley, chapter 2, 1986). These effects include the finding that articulatory suppression eliminates phonological similarity effects with visual but not with auditory presentation of verbal material (Levy, 1971; Vallar & Baddeley, 1984b) and evidence that irrelevant speech can interfere with memory for visually presented items (Colle & Welsh, 1976; Salame & Baddeley, 1982). In addition, articulatory suppression appears to disrupt rhyme judgments but not homophone judgments, suggesting that the latter task makes use of a phonological representation that is nonarticulatory in nature (Besner et al., 1981; Besner & Davelaar, 1982). In the neuropsychological literature, there have been several cases of selective impairments that argue for separable auditory and articulatory representations. One such case was presented by Shallice and Butterworth (1977); their patient had an impaired auditory memory span but no measurable articulatory deficit in speech. If we assume that the articulatory loop used in memory also underlies speech production, the memory deficit must reflect an impairment in a nonarticulatory code. Patient PV (Vallar & Baddeley, 1984a, b) also appears to have an impairment in phonological storage without evidence of a deficit in articulatory processes. In contrast, Vallar and Cappa's (1987) patients had severe impairments in overt articulation, but had memory spans within normal limits. Evidence of impairments in auditory discrimination tasks in patients with reduced memory spans supports the argument that the deficit reflects an impairment in an auditory "input" memory (e.g., Allport, 1984; Friedrich, Glenn, & Marin, 1984); it is interesting that in a group of aphasic patients the discrimination deficits do not appear to be particularly strongly related to comprehension deficits (Blumstein, Baker, & Goodglass, 1977) and so may not be readily apparent in patients with spared auditory comprehension. Allport (1984) suggested that the STM input deficit results from an "impaired distinctiveness" in the encoding of speech sounds: The phonological representation is unstable and cannot be adequately distinguished or maintained. However, Berndt and Mitchum's patient EDE (this volume, chapter 5), who showed no impairment in basic discrimination tasks, may provide an unusually pure example of a case in which the auditory information is encoded accurately, but cannot be maintained in a nonarticulatory form.


77

Overall, then, there seems to be substantial justification for separating the contributions of articulatory rehearsal from a nonarticulatory representation, from both the normal and the neuropsychological literature. The manner in which these separable speech-based representations might contribute to immediate memory performance and be selectively disrupted by articulatory suppression or brain injury has been described in some detail in the context of the working memory model (see Baddeley, 1986; this volume, chapter 2). Basically, the verbal subsystem of working memory is thought to have two components, an articulatory rehearsal loop and a phonological store. The latter is accessible directly by auditory material and by means of articulatory rehearsal for visual material. There are two features of the phonological store as described by the working memory model that are of particular interest here. First of all, in terms of the characteristics of the phonological code, the evidence suggests that information held there is sensitive to phonetic but not acoustic or semantic factors; that is, the unit of representation seems to be at the level of a phoneme or a syllable (Baddeley, 1986). However, evidence reviewed in the next section and by Campbell (this volume, chapter 11) suggests that there may in fact be several levels of closely interconnected representations within this phonological store. A second interesting point is that it appears that auditory information has "obligatory access" to the phonological store; that is, an auditory stimulus - even when it is irrelevant to the task - is placed in the phonological store automatically, without demands on an attentional control mechanism. Visual stimuli, in contrast, can be placed in the phonological store only by way of the resource-demanding articulatory control process. The mechanism for this type of obligatory access is not clear, however, and it is possible that an interactive model detailing the nature of the connections between different types and levels of representation may help account for these effects. This issue will be considered in more detail in the final section of this chapter.

3.2.2. Pre- and postcategorical representations The distinction between a representation that retains relatively "raw" acoustic information and a code consisting of abstracted phonemic features has been made in both the speech perception and immediate memory literatures. The characteristics of a "precategorical acoustic store," or echoic memory, have been studied extensively (e.g., Crowder & Morton, 1969; Morton, Crowder, & Prussia 1971; Crowder, 1976); however, because an acoustic representation seems to decay rapidly and is vulnerable to masking, it may not contribute a great deal to performance in immediate span tasks under normal conditions. Two lines of evidence in particular have been used to argue for a precategorical acoustic store: the modality effect and the stimulus suffix effect. The modality effect refers to the finding that the recall of auditorily presented material is superior to that of

78

Friedrich

visually presented material, particularly for the final items in the list. The suffix effect is demonstrated by presenting an irrelevant speech item at the end of the memory list; recall of the final item is reduced for lists presented auditorily but not visually. Crowder (1976, 1978) originally suggested that the modality effect is the result of "parallel and supplemental" information held in a precategorical store. From this view, most of the material retrieved in a memory span task is in a postcategorical code accessible by both visual and auditory input; with auditory presentation, the final item has the additional benefit of an unmasked sensory representation. When an irrelevant suffix is presented at the end of the list, however, the acoustic information for the final item is masked and the auditory benefit is eliminated. This pre- and postcategorical distinction between an echoic and a phonological code is similar to the distinction made in the speech perception literature between auditory and phonetic memory codes (e.g., Pisoni, 1973, 1975). The auditory memory code seems to retain acoustic features that are not available in the phonetic memory code. Differences that have been found in the way vowels and consonants are discriminated and recalled in immediate memory tasks have been related to the differential use of these codes (e.g., Crowder, 1971; Pisoni, 1973). For example, within-category discriminations for stop consonants tend to be poor and are not affected by delay interval; in contrast, within-category discriminations for vowels show good accuracy at short delay intervals but decrease significantly as the interval increases (Pisoni, 1973). Pisoni suggested that phonetic memory, based on derived phonetic properties, allows the listener to make between-category discriminations and is reliable for both vowels and consonants. An auditory memory code, on the other hand, retains acoustic features and facilitates within-category discriminations for vowels but not for consonants. The differential availability of acoustic feature information may not be based on classes of speech sounds, however; Darwin and Baddeley (1974) have argued that the acoustic discriminability of the stimuli will determine whether acoustic memory contributes to performance. That is, acoustic memory may be thought of as a relatively literal representation of the speech stimulus that degrades rapidly over time. With degradation, the acoustic representation becomes blurred and can make little contribution to fine acoustic discriminations. Darwin and Baddeley suggested that the stop consonants used in previous work tended to be more highly confusable than the vowels and thus showed little evidence of acoustic memory effects. The nature of the auditory memory representation has also been called into question recently by evidence that supposedly precategorical storage effects (such as the modality and suffix effects) are not limited to auditorily presented information. Lists of digits that are lipread, but not heard, produce many of the same effects as auditorily presented lists: Recall of the last item is superior to that of a written list and that advantage is eliminated by a heard suffix (Campbell & Dodd, 1984; Campbell, 1987, and this volume, chapter 11). Similarly, a lipread suffix has been shown to disrupt


79

recency effects in auditorily presented lists (Spoehr & Corin, 1978). The lipreading effects do not appear to be due to an articulatory process, since a suffix that is "mouthed" (but not spoken) by the subject only partially eliminates the recency effect (Nairne & Crowder, 1982; Campbell, 1987). Campbell (1987) has suggested that the auditory recency and suffix effects reflect a prelexical abstract representation that is not specific to input from the auditory modality. There may also be a specifically acoustic representation that is sensitive to acoustic characteristics, but the contribution of such a trace to recency and suffix effects seems to be considerably less than was originally thought (Campbell & Dodd, 1984). On the basis of these and other findings, Coltheart (1984) suggested that the very existence of an echoic memory may be called into question, particularly since the auditory recency and suffix effects, which provided the basis of the theorizing about echoic memory, have been shown not to be specific auditory effects. Although it is true that it is no longer appropriate to account for the suffix effect simply by reference to an acoustic memory, it does seem that there are different degrees of disruption caused by different types of suffixes. There seems to be an acoustic component of the auditory recency effect that a lipread suffix does not eliminate; similarly, a mouthed suffix is not as effective as a lipread suffix (Campbell & Dodd, 1984). One possibility is that the auditory recency effect reflects the simultaneous activation of several levels of speechbased representation, including acoustic features, a phonetic representation activated by both auditory and lipread input, more abstract phonemic units, and articulatory representations. The different types of suffixes may selectively interfere with specific representations that contribute to the effect; the different representations may therefore be separable, although closely connected (see also Campbell, this volume, chapter 11). An interesting question from the neuropsychological perspective is whether selective impairments occur between acoustic and more abstract auditory codes (or between the different "levels" within the input buffer in Campbell's scheme; see chapter 11) and what the consequences of different selective impairments would be. The existing neuropsychological data do not provide a clear answer to this question, but there is some evidence that is suggestive. One line of evidence comes from the literature on lipreading effects; Campbell (this volume, chapter 11), for example, suggests that patient DB may have a partial impairment of access to the phonological input buffer and that lipreading helps his repetition and comprehension performance by activating phonetic representations by another pathway. Another line of evidence suggests that more abstract representations can be selectively impaired. Caramazza, Berndt, and Basili (1983) suggested that their patient JS demonstrated a pure word deafness that resulted from a selective impairment in phonology with relatively spared auditory processes. JS could accurately identify auditory nonverbal sounds and showed excellent lexical decision abilities for visual words, but performed at chance on auditory lexical decision. In terms of auditory

SO

Friedrich

discrimination performance, the picture is rather mixed: Discrimination of computersynthesized syllables was impossible, and JS failed to identify (by pointing to written samples) either synthetic or natural speech sounds, including both vowels and consonants. On one task, however, when JS was required to make same-different judgments on natural speech syllables, he performed very well: 94% for vowels and 97% for consonants. Caramazza et al. concluded that the additional acoustic information available in natural speech allowed him to perform well on that discrimination task, and that the impairment involved processing at a phonetic rather than an acoustic level. The speech perception literature described earlier provides another possible means of distinguishing between pre- and postcategorical representations: the finding of categorical perception for stop consonants. A distinguishing feature between two consonants, such as voice onset time (VOT) for the syllables ba and pa, can be manipulated systematically by way of computer-synthesized speech so that an acoustic gradient is created (i.e., a series of synthesized consonants that differ in VOT to varying degrees). To the extent that the specific acoustic information is retained, identification of those consonants should show a similar gradient reflecting a gradual shift in perception from more bfl-like sounds to more pa-like sounds. The phenomenon of categorical perception is that this gradual shift is not found for normal subjects; instead, any given syllable is heard as either ba or pa and the point of transition, in terms of VOT, is very abrupt. Patient EA was evaluated on a test of categorical perception of stop consonants. EA fits the diagnostic criteria for conduction aphasia; in particular, she had a severe repetition deficit but her auditory comprehension for words and sentences was quite good (Friedrich et al., 1984; Friedrich, Martin, & Kemper, 1985). On the basis of extensive investigations of language and memory processes, Friedrich et al. (1984) concluded that EA had an impairment in phonological coding, specifically in the storage and maintenance of a postcategorical speech code. EA was clearly impaired in the categorical perception of stop consonants, but interestingly, her performance showed the kind of gradient that would be expected from the use of precategorical information; that is, she showed a gradual shift in classification of the speech sounds that followed the VOT gradient. In addition, EA showed a difference in discrimination tasks involving vowels and consonants; she was above chance for both, but her miss rate was 10% for the vowels and 27% for the consonants. Evidence from the speech perception literature that acoustic information is used to a greater extent in the perception of vowels than consonants might suggest that EA is better able to use precategorical than postcategorical information in these discrimination tasks, although the possibility that the vowel and consonant pairs may also differ in discriminability (e.g., Darwin & Baddeley, 1974) must also be kept in mind.


81

In general, EA's auditory comprehension was very good and may have contributed to her performance on basic perceptual tasks. Berndt and Mitchum (this volume, chapter 5) raise this issue with respect to EDE, who performed well on phoneme discrimination and identification tasks, and in single word auditory comprehension, but had difficulty in manipulating nonlexical phonemes (as in a rhyme task) and in making auditory lexical decisions. In particular, EDE had difficulty rejecting nonwords. Berndt and Mitchum suggest that although auditory representations were intact for EDE, phonetic representations could not be maintained in a purely phonological form, that is, without support from representations at a lexical level. Cases such as these suggest that there are separable auditory and phonetic representations, although it is clearly difficult to determine to what extent the basic auditory information may be impoverished and to what extent higher-level processes may provide "support" for more basic representations, as would be expected in an interactive system. It has become clear, however, that patients with good auditory comprehension may show basic impairments in phoneme discrimination and identification, and that this pattern may be related to impairments in immediate memory span, due to impoverished "input" information (e.g., Allport, 1984).

3.2.3. Pre- and postlexical phonology In the area of reading processes, several authors have distinguished between lexical and nonlexical pathways for grapheme-to-phoneme conversion (e.g., Coltheart, 1980; Patterson, 1982). Using the terms addressed and assembled phonology, Patterson distinguishes between putting together a phonological representation in order to determine the meaning of a word (assembled) and "looking up" an already-assembled phonological representation after cognition (addressed). The pronunciation of irregular words shows the distinction clearly: An assembled representation of the word island would include the s sound, while an addressed, or postlexical, representation would not. Although both phonological forms seem to have an articulatory character (Patterson suggests, for instance, that addressed phonology is used in producing spontaneous speech), they might be said to differ in the lexical value of the unit. Assembled phonology involves representations at a sublexical level that are then bound together, whereas in addressed phonological representations the sublexical components are already tightly associated, with the appropriate modification of the components in the word context. In addition to greater familiarity, the addressed or lexical units would also differ from sublexical components in that they are linked to associated semantic features and possibly other types of representation, such as visual images. Although the contribution of sublexical units to STM has been the focus of much of the work in the area (the working memory model, for instance, suggests that information is stored in the phonological buffer to phoneme- or syllable-sized units),

SZ

Friedrich

lexical factors do play a role in immediate memory performance (see also Saffran and Martin, this volume, chapter 6). In normal populations, for instance, memory span is greater for high-frequency than low-frequency words (Watkins, 1977; Watkins & Watkins, 1977). Similarly, memory span for words is greater than for nonsense syllables: Cavanaugh (1972) reported that the average normal span was 5.5 for unrelated words and 3.4 for nonsense syllables. The importance of lexical information, and of the access that it provides to semantic representations in particular, may be even greater for patients, especially those who appear to have some impairment in auditory input representations (Allport, 1984; Friedrich et al, 1984). As indicated earlier, examples of dissociations between pre- and postlexical processing can be seen in the reading literature, where the focus is on the activation of articulatory representations via visual pathways. For example, Patterson (1982) presents AM, a patient with phonological dyslexia who demonstrates the selective loss of assembled phonology with spared addressed phonology. The selective nature of the deficit is most obvious in that AM was able to read real words aloud quite well (except for function words), but was severely impaired in reading nonwords, even if they were orthographically regular. In fact, he even had trouble naming letters, presumably due to the relatively nonlexical nature of that material. Many of his attempts to read nonwords resulted in real words. Patterson suggests that if AM could recognize the word visually, the addressed phonological representation was available for pronunciation; if he had to assemble sublexical units in order to "sound out" a pronounceable nonword, he would fail. There are implications for repetition and memory performance as well. Articulatory control processes are generally thought to provide a means of "refreshing" information in an auditory input buffer (e.g., Baddeley, 1986). A loss of or inability to access articulatory representations that will in turn reactivate auditorily based representations would be likely to result in an STM deficit. AM, in fact, had a span of four digits. Although his span for nonsense syllables was not reported, we would expect that in a case such as AM, who may have a particular deficit in activating sublexical articulatory units, memory span should be particularly impaired for nonsense syllables. Moreover, as the length of the list to be recalled increases, the chances of retrieving previously activated auditory representations may diminish if they cannot be refreshed by reciprocal sublexical articulatory activation. Warrington and Shallice (1969) originally raised the point that memory performance declined with increased load, based on the probability of reporting the entire string correctly. To address the prediction of a string length effect for sublexical material, however, it is useful to consider the average number of items reported per trial. Evidence on this point is scanty, but there is some suggestion of differential effects of memory string length for lexical and nonlexical material in patients with STM deficits. For instance, EA showed an overall effect of word frequency and meaningfulness on


83

immediate memory performance (Friedrich et al., 1984). In addition, she showed a decreased probability of reporting even 1 nonsense syllable correctly as string length increased; out of 20 strings, she repeated 16 items correctly when there was only 1 item/string but only 4 items in total when there were 4 items/string. In contrast, her performance was stable for real words: she reported all 20 items correctly for 1 item/string lists, 26 words for 2 item/string condition, and 26 words for the 4 item/string condition. Thus for words, she consistently reported 1 or 2 items correctly on each trial regardless of list length. From the data that Vallar and Baddeley (1984b) present for PV, it appears that the same type of effect may be occurring for letter recall. PV reported, on average, 2.19 items correctly from a 3-letter string but only an average of 1.28 letters per list for the 4letter strings. The list length effect was even more notable for phonologically similar items: she recalled an average of 1.5 items per list for 2-item strings, but an average of less than 1 item per list with 3-item strings. The same pattern of decrement did not occur with word strings, however; for example, PV reported an average of 2.8 and 2.9 items per trial for 4-word and 5-word strings respectively, for two syllable words. In the case of EA, Friedrich et al. (1984) argued that the links between auditory and articulatory representations were impaired, and that in order to activate articulatory representations for words a lexical-semantic route had to be used, which would do little to activate sublexical articulatory units. It is, in fact, frequently reported that patients with STM deficits seem to depend heavily on lexical and semantic analyses (e.g., Friedrich et al., 1985; Saffran & Marin, 1975). Given these potentially important activation patterns, it is possible that an impairment in the lexical system would have consequences for STM performance as well. Indeed, Saffran and Martin (this volume, chapter 6) demonstrate that a patient with difficulty in lexical access also shows some impairment in immediate memory, although the pattern of impairment differs from that of a patient with an impairment at an auditory input level. Shallice and McCarthy (1985) have argued that rather than two levels of correspondence between orthographic and phonological representations, there may be multiple levels, ranging from graphemic to morphemic units and including syllabic and subsyllabic units as well. Each visual unit corresponds to a level of phonological representation, and the processes involved in visual-to-articulatory activation occur in parallel. Shallice and McCarthy apply this approach to account for "phonological" reading, that is, cases in which semantic information is not directly accessible from the visual information. The multilevel approach has the advantage of being computationally more powerful than a system postulating two sizes of unit; perhaps more important in the present context is the fact that to the extent that there are multiple levels of auditory input representations, as described earlier, we would postulate that corresponding articulatory representations are also available (see also Campbell, this volume, chapter 11).

84

Friedrich

3.2.4. Conclusions Overall, there seems to be ample evidence from both the normal and neuropsychological literatures of multiple phonological representations; in fact, it appears that there may be multiple levels of representation at both the auditory and articulatory levels. The value of making a precategorical-postcategorical distinction in terms of the contributions of each to immediate memory seems questionable at this point, given the evidence that the auditory recency and suffix effects are not strictly modality specific. There is evidence of separable representations at the perceptual level, but the pre- and postcategorical distinction does not go far enough in defining the nature of the different representations. There is evidence as well of at least two, and possibly more, dissociable levels of articulatory representation. In terms of neuropsychological evidence, the selectiveness of certain impairments and the fact that patients with striking deficits on one task (e.g., discrimination) show spared performance on seemingly related tasks (e.g., comprehension) suggest parallel activation of multiple, related representations. A framework for relating these representations and investigating the activation patterns among them will be explored in the next section.

3.3. Multiple representations and immediate memory: an interactive framework Although the empirical evidence supporting the notion of multiple speech-based representations comes from a variety of research areas, some of the most useful frameworks for considering how different levels of representation work together in a given task situation have come from the perceptual literature, and from the study of speech and visual word perception in particular. The class of interactive activation models that have emerged in recent years, such as the visual word perception model (McClelland & Rumelhart, 1981; Rumelhart and McClelland, 1982) and the TRACE model of speech perception (McClelland & Elman, 1986), have been designed to show that a variety of apparently rule-based and lexical-level effects may be the natural result of the simultaneous activation of several interconnected levels of representation. In the visual word perception model, for instance, McClelland and Rumelhart (1981) distinguished three levels of visual representation, for features, letter clusters, and words, with connections both within and between levels. The interactive nature of this system allows word level representations to enhance the activation of component features and letters, and therefore produce context effects such as the word superiority effect. Although these interactive models have been designed as models of perception, they clearly have implications for the study of immediate memory. Short-term memory can


85

be thought of in terms of the temporary activation of the network of interconnected representations. In the words of McClelland and Elman, with respect to the TRACE model of speech perception: The distinction between perception and (primary) memory is completely blurred, since the percept is unfolding in the same structures that serve as working memory, and perceptual processing of older portions of the input continues even as newer portions are coming into the system. These continuing interactions permit the model to incorporate right context effects, and allow the model to account directly for certain aspects of short-term memory, such as the fact that more information can be retained for short periods of time if it hangs together to form a coherent whole. (1986, pp. 9-10). The concept of immediate memory as a temporary activation of portions of the semantic memory network has been explicitly addressed in memory and attention models as well. Monsell (1984) proposed a system in which there are several levels and domains of representation that can be activated temporarily. Monsell distinguished between a persisting activation of a preexisting representation, and a process in which activated representations are copied into a limited-capacity workspace that is separate from permanent storage; the latter type of representation allows new representations and associations to be constructed. Although an interactive framework for coordinating information across domains is not made explicit, Monsell suggests that there are links between systems and that maintenance of information by rehearsal may reflect a cycling of information back and forth between different domains; consequently, one type of representation can be used to reactivate another temporary representation. Moreover, he suggests that some types of temporary activation occur automatically, while others give evidence of an active process, reflecting executive control. Schneider and Shiffrin's (1977) model looks specifically at the nature of these automatic and controlled processes, at the relationship between selective attention and short-term memory search, and at the way in which "the flow of information into and out of the short-term store" is determined. According to their general informationprocessing framework (Shiffrin & Schneider, 1977), the short-term store (STS) consists of concurrently activated memory nodes, and the loss of information from STS is the return of currently active information to an inactive state. The retrieval of information from this state of temporary activation can result from an automatic process, in which the retrieval processes are drawn to a subset of the activated nodes, or from a more resource-demanding controlled process, by which the contents of the store are searched in a serial order. The nature of these nodes or how they are connected within the system is not specified; what is important in the context of this model is how the activated information is used in detection and memory search decisions. Given the evidence of multiple representations described earlier and the possibility

86

Friedrich

that multiple representations may contribute to performance in immediate memory tasks, it may be useful to consider how normal and impaired immediate memory might operate in the context of an interactive framework. Six basic features of such a framework are suggested. (a) When verbal material is presented, preexisting representations are activated; certain types of representations (e.g., motor, visuospatial or imagery) can also be activated internally, that is, from within the system rather than by external stimuli. (b) Immediate memory tasks reflect what is currently active in the network. Although only a subset of the active representations may be accessible for response generation, the nature of the interactive system is such that the activation level of any one item reflects whether related representations are simultaneously active. (c) The most active representations will be most accessible to the response system for decision or response processes. In addition, motor-based or articulatory representations may be easiest to maintain in an active state, because they can be refreshed by means of an active "recycling" process, driven internally rather than by external stimuli. (d) The active rehearsal or control processes postulated by various accounts of immediate memory reflect this internal activation process, which may serve as a sort of "selective enhancement" of certain representations. (e) Different types of representations are interconnected and can be active simultaneously. The interactive nature of the system serves not only to spread activation to related representations but also to strengthen and enhance certain representations over time. For example, the interactive nature of the system means that sublexical representations may be "supported" by higher level representations, even if the quality of the input is poor. In addition, a representation with more connections to other codes may end up in a higher activation state than one with fewer connections, given the same initial stimulus input. These consequences may be particularly important in cases of selective deficits, in which one source of input is impaired or the activation of one type of representation is disrupted; for example, the articulatory representation associated with a visually presented word might be more accessible for concrete than abstract words in patients with acquired dyslexia because of the activation of and feedback from an associated "imagery" representation that is not available for the abstract words. (f) Certain associations between representations are stronger than others and will generate stronger activation patterns; indeed, in the connectionist interactive models mentioned earlier these "connection weights" in some sense define the representation. In terms of immediate memory performance, one implication of this feature is that how a particular representation is activated may determine its accessibility for a response output. For example, an articulatory representation might be more quickly or strongly activated when a word is presented auditorily rather than visually.


87

3.3.1. Association strength and obligatory access This final point deserves additional attention, not only because it is a central feature in connectionist models but also because it seems to relate to the issue of executive control discussed to varying degrees in most accounts of short-term memory (e.g., Salame & Baddeley, 1982; Monsell, 1984; Baddeley, 1986; Schneider & Shiffrin, 1977). This distinction in the strength of association seems to correspond to the notions of automatic processes, in the context of attentional models, and obligatory access in the working memory model. Given a particular input, certain kinds of information seem to be so strongly activated that they dominate the short-term store, as for the consistently mapped memory items in Schneider and Shiffrin's work, and the obligatory access of auditory items in the phonological buffer postulated by Salame and Baddeley (1982). For other types of input, a more active, resource-demanding recoding or matching is required, as occurs for varied mapping items or visually presented memory material. In the context of an interactive activation model, we can recast the automaticity issue in terms of the strength of the connections between different types of representations: Assuming that a certain level of activation is required for the response processes, the amount of internal enhancement needed will differ depending on the strength of activation coming from other representations. From this view, when verbal material is presented visually, the strength of activation of the articulatory representations is not sufficient, and a selective enhancement through attentional resources is necessary. The connections between auditory and articulatory representations, on the other hand, may be sufficiently strong that this selective enhancement is not initially required, although it may be necessary for maintenance of the activation patterns. The differential strength of connections within the interactive network may also be a critical feature of the system in terms of putting constraints on interactive models and fitting them to the empirical data. Given the potentially large number of available representations, and the potential for unlimited connections among the different types of representations, it seems difficult to predict in advance what the resulting activation pattern would be in any given situation. If we focus on the strength of the interconnections, in terms of the relative degree of attention required to produce the necessary level of activation, it may be possible to define the interactive processes more precisely. For example, there is evidence of a special class of connections among representations in the cognitive system that seem to reflect direct stimulus-response mapping (McLeod & Posner, 1984). McLeod and Posner suggested that verbatim repetition comprised a "privileged loop" that operates with little or no interference with other processes that are carried out at the same time. Of the several auditory-vocal or auditory-manual tasks investigated, verbatim repetition was the only input-output combination that McLeod and Posner found to show interference-free performance

88

Friedrich

when carried out concurrently with a visual letter-matching task. A "modality crossover" condition produced the greatest dual-task interference, when a manual response (moving a lever up or down) was made in response to the auditory stimuli up or down, and a vocal response ("same" or "different") was made to the visual lettermatching task. Surprisingly, substantial interference also occurred when a seemingly minor modification of a shadowing task was made: In the semantic associate condition the subject heard the words up or down and responded by saying "high" or "low." Although the stimulus and response are highly congruent and do not overlap in the use of input and output channels, the need to make a semantic transformation was sufficient to produce interference, in both latencies and error rates, which did not diminish over several days of practice. Verbatim repetition does indeed seem to be "privileged" in this context. Waters, Komoda, and Arbuckle (1985) have shown a similar effect of stimulusresponse mapping in a dual-task situation. They combined various interference tasks with reading for meaning and found no specific interference (i.e., interference other than increased processing demands) for verbatim repetition. Interference did occur, however, when the secondary task required a "meaning" transformation on the input, such as responding "A," "B," or "C" to the digits "1," "2," or "3." Additional evidence of the special auditory-articulatory link and of a similar visual-spatial connection have been reported by Posner and Henik (1983). These results suggest that there is a strong link between auditory input and articulatory output representations that is distinct from other connections using the same input and output modalities; responding to an auditory word with a close semantic associate rather than a verbatim repetition will forfeit the benefits of the direct link, even though the same input and output modalities are used in both cases. Friedrich et al. (1984) suggested that the short-term memory deficit shown by EA reflects the loss of this direct connection. Although EA could repeat individual words without difficulty, she showed considerable interference between single word repetition and a visual matching task in a dual task analogous to the one that was used by McLeod and Posner (1984). Interestingly, in this task she made a number of errors of repetition (saying "low" instead of "high" to the auditory stimulus high), which virtually never occurred in the McLeod and Posner studies. Given these data and her tendency to make semantic substitutions during repetition, it seemed that EA was unable to use the privileged loop for repetition and that the necessary articulatory representations were activated by way of lexical-semantic connections, which are slower, seem to require enhancement of the activation from internal connections, and are more vulnerable to interference. Given the complexity of this type of interaction system, there are a number of different types of deficits that could produce some degree of impairment in immediate memory performance. Indeed, many analyses of patient performance reported in this


89

volume suggest that STM impairments can result from disruption at a variety of points in the cognitive system (see, for example, Campbell, chapter 11; Saffran and Martin, chapter 6). The case of EA, however, provides a particularly useful demonstration of how a deficit in the "input" portion of the system may have consequences for the activation of articulatory representations as well.

3.3.2. Implications for immediate memory performance Although the work reflecting different strengths of association between representations comes primarily from the attention literature, it has interesting implications for the study of immediate memory performance as well. The strength of association concept may, in fact, provide a mechanism for the distinction made in the working memory model between the obligatory access of auditory stimuli into the phonological store and the resource-demanding encoding of visual stimuli into the phonological store by means of articulatory recoding. Obligatory access in working memory may reflect this strong auditory - articulatory link, which would result in a relatively rapid and strong activation of articulatory representations, given an auditory stimulus, with a minimal demand on attentional resources. A visual stimulus may also result in the activation of both articulatory and phonological representations, but because the visual-articulatory association is weaker (or in some cases must perhaps be constructed anew), significant attentional resources must be devoted to activating those representations and the resulting level of activation may be lower overall. The maintenance of activation over time would require attention and an active "refreshing" process regardless of the nature of the input. As suggested earlier, a motor-based representation such as an articulatory code might be more easily refreshed from an internal source than a more sensory-based code, since clearly some production mechanisms exist that allow motor planning and execution without external sensory stimulation. From this view, then, the functional characteristics of immediate memory performance reflect the connections and interactions between different types of representations. Irrelevant speech effects, as described by Colle and Welsh (1976) and Salame and Baddeley (1982), may be a manifestation of the relative strength of auditory—articulatory links; the auditory stimuli will be more effective than visual stimuli in activating articulatory representations in terms of both speed and strength of activation. In the same way, phonological similarity effects will be more robust (i.e., less affected by suppression) with auditory than with visual input. Immediate memory span may be better for words than for nonwords at least in part because activation from associated semantic representations can feed back more effectively for words than for nonwords to help keep the phonological and articulatory representations of words alive. The benefit of using an interactive framework may perhaps be more clearly seen by

90

Friedrich

viewing immediate memory from a rather different perspective. We might ask the question this way: what is the nature of the residual immediate memory for patients with STM impairments or for normals under interference conditions? There is virtually always some residual capacity, in the form of a span of two to three items in patients, and articulatory suppression does not completely eliminate immediate memory in normal subjects (e.g., Baddeley, 1986). An interactive view would suggest that a number of different types of codes are active simultaneously, and that residual capacity in patients reflects the activation of unimpaired representations. As was noted earlier, the immediate memory performance of patients with phonological impairments may reflect the use of semantic information, resulting in less vulnerability to list length effects for words than nonwords and in frequent word substitutions during nonword repetition (e.g., Patterson, 1982; Friedrich et al, 1984). That is, patients who do not show obligatory access of auditory information or who cannot "refresh" memories by way of the articulatory loop may nevertheless be able to retrieve some items as a result of semantic activation. Memory span based on semantic representations would necessarily be reduced relative to the articulatory loop because of the slower and resourcedemanding aspect of that pathway. In this context, the well-established finding that STM patients perform better with visual than with auditory stimuli also makes sense, in that a visual code is activated that can contribute to the maintenance and retrieval of the verbal material.

3.4. Conclusions Neuropsychological studies of selective impairments provide a means of evaluating the functional independence of cognitive systems and of basic processes, and cases that demonstrate a specific deficit of auditory-verbal memory seem to provide a particularly valuable avenue for tracing the role of various phonological relationships in memory and language tasks. One of the things that is particularly striking about this group of patients is that performance in many language and memory tasks can be so good, given the severity of the phonological disruption. Intuitively we would expect that comprehension would be disrupted in the case of a severe phonological impairment; and yet there does not appear to be a strong relation between auditory discrimination skills and auditory comprehension (e.g., Blumstein et al., 1977). This is an indication that the potential contribution of all available representations needs to be considered in the context of a given task; in this case sublexical phonology used in discrimination may be impaired while the lexical phonological representations required by a comprehension task may be supported by semantic connections. In addition, these dissociations suggest that we have at our disposal a good deal of seemingly redundant information that provides residual capabilities when one component of the system is not functioning, either because of an experimental manipulation such as suppression or because of brain injury.


91

The general scheme that emerges here is one of interactive associations among representations of different degrees of complexity and abstraction. The performance observed in any given task may in fact be a result of the entire pattern of activation rather than a small subset of representations. A key concept in this framework is the strength of the association among various speech-based, visual, and semantic representations; this factor may be important in determining what information is maintained in an articulatory form and how items are maintained when the usual speech-based processes are not available. The framework proposed here is, of course, merely a rough outline of how the concepts of multiple phonological representations and interactive connections might apply to questions of immediate memory performance. It does, however, suggest one direction for future work that might clarify the mechanisms of phenomena such as modality effects and help integrate related work from the fields of perception, attention, and memory.

References Allport, A. (1984). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma & D. Bouwhuis (Eds.), Attention and performance X (pp. 313-325). Hillsdale, NJ: Erlbaum. Baddeley, A. (1986). Working Memory. Oxford: Clarenden Press. Besner, D., & Davelaar, E. (1982). Basic processes in reading: Two phonological codes. Canadian Journal of Psychology, 36, 701-711. Besner, D., Davies J., & Daniels, S. (1981). Reading for meaning: The effects of concurrent articulation. Quarterly Journal of Experimental Psychology, 3 3 A, 415-437. Blumstein, S., Baker, E., & Goodglass, H. (1977). Phonological factors in auditory comprehension in aphasia. Neuropsychologia, 15, 19-30. Campbell, R. (1987). Common processes in immediate memory: Precategorical acoustic storage and some of its problems. In A. D. Allport, D. G. MacKay, W. Prinz, & E. Scheerer (Eds.), Language Perception and Production (pp. 131-150). London: Academic Press. Campbell, R., & Dodd, B. (1984). Aspects of hearing by eye. In H. Bouma and D. Bouwhuis (Eds.), Attention and performance X. Hillsdale, NJ: Erlbaum. Caramazza, A., Berndt, R., & Basili, A. (1983). The selective impairment of phonological processing: A case study. Brain and Language, 18, 128-174. Cavanaugh, J. (1972). Relation between the immediate memory span and the memory search rate. Psychological Review, 79, 525-530. Colle, H., & Welsh, A. (1976). Acoustic masking in primary memory. Journal of Verbal Learning and Verbal Behavior, 15, 17-32. Coltheart, M. (1980). Reading, phonological recoding and deep dyslexia. In M. Coltheart, K. Patterson, & J. Marshall (Eds.), Deep dyslexia (pp. 197-226). London: Routledge & Kegan Paul. Coltheart, M. (1984). Sensory memory. In H. Bouma and D. G. Bouwhuis (Eds.), Attention and performance X (pp. 259-285). London: Erlbaum. Conrad, R. (1964). Acoustic confusion in immediate memory. British Journal of Psychology, 55. 75-84. Crowder, R. (1971). The sounds of vowels and consonants in immediate memory. Journal of Verbal Learning and Verbal Behavior, 10, 587-597. Crowder, R. (1976). Principles of learning and memory. Hillsdale, NJ: Erlbaum. Crowder, R. (1978). Audition and speech coding in short-term memory: A tutorial review. In J. Requin (Ed.), Attention and performance VII. Hillsdale, NJ: Erlbaum.

92

Friedrich

Crowder, R., & Morton, J. (1969). Precategorical acoustic storage (PAS). Perception and Psychophysics, 5, 365-373. Darwin, C, & Baddeley, A. (1974). Acoustic memory and the perception of speech. Cognitive Psychology, 6, 41-60. Friedrich, F., Glenn, C, & Marin, O. (1984). Interruption of phonological coding in conduction aphasia. Brain and Language, 22, 266—291. Friedrich, F., Martin, R., & Kemper, S. (1985). Consequences of a phonological coding deficit on sentence processing. Cognitive Neuropsychology, 2, 385—412. Levy, B. (1971). Role of articulation in articulatory and visual short-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 123-132. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86. McClelland, J., & Rumelhart, D. (1981). An interactive activation model of context effects in letter perception: 1. An account of basic findings. Psychological Review, 88, 375-407. McLeod, P., & Posner, M. (1984). Privileged loops from percept to act. In H. Bouma & D. Bouwhuis (Eds.), Attention and performance X. Hillsdale, NJ: Erlbaum. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view. In H. Bouma & D. Bouwhuis (Eds.), Attention and Performance X (pp. 327-350). Hillsdale, NJ: Erlbaum. Morton, J., Crowder, R. G., & Prussia H. A. (1971). Experiments with the Stimulus Suffix Effect. Journal of Experimental Psychology, 91, 161-80. Nairne, J. S., and Crowder, R. G. (1982). On the locus of the Stimulus Suffix Effect. Memory and Cognition, 10, 350-357. Patterson, K. (1982). The relation between reading and phonological coding: Further neuropsychological observations. In A. Ellis (Ed.), Normality and Pathology in Cognitive Functions (pp. 77-111). London: Academic Press. Pisoni, D. (1973). Auditory and phonetic codes in the discrimination of consonants and vowels. Perception and Psychophysics, 13, 253-260. Pisoni, D. (1975). Auditory short-term memory and vowel perception. Memory and Cognition, 3, 7-18. Posner, M., & Henik, A. (1983). Isolating representational systems. In J. Beck, B. Hope, & A. Rosenfeld (Eds.), Human and machine vision. New York: Academic Press. Rumelhart, D., & McClelland, J. (1982). An interactive activation model of context effects in letter perception: 2. The context enhancement effect and some tests and extensions of the model. Psychological Review, 89, 60-84. Saffran, E., & Marin, O. (1975). Immediate memory for word lists and sentences in a patient with deficient auditory short-term memory. Brain and Language, 2, 420—433. Salame, P., & Baddeley, A. (1982). Disruption of short-term memory by unattended speech: Implications for the structure of working memory, Journal of Verbal Learning and Verbal Behavior, 21, 150-164. Schneider, W., & Shiffrin, R. (1977). Controlled and automatic human information processing: I. Detection, search and attention. Psychological Review, 84, 1-66. Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729-735. Shallice, T., & McCarthy, R. (1985). Phonological reading: From patterns of impairment to possible procedures. In K. Patterson, }. Marshall, & M. Coltheart (Eds.), Surface dyslexia (pp.' 361-397). London: Erlbaum. Shiffrin, R., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 127-160. Spoehr, K., & Corin, W. (1978). The stimulus suffix effect as a memory coding phenomenon. Memory and Cognition, 6, 583-589.


93

Vallar, G., & Baddeley, A. (1984a). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Vallar, G., & Baddeley, A. (1984b). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Cappa, S. (1987). Articulation and verbal short-term memory: Evidence from anarthria. Cognitive Neuropsychology, 4, 55-78. Warrington, E., & Shallice, T, (1969). The selective impairment of auditory verbal short-term memory. Brain, 92, 885-896. Waters, G., Komoda, M., & Arbuckle, T. (1985). The effects of concurrent tasks on reading: Implications for phonological recoding. Journal of Memory and Language, 24, 27-46. Watkins, M. (1977). The intricacy of memory span. Memory and Cognition, 5, 529-534. Watkins, M , & Watkins, O. (1977). Serial recall and the modality effect: Effects of word frequency. Journal of Experimental Psychology: Human Learning and Memory, 6, 712—718. Wickelgren, W. (1969). Auditory or articulatory coding in verbal short-term memory. Psychological Review, 76, 232-235.

4. Electrophysiological measures of short-term memory ARNOLD STARR, GEOFFREY BARRETT, HILLEL PRATT, HENRY. J. MICHALEWSKI, A N D JULIE V. PATTERSON

4.1. Introduction Event-related potentials can serve as objective indicators of the neural processes that occur during cognition with various components being correlated with particular mental activities such as attention (Hillyard, Hink, Schwent, & Picton, 1973), ease of discrimination (Naatanen, Simpson, & Loveless, 1982), stimulus classification (Donchin, 1981), and semantic incongruity (Kutas & Hillyard, 1980). These methods of study are noninvasive, as they utilize scalp electrodes, and the testing procedures are often derived from those used in experimental psychology. An underlying assumption of this type of work is that the event-related potentials reflect activity in the neural systems used during the performance of the cognitive tasks and thus provide information about the state of the nervous system for correlation with measures of performance. We have been studying event-related potentials in humans during the act of remembering using tasks that primarily test short-term memory. We have utilized a variant of a probe identification task (Steinberg, 1966) in which subjects are presented with a sequence of items to remember, followed by a probe that the subject must identify as being or not being a member of the memorized set. The task has been carried out both with normal subjects and with patients with disordered auditory short-term memory. The results we will present here bear on some of the models of short-term memory proposed in other chapters in this volume and, in particular, provide insights into the mechanisms underlying disordered auditory short-term memory in human beings following brain lesions.

4.2. Methods The subjects were 11 normal young individuals (average age, 29), 11 older subjects (average age, 66), and 3 patients with a reduced digit span that was poorer with auditory

Arnold Starr was supported in part by Grant 11876, National Institutes of Health.

94

Electrophysiological measures of short-term memory

.°

1.2 n

1.2

Probe

1

3sec

»

"nine"

STIMULI 0 ° 0 «

1 CM

1

Memorized Set

"four"

i

95

0

„

•

i

i

RESPONSE

Negative

n Reaction Time

Figure 4.1. Memory scanning paradigm: probe identification task for collecting eventrelated potentials related to short-term memory. Each trial begins with the word start followed by the items to be remembered (i.e., the memory set), then a brief pause, followed by the probe. The subject responds by a button press to indicate whether the probe item was a member (positive) or was not a member (negative) of the memory set. The number of items to be remembered, the mode of presentation (acoustic or visual), and the nature of the items (digits or musical notes) could be varied.

than with visual presentation. Details of the patients' clinical and neuropsychological examinations are summarized in a later section. Two methods were used to test short-term memory (see Starr and Barrett, 1987, and Pratt et al., 1989a, b, for details). The first consisted of a probe identification task that tested whether a set of items (such as digits) had been memorized. This task assesses, in part, the functions of an auditory-verbal short-term store. The second method required the identification of a particular tonal signal that occurred infrequently and was interspersed among another frequently occurring tonal signal differing from the target tone in pitch. We suggest that this task primarily engages a sensory acoustic store, since the stimuli have a constant pitch without the complex acoustical features characteristic of phonemic or lexical signals. Both of the tasks are suitable for recording brain potentials because they do not require a verbal output and thereby minimize unwanted and spurious potentials from tongue and facial movements. In the probe identification task (Figure 4.1), a computer initiates a trial with the word start followed in approximately 1 sec by an item for memorization. Subsequent items are presented at approximately 1-sec intervals. Two seconds after the last item of the set, a probe item is presented that the subject identifies, by an appropriate button press, as being or not being a member of the immediately preceding memorized set.1 The items and probe were arranged to differ from one trial to the next. Response accuracy and the time between the probe and the button press (reaction time, RT) were recorded. The number of items in the set was varied between one

96

Starr, Barrett, Pratt, Michalewski, and Patterson

and five and presented either in the auditory or visual modality. The trials were presented in blocks of 20 in which both the number of items in the memory set and their modality of presentation were fixed. The probability of a probe's being a member of the immediately preceding memory set was 0.5. The items selected for study were arabic numbers (1-9) or musical notes (middle C through D, one octave higher). At least 40 probe trials were presented in every tested condition: Stimulus modality (auditory or visual), number of items to be remembered (one, three or five), and the nature of the items (digits or musical notes) were factorially combined. In the tone identification task a rare tone (P = 0.2) of 1500 Hz was randomly interspersed with a frequent tone of 2000 Hz. The signals were presented every 1.2 sec, and the subject was instructed to press a response button when the rare tone occurred. A total of 200 trials were presented with 40 trials containing the rare tone and 160 trials containing the frequent tone. Measures of accuracy and RT were recorded. Event-related potentials were recorded from electrodes situated in the midline and laterally over the scalp (Fz, Cz, Pz, C3, C4, T5, T6) referenced to linked electrodes on the ear lobes. Eye movements were monitored by electrodes at the glabella and below the lateral aspect of the right eye to serve as the means for excluding those trials contaminated by potentials originating from the eyes. Electrical activity accompanying the presentation of each memory-set item and the probe was amplified, filtered, digitized, and stored on the computer for subsequent analysis. The time base of the event-related potential was 1000 msec for the rare target tones and 1280 msec for the probes. A 120 msec prestimulus baseline was included in each of these time bases. Each trial was visually examined for the presence of eye movements. If they were absent and the subject had responsed accurately, that trial was added to others recorded for the same stimulus variables to comprise an averaged event-related potential. The averaged potentials were digitally low-pass filtered at an upper cutoff of 17 Hz to facilitate subsequent measurements of amplitude and latency of the peaks of the components. The behavioral data were analyzed with regard to response accuracy and to latency of reaction time as a function of the number of items in the memory set, the modality, and the type of item to be remembered (verbal and nonverbal). The accompanying event-related potentials were measured for the latency and amplitude of the components and the data were similarly sorted. The components were identified by their polarity (P or N for positive or negative) and their latency in milliseconds (e.g., N100 refers to a negative-going potential occurring at 100 msec). Statistical analyses of the behavioral and electrophysiological data were conducted with repeated measures analysis of variance (ANOVAs) and post hoc comparisons of the means using a level of p < .01 as a measure of statistical significance.


97

4.3. Results 4.3.1. Normals A feature common to the event-related potentials accompanying correctly identified probes was a sustained (circa 800-msec duration) high-amplitude (approximately 15 uV), positive component peaking at a latency of about 450 msec that was largest in the midline over the parietal region without any gross hemispheric asymmetry. This positivity began approximately 200 msec earlier at a latency of 250 msec after the probe's presentation. We chose to take measures of the peak amplitude and latency of this potential, since this point on the positive shift was easier to define than the onset point. This distinction will become important later in this chapter when we try to relate this potential to memory processes. Figure 4.2 contains both the grand (also known as the group) and individual averages to correctly identified probes recorded from the midline parietal electrode (Pz) when the memory set contained one item in three different test situations: auditory digits, visual digits, and musical notes presented acoustically. The event-related potentials consist of N100, as well as P200 components whose latency, amplitude, and scalp distribution varied considerably with the modality and type of probe. In contrast, the P450 component was relatively unaffected by these variables but was affected by the size of the memorized set. These differences provide an initial separation of the components into two types: (a) exogenous (N100, P200), dependent on stimulus features, and (b) endogenous (P450), dependent on cognitive activity of the subject (Donchin, Ritter, & McCallum, 1978). This distinction is supported by the observation that if the sequence of memory items and probes were presented while the subject was attending to an unrelated task, the event-related potentials to the probes consisted of similar N100 and P200 components while the P450 component was absent. Accuracy was close to 100% for all set sizes for digits and for the one-item musical note memory sets but diminished to 82% and 77% for the three- and five-item musical note tasks, respectively. The functions relating reaction time and P450 latency to the number of items in the memory set had several significant differences (Figure 4.3). First, while the slopes for P450 latency were similar for the different types of memory items (digits and musical notes) and amounted to approximately 25 msec/item, the corresponding reaction time slopes differed both from the P450 measures as well as from each other. The RT slopes were approximately 50 msec/item for digits and 100 msec/item for musical notes. These results indicate that during the classification of a probe as a member of the memorized set there is a dissociation between brain potential measures (latency of P450) and reaction time. The lengthening of RT without a corresponding change in P450 suggests that the P450 event-related potential component reflects a processing stage relatively early in the short-term memory

98

Starr, Barrett, Pratt, Michalewski, and Patterson Pz

EOG

Auditory Digits

Visual Digits

Musical Notes

Figure 4.2. Event-related potentials associated with correctly identified probes obtained from a group of normal subjects in a one-item task using auditory digits, visual digits, and musical notes. The tracings on the left were recorded from a midline scalp electrode in the parietal region (Pz) referenced to linked ear lobes; the tracings on the right were obtained from electrodes above and below the right eye and are an indicator of eye movements (electrooculogram, EOG). Each pair of tracings consists of the grand average (above) and the superimposed separate averages (below) from 11 normal young subjects. The components occurring within 200 msec of the probe's appearance (indicated by the vertical bar near the beginning of the trace) represent sensory components (N100, P200). The high-amplitude positive deflection whose peak is indicated by the small vertical tick mark is the P450 component. scanning process that is independent of several stimulus features that significantly affect RT. There is additional evidence from our studies of short-term memory in an aged population that the P450 component has features that are distinguishable from RT. Figure 4.4 plots the relationship between P450 latency and RT as a function of

Electrophysiological measures of short-term

memory

99

400

-z noo e IOOO «

900

o

800

£

700

O Auditory Digits • Visual Digits • Musical Notes

5 eoo g

500

S

400

1

3 5 Memory Set Size

Figure 4.3. Reaction time (top graph) and P450 peak latency {bottom graph) as a function of set size for the memorized items in three conditions; auditory digits, visual digits, and musical notes. The slopes of the functions for RT differ for the musical notes compared to auditory and visual digits (100 msec/item vs. 50 msec/item respectively) and are also different from those for the P450 latency (25 msec/item).

memory-set size in an old and young population when the items to be remembered were digits presented in the auditory modality. Note that both the absolute values of RT and the slopes of the functions relating RT to memory-set size are accelerated with age (approximately 350-msec increase for absolute RT and 30 msec/item for the memoryscanning slope). In contrast, both the absolute latencies and the slopes of the functions relating P450 latency with set-size were not significantly different between the young and old subjects. These data suggest that the P450 represents a stage of processing in short-term memory that is relatively unaffected by age, whereas RTs that reflect both response selection and motor processes are distinctly affected with aging. There was a decrease in amplitude of P450 in the old population compared to the young group. The amplitude decrement of the P450 with aging may be related to the increase in absolute latency of RT, but both changes are more likely unrelated and reflect nonspecific effects of aging on brain function rather than being specific for aging effects on short-term memory. There are several features of the P450 component that bear on models of normal functioning of short-term memory. First, in the young normal subjects, there were no consistent differences in latency or amplitude between the event-related potentials

100

Starr, Barrett, Pratt, Michalewski, and Patterson Auditory Digits 1200

400

l

3

5

Memory Set Size Figure 4.4. Plots of RT and P450 peak latency to auditory digit probes as a function of set size for a young and an old group. Note the large increase both of absolute RT values and the slopes of their functions in the old group relative to the young group without any significant change in P450 latency with age. accompanying positive and negative probes, indicating that the neural processes governing such a choice are not distinguishable in the present evoked-potential data. In contrast is the finding that there are correlates of P450 latency and amplitude that correspond to the recency effect in short-term memory that is, those items most recently memorized are recalled with more facility than items presented earlier. A recency effect was apparent in the five-item digit list presented in the auditory modality with both RT and peak P450 latencies being shorter and P450 amplitude being larger to probes that matched the last item in the list compared to those representing items at earlier positions of the list (Figure 4.5). These studies in normal subjects demonstrate that a probe identification task can be used to measure features of short-term memory with both behavioral indices (accuracy and RT) as well as brain potential attributes (P450 peak latency and amplitude) that provide complementary insights into the workings of short-term memory. Reaction time and accuracy reflect the final output of the neural systems subserving short-term memory, whereas the latency of the peak of P450 brain potential reflects an early stage in the functioning of short-term memory that is relatively independent of stimulus features and subject age but sensitive to the memory load.

Electrophysiological measures of short-term

memory

101

Serial Position

1

2

3

4

5

Position 500

1000

msec

Figure 4.5. Recency effect in short-term memory of both RT and event-related potentials. The tracings in A are the grand averages compiled from 11 normal young subjects of the event-related potentials from Pz to correctly identified auditory digit probes as a function of their serial position in a five-item memory set. The amplitudes and peak latencies of the P450 component derived from the average of each subject have been measured and the means plotted in B; the mean reaction time data are presented in C. Only for the last item of the memory list (Item 5) were the amplitude and peak latency of the P450 component and the RT significantly different from each of these measures for the other items.

4.3.2. Patients Three patients with disordered auditory short-term memory were studied using behavioral (RT) and event-related potential techniques (Starr & Barrett, 1987). Table 4.1 presents their test scores on the Wechsler Adult Intelligence Scale (WAIS). Note the diminution of their auditory digit span compared to their visual digit span. The first two patients had stable and stationary deficits, whereas Patient 3 was studied acutely 3 days after a vascular lesion of the left hemisphere.2 The three patients had lesions of the left hemisphere involving the posterior temporal lobe and the inferior portions of the parietal lobe. Table 4.2 presents measures of the patients' accuracy and reaction times in the probe identification tasks using digits along with five age-matched controls. For both of the one-item digit tasks the absolute reaction times were slightly and insignificantly longer (approximately 150 msec) for the patients compared to the controls. With the threeitem task accuracy diminished to 69% using the auditory modality, whereas it remained high (96%) in the visual modality. RT increased on the three-item auditory task more than 250 msec to 966 msec, which was significantly different from normals, who showed a 76 msec increase to 549 msec (p < .01). The comparable RT values for the three-item visual task showed that the patients increased only slightly more than the normals (104 msec vs. 63 msec; p > .1). Thus, increasing the memory load from one to three items produced a memory-scanning rate for these patients of 166 msec/item in the

102

Starr, Barrett, Pratt, Michalewski, and Patterson Table 4.1. Neuropsychological Scores Patients Tests

1

Wais Verbal" Arithmetic Similarities Digit Span Vocabulary

61 4 7 1 6

8,7 8 11 3 9

61 2 6 0 7

Wais Performance

93 9 7 10

101 9 7 12

92 7 12 3

1 5

4 5

2 3

Picture Completion Block Design Picture Arrangement

3

2

Digit Repetition

Auditory Visual*

"Wechsler Adult Intelligence Scale. Subjects age adjusted with 10 as the mean. ^Tested with manually presented cards containing individual digits.

auditory mode compared to only 51 msec/item in the visual mode (p < .01). The scanning rate measures obtained for the normal subjects were approximately 35 msec/item for both the auditory and visual modalities. The data clearly define an abnormality of auditory short-term memory functions in these patients. Although the patients' performance in the visual modality was poorer than normal, the difference did not achieve statistical significance. The event-related potentials to the probes provide complementary evidence of the patients' profoundly disordered auditory short-term memory (Figure 4.6). The P450 component, when compared to normals, was either abnormally small or absent in the three-item auditory task and within normal limits in the three-item visual task. A more striking finding occurred on the one-item task when P450 was abnormally small with auditory but not visual presentations, even when the patients were performing accurately and at comparable normal RTs in both modalities. Table 4.3 lists the amplitudes and latencies of the P450 component in these different conditions. An abnormally small P450 amplitude in the one-item auditory task was also evident with the negative probes that were correctly identified as not belonging to the memory set. The data demonstrate a defect in these patients' auditory short-term memory processes detectable at a time when their performance (response accuracy and RT on a one-item load) could not be distinguished from normal. The ability of the patients to compensate for their deficit of auditory short-term memory processes at the level of overt

Table 4.2. Behavioral measures: accuracy (%)/reaction times (in msec) to correctly identified positive probes and rare target tones Visual

Auditory

Patients 1 2 3 X SD Normals SD

1-item

3-item

Rare

1-item

3-item

Rare

(100)/609 (100)/629 (100)/664 (100)/634 /111 (100)/473 / 16

(62)/915 (81)/862 (77)/1121 (73)/966a /385 (100)/549 / 82

(100)/578 (100)/426 (100)/414 (100J/473 /HO (100)/344 / 50

(100)/477 (100)/554 (100)/756 (100)/596 /146 (100)/458 / 11

(96)/607 (100)/710 {91)1119 (96)/699 /237 (100)/521 / 97

(100)/562 (100)/446 (100)/433 (100)/480 / 88 NT NT

NT, not tested; X, means; SD, standard deviations. a p < .01; patients compared with normals.

104

Starr, Barrett, Pratt, Michalewski, and Patterson Hit) AUDITORY

CRi

Hit3

Normal Hit3

CR3

450

450

450

1

450

450

Normal Grand Average

r^

Patient

yy

X VISUAL

j

Mormal Grand Averaae —

450

V

\

450

"

\

450

^

450

450

J_

Patient

sA

j

/V

~V_

Figure 4.6. Event-related potentials to correctly identified probes in patients with disordered auditory short-term memory using auditory and visual presentations of a oneor three-item memory list containing digits. The grand average from the normals is plotted above the individual tracings from the patients for positive (Hit) and negative (CR, correct rejection) probes. The size of the memory list is indicated by the numbers 1 or 3 following the probe type designation. The tracings from the normal individuals that comprise one of the grand averages (the three-item memory list to positive probes) is to the right (Normal Hit3). For auditory presentation the patients have a marked attenuation of a positive potential at approximately 450 msec, indicated by the vertical line at this latency, for both the one- and three-item lists compared to the normal subjects. In contrast, the potentials during visual presentations were of normal amplitude with the one-item list and reduced in amplitude slightly with the three-item list (see Table 4.3 for measures).

performance corresponds to other observations of the effects of neurological lesions that are not apparent unless special sensitive tests are used. An example of this phenomenon is the finding of abnormal visual evoked potentials in patients with otherwise asymptomatic optic neuritis (Halliday, McDonald & Mushin, 1973). An examination of each of the patients' individual event-related potential trials revealed an occasional P450 component to be of large amplitude, comparable to normals. These trials appeared to be associated with a relatively fast RT. Thus, new averages were computed for the patients and controls in which the trials were divided into three averages consisting of the fastest, the middle, and the slowest thirds of the RT


105

Table 4.3. Peak latency in msec and amplitude in fiV of P450 in the digit probe task and P300 in the rare-tone task (means and standard deviations in parentheses) Positive Probe Type Auditory Patients 1-item 3-item Rare tone Normals 1-item 3-item Rare tone Visual Patients 1-item 3-item Rare letter Normals 1-item 3-item Rare letter

Negative

Latency

Amplitude

Latency

Amplitude

441(19) 642(62) 387(34)

5.7(1.9)* 5.6(1.6)* 9.9(4.1)

438(95) 646(46)

4.3(0.9)" 6.6(3.9)"

NT

NT

488(83) 602(35)

14.9(5.3) 16.9(7.0)

NT

NT

455(91) 519(66) 342(18)

14.6(2.2) 13.6(5.2) 13.1(7.8)

436(31) 476(71) 441(25)

14.8(1.8) 10.2(1.5) 16.6(7.0)

545(74) 558(53)

10.1(3.1) 6.5(1.1)

NT

NT

424(51) 450(31)

18.0(3.5) 17.2(3.7)

469(68) 535(38)

15.6(6.4) 16.8(4.2)

NT

NT

NT

NT

NT, not tested or not applicable. a p < 0.01; patients compaired with normals. for that subject. Figure 4.7 compares in one patient and one normal control the resultant averages. Clearly, the amplitude and latency of the patient's P450 varied significantly (for latency p < .001, for amplitude p < .05) as a function of RT, both for auditory and for visual presentations, whereas these measures of P450 varied little with RT in the normal subjects. For the patient group the amplitude of the auditory P450 increased 11 fiV, going from 5 fiV for the slowest third of RTs to 16 fiV for the fastest third of RTs (the latter value is normal), while P450 latency decreased by 180 msec. The comparable values for the normal group was an increase of only 3 fiV (from 14 to 17 /iV) and a P450 latency reduction of only 60 msec. Thus, P450 could be considered normal in amplitude and latency in these patients when only those trials comprising their fastest RTs were averaged. Their deficit of short-term memory, viewed from the perspective of the P450 component of event-related potentials, has a fluctuating rather than a fixed character, at least with a one-item memory load. This is in keeping with the patients' anecdotal accounts of the varying nature of their memory for these items as being preserved on some trials and fleeting on others. It would be of interest if we could define the amplitude and latency of the P450 component on a trial-by-trial basis to obtain measures of their variability for correlation with behavior.

106


a.

NORMAL

PATIENT Auditory-1 50

A

Auditory-1

Visual-1 450

A

450

Visual-1 A

450

Fast

Med Slow

Grand Average

b.

Visual

i

20

or \)

-o

15

10 -

_

5 Fast

Med Slow

Fast

Med

Slow

Fast

Med

Slow

Figure 4.7. The effect of RT on the amplitude and latency of the P450 component to correctly identified probes. In a, the trials have been averaged according to the fastest middle and slowest thirds of RTs for both auditory and visual one-item memory sets. The grand average of all of the trials is plotted above. A vertical line is at the 450-msec latency to aid in visualizing the changes in latency. Plots of measures of RT, latency, and amplitude of P450 for the three patients and the controls are shown as a function of the trimester of RTs in b. Note the disproportionate change in amplitude for the P450 in the auditory mode as a function of RT for the patients compared to the normal subjects.

There is indirect evidence in the patients that the encoding or sensory aspects of auditory short-term memory functions were normal. The exogenous or sensory portions of the event-related potentials (NlOO, P200) to the auditory presentation of the digits in the memory-set were normal with regard to latency and amplitude. We interpret these exogenous components as deriving from activity in those portions of auditory cortex engaged in sensory processing, that is, those portions responsive to


107

physical attributes of the stimulus such as spectrum, intensity, and so on. Thus, the alterations in the patients' P450 component are unlikely to be due to a sensoryprocessing disorder of auditory cortex but, more plausibly, to another type of deficit of auditory processing such as storage or retrieval functions related to auditory short-term memory. Other results further help to clarify these possibilities by showing that not all forms of the auditory storage over the short term were abnormal in these patients. In the infrequent-tone identification task, both the amplitude and latency of the endogenous or cognitive P300 component associated with the identification of the rare tone were normal. The "remembering" required for accurate response to the rare tone must involve an acoustic store that is different from the store involved in "remembering" the more complex acoustic stimulus features of a word. Thus, these electrophysiological results suggest that the patients' deficit of auditory short-term memory functions appears to be localized primarily to an auditory-verbal short-term memory store and not to an auditory sensory store.

4.4. Discussion The study of short-term memory processes has traditionally utilized behavioral measures of accuracy and RT to define the variables influencing encoding, storage, and retrieval processes (Peterson & Peterson, 1959; Steinberg, 1966). Event-related potentials recorded from the scalp of individuals during short-term memory tasks have been introduced as a means of gaining insight into neural processes subserving shortterm memory that precede the behavioral output (Marsh, 1975; Ford, Roth, Mohs, Hopkins & Kopell, 1979). It has been necessary to employ tasks that minimized movements to ensure that the event-related potentials reflected brain activity and not muscle or eye movement artifacts. A modification of the probe identification task (Sternberg, 1966) has been used by many investigators testing visual (Marsh, 1975; Gomer, Spicuzza, & O'Donnell, 1976; Adam & Collins, 1978) or auditory (Gaillard & Lawson, 1984; Starr & Barrett, 1987) short-term memory. The results from these electrophysiological investigations have indicated that a high-amplitude positive component appears at a latency of approximately 450 msec during the act of remembering whether or not the probe being classified is a member of a previously memorized set. Our studies in normal subjects (Pratt et al. 1989a, b) summarized in this chapter confirm that the P450 peak latency was affected by the size of the memory load but in a different fashion from RT: P450 latency increased more gradually (approximately 25 msec/memory item) than did the RT functions (50-100 msec/item). Moreover, our latest data indicate that P450 latency functions are relatively independent of the nature of the item being remembered (digits vs. musical notes) and the subject's age (when auditory digits are classified), whereas RT is very greatly influenced by these variables

108


(Ford et al, 1979; Pratt et al., 1989b). We have concluded that the P450 component reflects activity of those neural processes relating to memory scanning of the stored items prior to their comparison with the probe and response selection. The P450 represents activity in an early stage of the short-term memory processes and is thereby relatively unaffected by those subsequent neural stages of short-term memory that influence RT. We would like to stress that the positive potential shift designated by its peak latency as P450 is initiated at approximately 250 msec after the probe's appearance and thus can be considered to represent an "early stage" of memory. A strength of the electrophysiological measure of short-term memory is in its application to clinical disorders of short-term memory (Warrington & Shallice, 1969). We have had the opportunity to study several patients with a selective impairment of auditory short-term memory using event-related potentials during the probe identification task. The functional localization of these patients' deficits has been postulated as occurring at several sites. First, there may be an abnormality of encoding of auditory-verbal information into an auditory store or, as suggested by Allport (1984), into a store that is phonological. Second, the store itelf could have an abnormally small capacity and/or rapid decay (Warrington & Shallice, 1969; Shallice & Warrington, 1970). Third, there could be alterations in the manner of retrieval of the items from the store(s) (a possibility not explicit in the literature but certainly reasonable), or, fourth, there could be problems in utilizing the retrieved items for response selection (Kinsbourne, 1972). Behavioral methods have suggested that the locus of these patients' deficits is in a verbal and not a sensory auditory store, since they can remember environmental sounds but not lists of digits (Shallice & Warrington, 1974). The possibility of this disorder's being at the output stage has had several supporters (e.g., Kinsbourne, 1972; Ellis, 1979). The event-related potentials obtained from these patients indicate that encoding into the auditory store is normal: The sensory components of the event-related potential to the digits in the memory set were normal, as were the sensory components of the potentials evoked by tones used in the rare-signal classification task. Moreover, in this latter task the potentials accompanying correct discrimination (the P300) were normal both in latency and amplitude. Thus, both the encoding of auditory information (tones and digits) as well as the retrieval from at least a sensory auditory short-term store (the one used for classification of a rare tone) were normal. We reason that the rare-tone task engages short-term memory processes because if the interval between the tones were increased, a point would be reached at which the ability to detect the change of stimulus from frequent to rare would certainly be impaired. This interval is probably in seconds rather than in minutes. In contrast, the potentials associated with the scanning of an auditory short-term store, as reflected by the P450 component to the digit probes, were abnormally small. The abnormality was evident even with a one-item load in the auditory mode when


109

performance was not distinguishable from normal. These data are compatible with the presence of a modality-specific defect of short-term memory. It is most economical to suggest that our patients' deficit lies primarily in an auditory-verbal short-term store that may interact to some extent with a visual short-term store of limited capacity. Their sensory auditory short-term store appears normal. There are many unexplained details about the electrophysiological analysis of shortterm memory processes both in normals and in these patients. For instance, we are exploring the effects of interference (Peterson & Peterson, 1959) on the event-related potentials. Will such an analysis provide a model system in normals for studying the clinical disorders of short-term memory? Can the digit probe identification task be used to assess short-term memory processes in demented patients using several types of stimuli (digits, notes, words, visual figures, etc.)? The results from the studies reported in this chapter provide encouragement that event-related potentials do yield significant data relevant for understanding the neural processes underlying short-term memory in human beings.

Notes 1. Two buttons were used in the normative studies; one of the buttons was to be pressed if the probe was a member of the preceding memory set and the other button was to be pressed if the probe were not a member of the preceding memory set. For the clinical studies of patients with disordered auditory digit span and their age-matched controls only a single button response was made to probes that were members of the preceding memory set. The button was not to be pressed if the probe were not a member of the preceding memory set. 2. Cases 1 and 2 have been described in other studies as RAN (see chapters 1 and 7) and JB (see chapters 1 and 8), respectively.

References Adam, N., & Collins, G. I. (1978). Late components of the visual evoked potential to search in short-term memory. Electroencephalography and Clinical Neurophysiology, 44, 1 4 7 - 1 5 6 .

Allport, D. A. (1984). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.) Attention and performance X: Control of language functions (pp.

313-325). London: Erlbaum Donchin, E. (1981). Surprise!... Surprise? Psychophysiology, 18, 493-513. Donchin, E., Ritter, W., & McCallum, W. C. (1978). Cognitive E. Psychophysiology: the endogenous components of the ERP. In E. Callaway, E. Tueting, S. Koslow, (Eds.), Eventrelated brain potentials in man (pp. 415-430). New York: Academic Press. Ellis, A. W. (1979). Speech production and short-term memory. In J. Morton & J. C Marshall (Eds.), Psycholinguistic series: Vol. 2. Structures and processes (pp. 157—187). London: Elek.

Ford, J. M., Roth, W. T., Mohs, R. C, Hopkins, W. F., & Kopell, B. S. (1979). Event-related potentials recorded from young and old adults during a memory retrieval task. Electroencephalography and Clinical Neurophysiology, 47, 4 5 0 - 4 5 9 .

Gaillard, A. W. K., & Lawson, E. A. (1984). Evoked potentials to consonant-vowel syllables in a memory scanning task. Annals of the New York Academy of Sciences, 425, 204—209.

110


Gomer, F. E., Spicuzza, R. J., & O'Donnell, R. D. (1976). Evoked potential correlates of visual item recognition during memory-scanning tasks. Physiological Psychology, 4, 61-65. Halliday, M., McDonald, I., & Mushin, J. (1973). Visual evoked response in diagnosis of multiple sclerosis. British Medical Journal, 4, 661-664. Hillyard, S. A., Hink, R. F., Schwent, V. L, & Picton, T. W. (1973). Electrical signs of selective attention in the human brain. Science, 182, 177-ISO. Kinsbourne, M. (1972). Behavioural analysis of the repetition deficit in conduction aphasia. Neurology, 22, 1126-1132. Kutas, M , & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207, 203-205. Marsh, G. R. (1975). Age differences in evoked potential correlates of a memory-scanning process. Experimental Aging Research, 1, 3-16. Naatanen, R., Simpson, M., & Loveless, N. E. (1982). Stimulus deviance and evoked potentials. Biological Psychology, 14, 53-98. Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193-198. Pratt, H., Michalewski H. ]., Barrett, G., & Starr, A. (1989a). Brain potentials in a memoryscanning task. I. Modality and task effects on potentials to the probes. Electroencephalography and Clinical Neurophysiology, 72, 407-421. Pratt, K, Michalewski, H. J., Patterson, J. V., & Starr, A. (1989b). Brain potentials in a memoryscanning task. II. Effects of aging on potentials to the probes. Electroencephalography and Clinical Neurophysiology, 72, 507-517. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261-273. Shallice, T., & Warrington, E. K. (1974). The dissociation between short term retention of meaningful sounds and verbal material. Neuropsychologia, 12, 553-555. Starr, A., & Barrett, G. (1987). Disordered auditory short-term memory in man and event-related potentials. Brain, 110, 935-959. Sternberg, S. (1966). High-speed scanning in human memory. Science, 153, 652-654. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory verbal short-term memory. Brain, 92, 885-896.

Part II. Phonological short-term memory and other levels of information processing: studies in brain-damaged patients with defective phonological memory Part II of this book comprises four chapters dealing with the relationship between phonological short-term storage and the other processes involved in the retention of verbal information. All four chapters report data from individual case studies of shortterm memory patients. They emphasize that immediate retention of verbal material involves more than one level of representation and attempt to specify how these different levels contribute. Phonological storage is not carried out in a single isolated system (see, e.g., Monsell, 1984; Barnard, 1985). The storage system responsible interacts with low-level (acoustic, nonphonological) components (see Berndt & Mitchum, chapter 5; Campbell, chapter 11). Within the phonological level, interrelated input and output subcomponents may be distinguished (see Campbell, chapter 11; Howard & Franklin, chapter 12) and possibly also lexical and prelexical ones (see Saffran & Martin, chapter 6). At a higher level of processing, interactions with lexical-semantic (Saffran & Martin, chapter 6) and syntactic-semantic (Butterworth, Shallice, & Watson, chapter 8) systems with longterm memory properties may occur. Finally, short-term memory tasks also involve highly controlled executive subcomponents (see Baddeley, chapter 2; McCarthy & Warrington, chapter 7; Craik, Morris, & Gick, chapter 10; Crain, Shankweiler, Macaruso, and Bar-Shalom, chapter 18). The discussion of the relationships of the phonological short-term store with other functional components of mental functions is of course not confined to this part, but may be found, in a variety of theoretical approaches, in most chapters of the book. This may be taken as an indication of the highly interactive nature of the system. A first issue concerns a possible distinction between acoustic, nonphonological and phonological input codes in speech perception and immediate memory (see also Campbell, chapter 11). Within the framework of a multilevel of representation model, Berndt and Mitchum (chapter 5) suggest that immediate retention of verbal material may be supported by both nonlexical acoustic and lexical codes. They report a detailed study of EDE, a patient with a reduced memory span who, they argue, has an impairment at the phonological level. Berndt and Mitchum point out that when a given level of 111

112

Part II

representation is defective, residual performance should reflect contributions from other processing domains, such as auditory nonlexical and semantic systems (see related data from patient RE: Vallar, Basso, & Bottini, chapter 17). Berndt and Mitchum investigate these issues by assessing nonverbal auditory memory performance, the effect on span of an auditory suffix, and how variables such as frequency and imageability affect immediate recall. The following three chapters (Saffran & Martin, chapter 6; McCarthy & Warrington, chapter 7; Butterworth et al., chapter 8) report investigations into the role of higher-level (mainly semantic) representations in immediate retention of word lists and in sentence repetition and comprehension (chapters 7 and 8). Saffran and Martin (chapter 6) deal with the contribution to immediate retention of nonphonological lexical-semantic system(s). They assess the effects of lexical and semantic variables (word frequency and imageability; see also Berndt & Mitchum, chapter 5) and of semantic similarity on immediate memory performance for word lists. Saffran and Martin report findings from two patients who have a defective span, but who differ in the contribution made by nonphonological systems to immediate recall. They discuss their neuropsychological results in the light of evidence from normal subjects and suggest an explanation of the well-known lexical-semantic effects in immediate memory in terms of a multilevel interactive framework, which they distinguish from the more traditional short-term (phonological)/long-term (semantic) dichotomy (see, e.g., Baddeley, 1966a, b). McCarthy and Warrington (chapter 7) investigate the immediate repetition of word lists and sentences in three patients. They find different patterns of impairment between two patients who have a severe span deficit, associated with a comparatively preserved sentence repetition, and one patient who shows the complementary disorder. On the basis of this double dissociation, McCarthy and Warrington argue for a distinction between two short-term memory systems: (a) a first phonological component, involved in span for unrelated words; and (b) a second component, semantic in nature, which is an anticipatory and integrative processing system, involved in sentence repetition and comprehension. The on-line construction of a central cognitive representation for a given sentence, a level of processing that encompasses semantic, pragmatic, and other contextual information, requires the operation of the integrative system. The phonological short-term memory system is not required for on-line comprehension, but becomes important when backtracking over spoken input is needed. McCarthy and Warrington suggest that there are three conditions in which the on-line construction of a central cognitive representation cannot take place and a verbatim record provided by the phonological system needs to be available: (a) when the rate of information presentation is too great (e.g., the Token Test); (b) when extralinguistic assumptions bias the spoken message; and (c) when the achievement of an adequate central cognitive representation requires supplementary cognitive operations to be performed on the spoken input. Under these conditions, they argue, span-impaired patients (i.e., patients

Studies in brain-damaged patients

113

with a selective phonological memory deficit) should show a defective comprehension performance. McCarthy and Warrington's suggestion posits a putative role for phonological memory other than facilitating immediate, on-line, syntactic and lexical processing. Under specific circumstances the phonological store might cooperate with the integrative system in higher-level (semantic) mental activities such as reasoning and decision making. It should also be noted that other contributors distinguish between a passive phonological component, involved in span-type immediate retention, and more controlled systems, which may (Crain et al, chapter 18) or may not (Craik et al, chapter 8; see also Baddeley, chapter 2) be conceived as specific to speech comprehension. Butterworth et al. (chapter 8) investigate the role of higher-level nonphonological components in immediate verbal memory and address the issue of the role of the phonological short-term store in sentence comprehension and repetition. They explore the memory performance for lists of unrelated words and sentences of JB, a well-known short-term memory patient, and of matched control subjects both in immediate recall and in a filled delay (20-sec) condition, where the contribution from the phonological short-term store should be negligible. Butterworth et al. interpret their data within a multistore framework, distinguishing two mutually supportive components, which, in classic memory terms, have short- and long-term characteristics: a low-level store (A), in which information is coded phonologically and may be maintained for a limited amount of time; and a higher-level store (B), which holds syntactic—semantic representations, at least over 20 sec. Butterworth et al. argue, however, that their findings suggest a position different from the classical one in that retrieval from the two stores needs to be interactive, not independent. This leads them to introduce the concept of (in)adequately supported information. For instance, the elements of an input string would be adequately supported if it is possible to construct an interpretation with a very small number of semantic components, while a structure in which multiple elements have equivalent roles would be inadequately supported. The higher the level of representation that can be constructed, the better supported the elements will be. Conversely, the less adequately supported the elements of a given string are, the more important the contribution from the phonological A store becomes. Like McCarthy and Warrington, Butterworth etal. argue that the process of building up syntactic-semantic representations does not necessarily require the maintenance of a phonological record in the A store. In the case of adequately supported information, damage to the phonological short-term store would not affect comprehension. Short-term memory patients would, however, be impaired in the case of inadequately supported material, for example, sentences with a number of structurally interchangeable elements, like the Token Test. Butterworth et al. are concerned with the role of phonological, syntactic, and

114

Part II

semantic representations in immediate repetition and comprehension of sentences. They do not discuss the relative contribution of their two stores at the discourse level, even though they mention that, for instance, text materials and ordinary conversations may induce higher-level structures, where syntactic-semantic representations could be linked to mental models (see Johnson-Laird, 1983). This issue is, though, treated in McCarthy and Warrington's chapter. Their central cognitive representation includes not only the products of linguistic analysis but also information from preceding or anticipated speech and aspects of real-world knowledge, namely, expectancies. The verbatim record provided by a phonological short-term store may, they argue, be useful at a discourse level of processing, when the central representation cannot be constructed on-line and backtracking is needed.

References Baddeley, A. D. (1966a). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 362-365. Baddeley, A. D. (1966b). The influence of acoustic and semantic similarity on long-term memory for word sequences. Quarterly Journal of Experimental Psychology, 18, 302-309. Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In. A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). London: Erlbaum. Johnson-Laird, P. N. (1983). Mental models. Cambridge: Cambridge University Press. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view. A tutorial review. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 327-350). London: Erlbaum.

5. Auditory and lexical information sources in immediate recall: evidence from a patient with deficit to the phonological short-term store RITA SLOAN BERNDT AND CHARLOTTE C. MITCHUM

5.1. Introduction Many of the recent efforts to provide functional explanations for the deficits of patients with limited repetition span have used the theoretical framework available in the working memory model (Baddeley & Hitch, 1974; Baddeley, 1986). According to this formulation, a limited-capacity memory system composed of several distinct subcomponents is available for temporary storage of information that is needed for a variety of cognitive tasks. Recent modifications of the working memory model relevant to verbal tasks have postulated two separate components for the storage and maintenance of verbal information. An articulatory loop system, which provides for the subvocal rehearsal of verbal information in an articulatory code, has been supplemented in recent discussions by a more passive phonological short-term store, to which auditory-verbal information gains obligatory access. A variety of experimental results obtained with normal subjects necessitated the postulation of the nonarticulatory phonological short-term store (STS) (see Baddeley, 1983, for review). The most important of these for present purposes is the following: Prevention of subvocal articulation (which presumably blocks effective use of the articulatory loop) does not prevent the phonological similarity among to-be-remembered items from interfering with recall when stimuli are presented aurally. This "phonological similarity effect" is presumed to reflect phonological factors intrinsic to storage in some nonarticulatory component of the system. Another finding with normal subjects supports the articulatory basis of the rehearsal loop. Longer words (those that take longer to articulate) are not recalled as well as shorter words, but this "word length effect" can be removed by the prevention of The authors are grateful for the contributions of Barbara Ritgert, who tested the controls, of John Berndt, who prepared the auditory probe recognition task, and of Maryne C. Glowacki, who typed the manuscript. Special thanks to Eleanor Saffran for many useful discussions. This project is supported by NINCDS Grants ROI-NS21054 and KO4-NS-00851 to the University of Maryland Medical School.

115

116

Berndt and Mitchum

rehearsal through concurrent articulation. Thus, the advantage of short over long words in immediate recall presumably reflects the influence of an articulation-based rehearsal system. However, patients' failure to use the articulatory loop may not necessarily indicate that that system is impaired. Vallar and Baddeley (1984) have argued that their patient PV, who shows no tendency to use subvocal rehearsal, suffers from an impairment to the phonological STS. The articulatory rehearsal process is strategically not used, they argue, because without a functioning phonological STS, rehearsal would serve no purpose (for further discussion, see Shallice & Vallar, this volume, Chapter 1). Thus, the failure to find effects of word length in patients' repetition might indicate a failure of rehearsal or a reduction of the phonological STS. Several patients with apparent memory limitation who have been reported in recent years, in addition to PV, might be interpreted as suffering from a limitation at the level of this phonological short-term store (Shallice & Warrington, 1970; Shallice & Butterworth, 1977; Caramazza, Basili, Koller, & Berndt, 1981; Friedrich, Glenn, & Marin, 1984). However, since these patients had difficulty with phonological processing in other tasks (discrimination, production), it might be argued that their impairment is not so much with the maintenance of phonological information as with its registration. Allport (1984) tested several "conduction aphasic" patients with tasks designed to detect subtle deficits in phoneme discrimination. Allport argues that these patients' inability to repeat arises from an inability to represent stably the sounds of the language, as a consequence of a disruption to a "central" inventory of phonological word forms. Such a deficit would necessarily result in problems in both the expression and the reception of speech sounds. Although Allport does not distinguish these problems with registration of speech sounds from their maintenance, the processes and representations postulated are more clearly linguistic than memorial. There are other considerations concerning the storage of phonologically coded information that have not been given much attention in the working memory model, nor in discussion of patients with memory deficits (but see Saffran & Marin, 1975). One of these is the relationship between the phonological code that is presumably stored when visually or aurally presented words are to be repeated, and the auditory trace that is argued to persist for some period of time following the aural presentation of stimulus items. A pre-phonological, auditory memory of this spoken information was postulated to account for the superior recall of aurally presented over visually presented items by normal subjects, which is especially evident at the ends of lists ("recency" effects) (Crowder & Morton, 1969). Information from this auditory trace was argued to supplement, in conditions of verbal presentation, the phonological code that could be generated with written as well as verbal presentation. A second type of support for the separate existence of an auditory code is available from studies of the error patterns that occur with auditory and written presentation. Intrusion errors produced in serial recall tasks tend to preserve auditory, nonsegmental features of the targets (stress pattern,

Information sources in immediate recall

117

number of syllables, and identity of stressed vowel) significantly more often with aural than with visual presentation (Drewnowski & Murdock, 1980). The nature of this auditory code, especially its characterization as a veridical "echoic" memory, has recently been questioned. It has been shown, for example, that effects that have been attributed to an auditory store (e.g., recency effects), can be produced when modality of input is not explicitly auditory, as when stimuli are lipread (see Coltheart, 1984, and Campbell, this volume, chapter 11, for reviews). Since no actual auditory information is supplied in mouthed stimuli that must be lipread, the form of this code cannot be a simple acoustic "echo." Moreover, since written stimuli do not elicit recency effects to the same degree as do mouthed stimuli, the code responsible for recency effects when stimuli are lipread is not the phonological code that is presumed to support retention of written word lists. It has been argued on the basis of data on recall of lipread stimuli that the code underlying retention is an abstract sound-based (but nonechoic) representation (Campbell & Dodd, 1984); alternatively, it may be that more than one kind of "auditory" code is generated to account for this complex set of results (Gathercole, 1987). The levels of processing involved in translating the acoustic information of the speech signal into the phonemically coded units that are presumably stored in the phonological short-term store have long been of concern to researchers interested in speech perception (Pisoni & Luce, 1987). Of further interest has been the type of information that is generated along the continuum from acoustic to phonetic representations that is needed to gain access to word forms in the lexicon. At one extreme, Klatt (1979) has formulated a model in which access to lexical entries can be gained on the basis of context-sensitive spectral characteristics, without the computation of a distinct level corresponding to discrete phonetic segments. At the other extreme, the interactive activation model of McClelland and Elman (1986) explicitly assumes a segmental representation for speech, with the phonetic feature constituting the most "basic" level of representation available to the word recognition system. The questions posed by this review relevant to patients with putative deficit to the phonological short-term store, which are not answered by findings in the normal literature, are the following: (a) Can a deficit of the phonological STS be distinguished from a deficit at the level of registration of phonological information? (b) If a deficit of the phonological STS is implicated as a cause of a patient's symptoms, what is the basis of the repetition responses that the patient is able to produce? In other words, what information sources serve as a basis for the patient's responses? and (c) What is the effect of a deficit of the phonological STS on auditory word recognition? This chapter attempts to address these questions by considering the performance of a patient with severely reduced repetition span in a variety of tasks investigating her list repetition, phonological processing, and lexical decision performance. It is argued first that the primary deficit involves a reduction of the phonological short-term store, but

118

Berndt and Mitchum

that the ability to discriminate pairs of phonemically related items is intact. Next, an attempt is made to attribute residual repetition abilities to reliance on information from an auditory trace and from the lexicon. Finally, the patient's unusual performance in auditory lexical decision is discussed in light of these hypothesized memory characteristics.

5.2. Case report The topic of this report is EDE, a 56-year-old housewife who suffered a right cerebrovascular accident in August 1982. Primary symptoms were disordered speech, agitation with pronounced mood swings, and left arm drift. Acute CT scan showed a right temporoparietal infarction; angiography uncovered tight stenosis of the proximal right carotid artery. Endarterectomy was performed in September 1982. The patient reportedly improved gradually for several months, although aphasia and labile effect persisted. She was admitted to the University of Maryland Hospital in June 1983 for treatment of Jacksonian seizures that began in the left hand and progressed to become generalized. EDE is a native speaker of American English who completed high school and formerly worked as a secretary. Both she and her husband report that she has always been right-handed, with no history of left-handedness in the immediate family. She scored as a dextral on the Harris Test of Lateral Dominance. EDE was seen intermittently over the next four years. During that time, repetition span and sentence comprehension performance improved very slightly, as noted in section 5.3.2. The patient complains of problems "not getting" what people say, which she says are exacerbated by noise in the background and by people speaking too quickly. During this same period, reading and writing have improved more dramatically; both are currently functional and frequently used communication modes.

5.3. Clinical assessment Clinical assessment of cognitive functions was carried out at the beginning of this study, approximately 1 year following the right cerebrovascular accident, and 3 months following the onset of seizure activity.

5.3.1. Language The Boston Diagnostic Aphasia Examination (Goodglass & Kaplan, 1972) failed to classify EDE into one of the classical aphasia types. Although she generally matched the profile of a conduction aphasic in producing fluent and (infrequently) paraphasic speech, with poor repetition of sentences, her auditory comprehension was too depressed to


119

allow this classification (z-scores for body part identification, commands, and complex material all < — 1). Confrontation naming (Boston Naming Test = 43/60) was moderately impaired, with about half of all errors attributable to phonemic paraphasias. These errors were much more prevalent in naming tasks than in spontaneous speech. Single word comprehension was good (Peabody Form M raw score = 1 1 5 , ceiling = 140), but sentence comprehension was very poor on the Token Test (DeRenzi & Vignolo, 1962); (Part 1 = 6/10; Part 11 = 2/10; Part 111 = 0/10). Oral reading of single words was excellent for highly imageable nouns (27/28) and function words (43/43), but somewhat worse for low-imageable nouns (17/28) (Coltheart, unpublished materials). Reading of single-syllable nonwords was good (27/30), with very occasional lexicalization errors. Comprehension of single printed concrete words was excellent. Writing of words to dictation continues to show a slight effect of imageability, and EDE complains of difficulty writing function words spontaneously, although she has few problems with them in writing to dictation.

5.3.2. Memory Remote memory

Remote memory appears to be intact. EDE can relate in considerable detail the specifics of her illness and of her daily routine; she has no difficulty discussing her son's childhood (some 20 years earlier) in terms that her husband corroborates. She recalls easily the plots of fairy tales and TV shows. On a shortened version of Butters's Famous Faces Test (which excluded sports figures), she demonstrated recognition of 23/40 pictures, although she could name only 13. Immediate memory

Digit span was consistent at two in earliest testings, with occasional correct repetitions of three digits forward. Digit span was retested periodically over the next several years; in February 1985 digit span was consistent at three whether responses were spoken or required pointing to a digit list. Currently, EDE is rarely successful repeating four digits in order. Repetition of monosyllabic concrete nouns shows an input-modality dissociation. When 20 four-item lists were presented aurally, none was repeated in order (as instructed) and an average of two words per list was reproduced with a marked recency effect. When presented visually, four-item lists could be repeated without error. Of 20 visually presented five-item lists, 6 were repeated in order, with an average of four of five words reproduced from each list. Repetition of nonword lists was very poor (2/12 two-word lists).

120

Berndt and Mitchum

Verbal learning

Paired-associate learning (Wechsler Memory Scale) was poor (4/6 easy; 1/4 difficult pairs after three trials). Word list learning was tested with high-frequency monosyllabic noun lists of various lengths. EDE was able to learn a 5-item list (in order) after four trials, and a (different) 10-item list (in order) after seven trials. Thus, verbal learning was impaired but clearly not abolished. Visual memory EDE's ability to reproduce simple designs from memory (Wechsler Memory Scale) was excellent. Her reproduction of the complex figure of Rey (after 1 minute of study) was scored at the 50th percentile (score = 22; see Lezak, 1983, p. 400).

5.4. Immediate memory and phonological processing Experiments were carried out with EDE that were designed to (a) locate her span deficit within the working-memory system, and (b) evaluate the status of the phonological information she was able to enter into the phonological STS.

5.4.1. Experiment 1. Phonemic similarity effect: repetition of phonemically similar and dissimilar consonant lists Procedure

Lists of consonant letters two, three, and four items in length were constructed from phonemically similar (B, D, V, C, G, T) and dissimilar (F, Q, Z, X, R, K) consonants. Ten lists at each length were presented, blocked as to similarity condition. In the visual presentation condition, printed block letters were presented on index cards, at the rate of approximately one per second. In the auditory presentation condition, the letters were read to EDE at the same rate, without allowing her to view the tester's lips. She was requested to repeat the letters in the order in which she had heard or seen them.

Results

As shown in Table 5.1, EDE's performance showed a reverse of the normal modality effect: Repetition of visually presented stimuli was superior to repetition of aurally presented stimuli, regardless of whether scoring was based on ordered strings (z = 4.68, p < .001) or total number of items recalled (z = 2.5, p < .01) (normal test). In addition,


121

Table 5.1. Repetition of phonemically similar and dissimilar consonant lists (number correct in order and proportion correct, without respect to order, in parentheses) Similar Auditory presentation 2 letters 3 letters 4 letters

Dissimilar

2(.35) 2(.63) 0(.23) 4(.39)

5(.70) 2(.67) 0(.48) 7(.59)

2 letters 3 letters 4 letters

10(1.0) 3(.8O)

10(1.0) 8(.9O) 0(.75)

Total

I8(.86)

18(.86)

Total

Visual presentation

Note: N = 10 in each condition.

the phonemic similarity effect was apparent only when stimuli were presented aurally. Significantly more dissimilar than similar consonants were repeated over all list lengths when presentation was auditory (z = 2.5, p < .01). In contrast, there was no difference between number of similar and dissimilar items recalled when presentation was visual.

5.4.2. Experiment 2. Word length effect: repetition of aurally presented one-syllable and three-syllable words Procedure Lists of 10 one-syllable and three-syllable nouns were selected from frequency norms (Kucera & Francis, 1967) such that each one-syllable word was frequency-matched to one three-syllable word. Frequencies ranged from 1/million to 403/million. For each syllable length, 10 lists of two-, three-, and four-word sets were constructed such that words were repeated approximately equally in the various serial positions within the sets, and no word was repeated in a single list. Words were presented aurally to EDE, who was asked to repeat the words in the order given, at the rate of approximately one per second.

Results Performance was scored in terms of number of lists repeated in order and total number of words recalled without regard to order. Phonemic paraphasias, which were infrequent, were scored as errors. As shown in Table 5.2, there was no consistent effect

122

Berndt and Mitchum Table 5.2. Repetition of one- and three-syllable noun lists {number correct in order and proportion correct, without respect to order, in parentheses) 1-syllable words

3-syllable words

2 words 3 words 4 words

3(.65) 4(.77) 0(.60)

7(.9O) 2(.57) 0(.63)

Total

7(.67)

9(.67)

Note: N = 10 in each condition.

Table 5.3. Number repeated without regard to order from each serial position: four-item, one- and three-syllable word lists Serial position

1-syllable words 3-syllable words Mean frequency, items recalled

6 8

6 7

4 4

8 6

93

94

144

78

Note: Maximum possible = 10 in each serial position.

of word length (as measured in number of syllables) on EDE's ability to repeat the lists. The largest difference (favoring longer words) occurred at the shortest list length, but was not statistically reliable (z = 1.34, p > .15). Inspection of errors by serial position of the target words (Table 5.3) indicates that recall of the third item was worse than recall of the other items, suggesting a modified serial position curve over this abbreviated list length. It should be noted that although the last item was recalled as often as the first, it was much less likely to be recalled in correct position. Although word length did not have a significant effect on recall, EDE's errors on this task indicate some differences between the short and long words. Three of seven list repetition errors in the two-word monosyllable condition involved substitution of phonologically similar words (bugs -» rugs) that were not part of the list. Three other errors with one-syllable words occurred as EDE said "something... something... I didn't get that word." In contrast errors with three-syllable words were order errors or intrusions of other (nonphonologically related) words from within the memory set. This pattern suggests better identification, if not retention, of the longer words.


123

5.4.3. Experiment 3. Phoneme discrimination and identification Failure to find a consistent (across modalities of presentation) and strong effect on repetition of phonemic similarity among list items suggests a deficit to the phonological short-term store (Vallar & Baddeley, 1984). Before such an argument can be advanced, however, it is necessary to demonstrate that the patient can successfully discriminate between minimal pairs of phonemes. If she cannot, the deficit cannot be attributed entirely to a storage problem, but must be viewed as involving inadequate entry of phonemically coded information into storage.

Procedure

A 30-item phoneme discrimination task was constructed using pairs of CV syllables. For 15 pairs, both members of the pair were the same syllable. The other 15 pairs differed minimally from each other in voice onset time (5 pairs), place of articulation (5 pairs), and vowel quality (5 pairs). Pairs were randomized and spoken by the experimenter to EDE, who was prevented from using lip cues. The patient was instructed to indicate whether or not the two members of the pair were identical sounds. In an identification test (presented later in the same session), EDE heard 30 spoken syllables and was requested to point to the one of two written choices that corresponded to each syllable. Distractors were written versions of the minimal contrasts used in the nonmatching trials for the discrimination tasks; vowel sounds were represented in real written words (e.g., beat, bit).

Results

EDE made two errors on the discrimination task (both on nonmatching voice onset time contrasts). She made only one error on the identification task (a place contrast).

5.4.4. Experiment 4. Benton Phoneme Discrimination Task This standard test (Benton, Hamsher, Varney, & Spreen, 1983) was chosen to supplement Experiment 3 because it is a difficult task, presented in standardized taped format. Thirty pairs of nonsense words (half matching) are presented by a male speaker. Twenty-two of the pairs are two-syllable items, and the contrast, if it is present, can occur anywhere in the sound sequence (e.g., /kwefab/ ,/kwefad/). Six of the nonmatching pairs contrast consonants; nine contrast vowels. The tape was presented in accordance with the standard procedure.

124

Berndt and Mitchum

Results

EDE made three errors (27/30), all failures to detect contrasts at the ends of items (two vowels, one consonant). This level of performance was well within the normal range (cutoff score for normal = 22), and thus indicates intact phoneme perception and discrimination even for very subtle differences between nonsense words.

5.4.5. Experiment 5. The Denver Auditory Phoneme Sequencing Test This test (Aten, 1979) was given to explore further the ability to distinguish minimal sounds in real words. Fifty words are read by the tester, and each is paired with an array of four pictures. One of the four is a correct depiction of the word, and the other three depict objects with names phonemically similar to the target. For half the items, the distractors are similar to the initial segment of the target (rose —> rope); for half they are similar to the end segments (pan —>fan). The patient was asked to point to the picture named; she was not allowed to request repetitions.

Results

EDE produced three errors on this task. Although norms for adults are not available, this level of performance is assumed to be within the normal range for her age.

5.4.6. Experiment 6. Matching span with phonemically similar and dissimilar consonants The results of Experiments 3-5 indicate that EDE's ability to discriminate among phonemes is good when the storage requirements of the task are minimal. The following task was designed to involve the same kinds of input processing as the phoneme discrimination tasks, but to increase the memory requirements necessary for good performance.

Procedure

The same sets of phonemically similar and dissimilar consonant names used in Experiment 1 provided a basis for this three-item matching span task. A sequence of three spoken consonants (all either phonemically similar or dissimilar) was followed after a pause by another set of three consonants, and EDE was asked to say whether or not the two sequences were identical. Identical sequences were given in 32 pairs, 16


125

Table 5.4. Number correct, three-item matching span task

Phonemically similar

"Same"

"Different"

13/16

0/8 5/8

Phonemically dissimilar

13/16

4/8 8>/8>

order distractors substitution distractors order distractors substitution distractors

composed of phonemically similar consonants and 16 sequences composed of dissimilar consonants. Half of 32 nonmatching sequence pairs contained changes in the order of the consonants in the sequence; the other half contained a substitution from within the appropriate similar or dissimilar set for one of the three items in the string. Strings were presented in random order; lip placement cues were prevented.

Results

As shown in Table 5.4, EDE showed a tendency to respond that the pairs matched, making few errors on matching trials. This level of response bias in a forced choice task indicates poor performance. However, the pattern of errors on the nonmatching trials suggests more than simple response bias. Changes of the order of the sequences were usually undetected, and were always missed when items were phonemically similar. Substitution of one item in the sequence for another was detected consistently only when they were phonemically dissimilar.

Discussion

When stimuli were presented visually, there was no effect on repetition performance of the phonological similarity among stimulus items. This finding has been interpreted as a reflection of a deficit to the phonological STS (Vallar & Baddeley, 1984). When presentation was auditory, however, the phonological similarity of to-be-repeated items did undermine performance. This finding can be reconciled with the hypothesis that EDE has a deficit to the phonological STS if it is assumed that her repetition is based on residual auditory information, as well as perhaps on fleeting phonological information. An auditory trace would be expected to be as susceptible to similarity

126

Berndt and Mitchum

effects as would a phonological code. In contrast, the code generated with written stimuli is apparently not susceptible to sound-based interference. The failure to find an effect of word length on repetition might indicate that EDE is not relying on subvocal rehearsal to maintain information in memory. As we discussed earlier, rehearsal failure might arise because of a pathologically induced inability to carry out subvocal rehearsal, or, as argued by Vallar and Baddeley (1984), because rehearsal may be of no use when the phonological STS is not functioning properly. Since EDE demonstrated no problem with the articulation of speech, and was able to recite the alphabet and to count at normal rates, there is no independent evidence that an articulation-based rehearsal system might be compromised. Thus, the results of Experiments 1 and 2 are interpreted as indications that EDE's poor performance in immediate serial recall tasks results from a deficit to the phonological short-term store. This impairment is not apparently secondary to imprecise registration of phonological information. When the storage requirements are minimal, EDE demonstrates a high level of ability to discriminate among similar phonemes, in words and nonwords. When required to maintain phonemes beyond a pairwise comparison, however, her performance breaks down in a manner suggesting that order is particularly at risk, and that items that sound alike are confused. Other aspects of the results of these experiments suggest a possible basis of EDE's repetition performance in light of a deficit to the phonological STS. In Experiment 2, although the expected advantage of short over long words was not obtained, there was a small trend in the reverse direction at the shortest list length. This may indicate that the additional phonological information in the longer words (relative to the monosyllables) provided a richer basis for lexical identification. This interpretation is supported by the nature of EDE's errors, which suggested that more of the short than the long words were not identified. Second, EDE's repetition in this task shows a trend toward a recency effect for the last item, as well as some primacy effect. This finding contrasts with the typical absence of a recency effect in serial recall tasks in patients of this type (see Shallice & Vallar, this volume, chapter 1). Although failure to find a recency effect is often attributed to a deficit of the phonological short-term store, a positive recency effect is not inconsistent with the same deficit if it is assumed that the last item is repeated on the basis of an auditory trace of residual acoustic information. Such an explanation necessarily implies a distinction between auditory and phonological codes, with the sparing of the former and the involvement of the latter. The primacy effect might be attributed to support from lexical information, which is argued to be more effective early in the list (Saffran, in press).The following set of experiments was designed to investigate the possibility that EDE is performing the list repetition task on the basis of information from auditory and lexical codes, with minimal support from information in the phonological short-term store.


127

5.5. Establishing the basis for list repetition: auditory and lexical effects In this section, experiments are described (a) that investigate EDE's auditory, nonphonological memory; (b) that manipulate the conditions under which the digit repetition task is performed to vary the extent to which auditory information could serve as a basis of performance; and (c) that manipulate the lexical composition of lists to determine the role of lexical factors in performance.

5.5.1. Experiment 7. Nonverbal sound recognition Procedure and results

A tape recording was made of 14 common nonverbal environmental sounds. Each was paired with a picture of the object that produces that sound, along with a distractor picture of an object that produces a similar sound. For example, the sound of an electric shaver was presented with pictures of an electric shaver and a vacuum cleaner. EDE was asked to point to one of the two pictures that were presented with each sound. She performed perfectly, without hesitation.

5.5.2. Experiment 8. Probe recognition of nonverbal sounds Procedure

Fifteen discrete sounds of approximately 1 sec duration were generated on a Yamaha DX-7 synthesizer by varying the parameters of harmonic composition, duration, amplitude, and pitch, with the aim of creating a palette of sounds that could be easily distinguished from one another, but which were not recognizable as produced by specific musical instruments. Thirty sets of four sounds were generated by pseudorandomly combining tokens sequentially so that none was repeated within a set and each was used eight times. Interstimulus interval was approximately 0.8 sec. Fifteen sets were followed after a 2-sec interval by a probe sound that had not been part of the set, and fifteen were followed by a sound contained in the set. Serial positions 1 and 4 were each probed five times, and Positions 2 and 3 a total of five times. Stimulus sets were organized pseudorandomly into one of three blocks, with half matching and half nonmatching trials in each block. Four practice trials were similarly constituted, and a set of all sounds presented individually preceded Block 1. The task was to judge whether or not the probe sound occurred as part of the set. Ten normal adults served as controls. Since it was necessary to compare EDE's performance to the best possible normal performance, and since the task was subjectively quite difficult, control subjects

128

Berndt and Mitchum

were younger (mean = 31) and better educated (mean = 17 years) than the patient. Controls were tested in one session; EDE was tested in three separate sessions on different days.

Results

Mean correct performance for controls was 24/30, range 21-27. EDE was correct on 21 trials, thus scoring within the range of the normal subjects.

5.5.3. Experiment 9. Manipulations of auditory information in digit repetition Results of Experiment 8 suggest that EDE's nonverbal auditory memory is normal. Thus, it is reasonable to investigate the possibility that repetition might be based, at least in part, on a pre-phonological auditory code, if one exists for verbal sounds. The digit repetition task was used so that lexical content of stimuli could be kept as neutral as possible, while auditory input was manipulated. Although at the time of testing EDE's digit span was three, five-digit lists were used in these tasks to assure that span was consistently exceeded. Ten lists of five digits were presented in the following four conditions: 1. Look and listen (EDE was allowed to look at the the experimenter's lips while listening to the stimuli); 2. Look only (stimuli were mouthed silently); 3. Listen only (EDE was not allowed to use lip cues. Note that this presentation condition is the one used for all other repetition tasks reported in this chapter); 4. "Suffixed" digit list (each list of digits was followed by the word zero, which EDE was asked not to repeat). The presence of an auditory suffix - a word at the end of the stimulus list that the subject is requested to ignore - has been shown to attenuate significantly the recency effect (Crowder & Morton, 1969). The last condition was also carried out with fourdigit lists, since it was possible that the extra word in this condition would render the five-digit lists impossible for EDE even to attempt.

Results

Since the unordered recall of digit lists cannot meaningfully be interpreted in view of the high probability of correct guessing, the results of these tasks were scored strictly with regard to ordered production. However, in all serial recall tasks, but especially in digit repetition, EDE used the verbal marker something to signify that she knew there


129

Table 5.5. Number of digits repeated in correct serial position Serial position Condition

1

2

3

4

5

Total

Look and listen Look only Listen only Suffixed (5 digits) Suffixed (4 digits)

10 7 8 9 8

10 8 6 9 6

7 8 1 7 4

2 6 3 2 2

5 6 7 3

34/50 35/50 25/50 30/50 20/40

N/A

Note: N = 10 in each serial position.

was a digit in a certain position that she could not recall. These were interpreted as bona fide placeholders, and subsequent digits were scored as if something had been indeed present in the earlier position. For example, if the digit list 1-2-3-4 were repeated " 1 something-3-2," the first and third digits would be scored as correct; no credit would be given for production of a correct digit 2 in the fourth position. As shown in Table 5.5, EDE repeated the most items in conditions in which she could view the speaker's lips, and there was a significant difference between the total number of items repeated in the "look only" and the "listen only" conditions (z = 4.2, p < 0.001). There was no advantage over the "look only" condition from also being allowed to listen to the stimuli. In addition, the recency effect was greatest when EDE was forced to rely on the heard stimuli alone, and was not present when she relied on visual input alone. There was an attenuated but still evident recency effect in the mixed condition, perhaps indicating a strategic shift toward reliance on the preferred, visual information. The presentation of a suffix removed the recency effect in the four-digit lists, and greatly diminished it in the five-digit lists, despite the fact that total recall was slightly better in the suffixed than in the "listen only" condition (z = 1.21, n.s.).

5.5.4. Experiment 10. Repetition of non words The results of Experiment 9 indicate that the recency effect in digit list repetition is attributable to information specific to actual spoken input, which is easily overwritten by a successive word. The next experiment investigated the question of whether or not characteristics of the spoken output in repetition could be found to reflect the same information source. Intrusion errors in normal list repetition tend to retain nonsegmental features (stress, number of syllables, and vowel quality) when presentation is auditory. These types of intrusion errors are considerably less frequent with visual presentation (Drewnowski & Murdock, 1980). Repetition of non words, which must be accomplished without the advantage of lexical support, was used.

130

Berndt and Mitchum

Procedure

Twelve multisyllabic nonwords (e.g., /Agrosandli/) were constructed, ranging from two to four syllables and 6-11 phonemes in length. Each stimulus was spoken twice before repetition began.

Results

EDE repeated only 3/12 of these complex non words correctly. However, errorful repetitions retained the nonsegmental auditory features of the target to a substantial degree. Only one of the nine errors failed to maintain the stress pattern, and only two contained a different number of syllables. The stressed vowel of the sequence was always reproduced. Errors were overwhelmingly substitutions of incorrect consonants.

5.5.5. Experiment 11. Repetition of word lists It appears that EDE's repetition of the last item in lists is based at least partly on residual information in the auditory signal, and that her nonword repetition errors tend to retain best the nonsegmental auditory information in the target. The primacy portions of lists, typically well retained, cannot be supported by the same source, since subsequent incoming stimuli would obliterate the first item with new auditory information. The first item may have the advantage in terms of lexical or semantic factors, since it is salient in the list and has no competition (when it is heard) for access to the lexicon. In addition, later-presented list items may suffer some decrement in lexical activation arising from allocation of attentional resources to the maintenance of earlier items. Since high-frequency words are assumed to gain fastest access to lexical entries, highfrequency words might be expected to be particularly well retained in primacy position. Further, since the recency position has been argued to rely largely on auditory information, it might be expected to be less influenced by frequency. In fact, post hoc inspection of frequency effects in Experiment 2 (see Table 5.3) suggested that that might be the case. In addition, concrete words have been shown to have an advantage over abstract words in lexical decision (Kroll & Mervis, 1986), although the precise locus of this effect is unclear.

Materials and procedure

Words of high and low frequency, and high and low imageability, were drawn in general from two sources: Kroll and Mervis (1986) and Saffran and Martin (this volume, chapter 6). High-frequency words have frequency > 40/million, and low-frequency


131

Table 5.6. Repetition of words of high and low frequency and imageability {number correct in order, and without regard to order, in parentheses) Serial position Frequency

Imageability

1

2

3

4

Total

High Low High

high high low

5(5) 5(5) 5(6)

1(6) 4(7) 0(0)

0(6) 1(4) 0(4)

K9) 1(5) 1(10)

7(26) 11(21) 6(20)

Slowed presentation Low

high

6(6)

5(5)

2(3)

3(7)

16(21)


words < 10/million (Kucera & Francis, 1967). Word sets were chosen so that no word was ever presented more than once; word frequencies and word lengths (in syllables) were equivalent for each serial position in each set. The following conditions were constructed so that each contained 10 four-word lists: 1. High-frequency, high-imageability words; 2. low-frequency, high-imageability words; 3. high-frequency, low-imageability words.

An additional (different) set of low-frequency, high-imageability words was presented in slowed presentation rate, at approximately one item per 2 sec.

Results

Although EDE was instructed to repeat in order, she rarely repeated back more than the first item in correct serial position. Again items were scored as occurring in correct order if she said the word something to mark the position of an omitted word. (This was done less frequently than in digit repetition.) Serial position data for the four conditions are shown in Table 5.6. Although high-frequency words enjoyed a very small advantage over lowfrequency ones, the hypothesis that this effect would occur early in the list was not upheld. A comparison of recall from early (Positions 1 and 2) and late (Positions 3 and 4) in the list for the high- and low-frequency, high-imageability conditions indicates, if anything, a nonsignificant trend toward a high-frequency advantage in the recency position (%2(1) = 1.02, p > 0.30). There is similarly no effect of imageability when the first two positions are compared to the last two Qf2(l) = .735, p > 0.30). It is clear, however, that low-imageability words are considerably less likely than highimageability words to be retained in Position 2. If low-imageability words are more

132

Berndt and Mitchum

difficult to gain access to in the lexicon, or to retain for some other reason, reliance on auditory information (and consequently good recency performance) would be expected and was obtained (all 10 final-position items recalled in the low-imageability condition). This argument is supported by an analysis of the quality of EDE's responses: Minimal phonological distortions occurred only on items presented in final position. Of 10 correct responses in the recency position of the low-imageability set, 5 contained such distortions (e.g., anxiety —> /lns9Ati/), which did not occur in other positions. These distortions suggest that EDE's production of final-position words was reliant on the sounds of the stimuli in the low-imageability condition rather than on their lexical identity. Slowing presentation rate, which might be expected to maximize lexical access and thus result in better recall from early positions, resulted in minimal improvement. Four semantic errors occurred overall, equally divided between Position 1 and 3; all of these were found in high-imageability lists.

5.5.6. Experiment 12. Free recall, mixed lists of high and low imageability All of the repetition results reported thus far have required serial recall. Although EDE was not able to repeat four words in order, she struggled to do so. Thus, the primacy effect may result not from better lexical-semantic support in that position, but from a straightforward attempt to follow instructions. This experiment was carried out to determine if the advantage of imageability, occurring early, would result in serial position effects even without the constraint of repeating in order.

Procedure

Ten four-word lists were constructed of high-frequency words such that the first and second positions were composed of high-imageability words, and positions 3 and 4 contained low-imageability words. Mean frequencies and syllable lengths were equivalent for each serial position. EDE was instructed to repeat as many words as possible, in any order.

Results

As shown in Table 5.7, the high-imageability words were repeated about twice as well as the low-imageability words (z = 2.48, p < 0.01), but there was still a recency effect for final position. Inspection of the order in which words were repeated indicates a continuing tendency to start with the first item (5/10 trials), which represents a deviation from the normal tendency to recall first from the end of the list in free recall (Postman & Phillips, 1965). This may simply reflect a long-established habit, although EDE was noticeably relieved not to be constrained to repeat in order.


133

Table 5.7. Free recall of high- and low-imageability words (high frequency) {number repeated correctly) Serial position

Order of report First Second Third Fourth Total

1 2 (high imageability)

3

5 2 2 0 9

2 5 0 0

1 0 1 0

7

2

4

(low imageability)

2 3 1 0 6


5.5.7. Discussion EDE appears to process and to maintain normally nonverbal auditory stimuli. This finding is consistent with other results from patients with short-term memory deficit (e.g., Shallice & Warrington, 1974). Moreover, she retains auditory features of nonword stimuli while repeating, even though she fails to retain segmental phonetic information. These results suggest that an auditory, pre-phonological memory code exists, and that it is spared in a patient who shows a severe reduction of phonological memory. An "auditory" memory component has been argued to be responsible for the privileged status of list-final items when normal subjects are asked to repeat aurally presented word lists. A recency effect was found for EDE in all repetition tasks involving auditory presentation. When stimuli were mouthed, and EDE's performance was based entirely on lipreading, the recency effect was not evident, though performance was relatively good. Normal subjects typically show a recency effect with lipread stimuli (Campbell & Dodd, 1984), an effect that has been attributed to the generation of an auditory-like code to mouthed stimuli. In contrast, EDE seems to require actual auditory sensory input to generate the memory code responsible for recency. Although EDE performs better when lip-reading alone than when listening alone, this advantage is most evident in list positions most at risk with auditory presentation: Positions 2 and 3. When lip cues are available, EDE apparently relies on them at the expense of auditory cues, since there was no improvement relative to the mouthed condition when both types of information were available. An auditory suffix word at the end of the list diminishes the recency effect, a result that when found with normal subjects has been used to support the auditory basis of the recency effect. Thus, it appears that there is evidence that EDE's repetition is supported at least in part by information held in a pre-phonological auditory code, which is spared despite a severe reduction of phonological memory. Other patients with deficits to

134

Berndt and Mitchum

phonological STS, who do not show recency effects, presumably do not have this selective sparing. List positions earlier than final position are not likely to be recalled from an auditory memory. Lexical information is arguably a source of support for earlier list positions when phonological memory is compromised. We did not find strong support for this hypothesis. Reasoning that high-frequency words would gain access to the lexicon more easily than low-frequency words, we predicted that high-frequency lists would enjoy particularly good recall in early list positions. This result was not obtained; if anything, high-frequency words were better recalled than low-frequency words only in list-final position. Two possible explanations for this result can be entertained. First, EDE's ability to gain access to information in the lexicon may be abnormal such that frequency advantages for high-frequency words found with normal subjects are attenuated. Further, better retention of higher-frequency words in list-final position (a position that has been argued to enjoy benefit from auditory memory) suggests that lexical access may have been enhanced by continued maintenance of the auditory signal. That is, an interaction between information sources may have occurred to favor retention of words with easiest accessibility (high-frequency words) in the list position with the strongest auditory trace. A second possibility that must be considered is that purely lexical factors (such as frequency) do not provide a useful basis for word retention for this patient. A more obviously semantic than lexical factor - imageability - had a discernible effect on performance in Position 2, though no reliable effect on the early list positions in general. On the assumption that words of low imageability are less easily accessed (Kroll & Mervis, 1986), they should be harder to retain in early list positions. In this case, any available auditory information should be exploited to the fullest extent possible, since only weak competing sources of memory support are available from the lexical-semantic system. There was a trend in this general direction; items were most likely to be lost with low-imageability relative to high-imageability words early in the list (Position 2); all final-position, low-imageability words were retained, with the quality of responses indicating that recall from that position was more sound based than meaning based. Further, results of the free recall task suggest that the advantage of high- over low-imageability words is real, as EDE did not choose to exploit the auditory advantage of list-final items, but overwhelmingly recalled from highimageability early positions. Several aspects of EDE's free recall performance are worthy of note. The patient's failure to recall the last items first is similar to the performance of PV, another patient with short-term deficit who demonstrated no recency effect in free recall (Vallar & Papagno, 1986). When PV was requested to begin recall with list-final items, she could do so, although recall from recency positions remained defective with auditory presentation. Vallar and Papagno argue that this finding supports the view that the


135

recency effect in free recall reflects the output of the phonological short-term store. When the phonological STS is defective, recency performance should be affected. Despite the presence of a recency effect in EDE's list repetition, there are indications that recency performance is not normal. In addition to her failure to repeat last items first in free recall, the recency effect is limited to only one item. Thus, EDE's data do not contradict Vallar and Papagno's conclusion that recency reflects the output of a phonological STS. In EDE's case, the disordered STS may be the source of the abnormal recency performance, while as argued here, an intact auditory, nonphonological code provides the basis for her minimal production of a recency effect. The results presented in section 5.2 of this chapter indicate that EDE's list repetition relies on auditory and, to some extent, on semantic information, and they also suggest possible problems of lexical access in that expected frequency effects did not occur. The next set of studies was designed to explore EDE's single-word processing abilities in comprehension and lexical decision tasks. 5.6. Lexical and semantic processing of single words Three types of tasks were carried out to investigate EDE's ability to gain access to words in the lexicon and to understand single words. 5.6.1. Experiment 13. Word-picture matching with related distractors

Procedure and results Sixty concrete nouns from two frequency ranges ( < 2 5 or > 25/million) were presented aurally to EDE along with a pair of pictures. Distractors included 20 pictures in which the name was phonemically related to the target, 20 pictures that were semantically related to the target, and 20 pictures that were unrelated to the target. Distractor types were presented randomly, and repetition of the target name was not allowed. EDE was asked to indicate which of the two pictures showed the target items. Pictures were in view when the name was presented. EDE performed this task with no hesitation and made no errors. 5.6.2. Experiment 14. Synonym judgments

Procedure A subset of the high- and low-imageability synonym judgments prepared by Coltheart (unpublished) was administered. These items have been selected such that each highimageability pair (e.g., grave/tomb) is matched in frequency, length, and degree of

136

Berndt and Mitchum

synonymy to one of the low-imageability pairs (e.g., idea/notion). Twenty pairs of each of the two imageability types (10 synonyms and 10 nonsynonyms) were spoken aloud to EDE, who indicated whether or not the meanings of the two words were essentially identical. She was allowed to have the items repeated once, and she requested eight repetitions.

Results

EDE was correct on 19/20 high-imageability pairs and 17/20 low-imageability pairs. This is interpreted as good performance of this task.

5.6.3. Experiment 15. Lexical decision Limited testing was carried out with regard to EDE's reading skills, since deficits in this area were not among her primary complaints. However, in light of her performance with auditory lexical decision, to be presented, her ability to make lexical decisions to visually presented stimuli provides an important index of her understanding of the task demands.

Procedure, 15a

Materials for visual lexical decision were again borrowed from Coltheart (unpublished). An "easy" lexical decision task required judgments on 20 words (short, common nouns) and 20 nonwords (formed by changing one letter of words); mean frequency was approximately 450/million. A "difficult" lexical decision contained the same number of items, but length was always greater than 11 letters and five syllables. Frequency was constant at 1/million. Nonwords were generated by interchanging two syllables (e.g., cirsemicular from semicircular). EDE was presented with these items, typed on cards, in separate administrations during the same session.

Results, 15a

EDE performed the "easy" task effortlessly and without error. She made seven errors on the "difficult" version (83% correct), equally distributed between misses and false alarms. Since this latter task contains lengthy and low-frequency words, this level of performance is considered to be quite good. Most important, it indicates that EDE understands what is involved in making lexical decisions.


137

Procedure, 15b

A more extensive and controlled lexical decision task was administered to EDE through the auditory modality. A 100-item task was constructed with 50 words of two levels of frequency ( > 50 or < 50/million) and of high and low imageability. Nonwords were constructed by making minimal changes (one or two phonemes) to real words. Items were randomized and presented to EDE in one block. She was permitted one repetition of the stimulus, which she requested only on nonwords.

Results, 15b

EDE responded that 82/100 items were real words; she correctly rejected only 18 (36%) of the nonwords. When a nonword was correctly rejected, it was done with certainty. Likewise, real words of all kinds were accepted without hesitation. Of the many incorrect judgments that nonwords were words, 12 (38% of errors) were accepted with the appearance of certainty, that is, without hesitation. Sixteen errors (50%) were committed quite hesitantly, usually with the accompanying remark "I think it's a word, but I don't know what it means." An additional four errors (12%) were overt guesses. It was felt that EDE might have set a very loose criterion for the acceptance of stimuli as real words, based on her feeling that her comprehension was poor. That is, she did not appear surprised that she thought items were words that she did not know, despite our rinding that her single-word comprehension was quite good. To investigate the possibility that EDE's criterion could be shifted, we readministered the task using an exactly comparable set of stimulus items. Words were matched in frequency, concreteness, and length to the first set; nonwords were formed in the same way. EDE was told that she had made many errors on the first administration and that we thought she should have more confidence in her own ability to detect nonwords. She was also told that if she did not immediately recognize the item, it was probably not a real word and should be rejected. Performance on this administration was better, although her responses on nonwords were hesitant and she requested that 20% of the nonword stimuli (and none of the word stimuli) be repeated. She failed to reject only four (8%) of the nonwords and incorrectly rejected one real word. However, she gave the impression of uncertainty when responding to a large number of nonwords. This kind of behavior suggests that EDE's problems in the lexical decision task may be caused not by difficulty gaining access to lexical items but by the adoption of an inappropriate response criterion, or by an impairment to some process whereby items are validated after they have been accessed. The next experiment was designed to investigate these possibilities.

138

Berndt and Mitchum

Procedure, 15c

An additional set of lexical decision tasks was administered in an attempt to determine whether or not EDE was experiencing true lexical access difficulties. This task used stimuli that were made up of real word stems and inappropriately joined suffixes. Realword stimuli (n = 52) included regularly inflected words, uninflected words, and functors. Nonwords were of two general types : 32 nonwords were formed by joining a real word stem to a legal word ending that was inappropriate for that stem (e.g., tennised). Half of these endings were grammatical inflections (e.g., -ing), half derivational and other suffixes (e.g., -ate). An additional 20 nonwords were formed by changing two letters of words matched in frequency and concreteness to the real words (e.g., spract from strict). Data obained from normal subjects on this task indicated that nonwords made up of real lexical roots and inappropriately attached inflections are very difficult for subjects to reject (error rates > 20%; reaction times significantly slowed; see Salasoo & Berndt, 1986). This effect was not simply an artifact of having real words in the initial segments of the nonwords, because real lexical roots paired with noninflectional word endings (e.g., derivational suffixes) did not cause particular difficulty. Other nonwords included in the task differed from real words in the usual way, that is, by a change of two phonemes. The expectation was that if EDE's poor lexical decision performance was the result of some strategic response to her uncertainty about the task demands, then errors should occur somewhat randomly across the nonword types. If lexical access problems are the cause of her poor performance with nonwords, then the normal pattern of difficulty with inappropriately inflected real word roots should be exacerbated. In addition, a reaction time format was used in an attempt to force EDE away from a reflective strategy toward giving an immediate response. Stimuli were digitized for auditory presentation by computer, so that stimulus duration could be equated across types. Each trial began with a READY prompt in the middle of the CRT screen. When EDE responded with a button press, the screen went blank and the spoken stimulus was presented through headphones after a 75 0-msec pause. She responded by pressing buttons marked WORD or NONWORD on a response box. Response latencies and errors were recorded automatically. EDE was tested in four separate sessions consisting of two or three blocks/session, which were conducted approximately 2 weeks apart. She received no special instructions for this task, but was asked to respond as quickly and accurately as possible. Two female control subjects, matched to EDE in age and education, were tested in the same manner. Results, 15c

As shown in Table 5.8, EDE continued to show a strong bias to respond that nonwords were words. Her response latencies indicate that real words were accepted quickly


139

Table 5.8. Reaction times for correct responses (and error rates) in auditory lexical decision

Words No suffix (N=36) EDE Control 1 Control 2 Mean control

1637(.O8) 1521 (.10) 1285 (.12) 1403 (.11)

Nonwords Properly suffixed (N=16) 1599(.18)

1662(25) 1530(.18) 1596(.22)

No suffix (N=20)

Inappropriate inflection (N=16)

Inappropriate other suffix (N=16)

325 7 (.43) 1446(.27) 1771(.12) 1608 (.19)

3118(.7O) 1662 (.25) 2101 (.23) 1881 (.24)

2536(35) 1412(.1O) 1876(.O8) 1644(.O9)

(comparable to the response times of the controls), whereas correctly rejected nonwords were responded to very slowly. In addition, however, EDE had serious problems rejecting inappropriately inflected word roots. Although her response latency suggests that she had doubts about the lexicality of these items, she accepted the majority (70%) as words. Although performing poorly in general, she was considerably better at rejecting nonwords formed by changing phonemes or by inappropriate joining of noninflectional suffixes to real word roots.

5.6.4. Discussion EDE performs well in tasks requiring interpretation of the meanings of individually presented words, both in picture pointing and in synonym judgments. In contrast, she has difficulty performing auditory lexical decision, demonstrating a strong tendency to accept most nonwords as words. This problem does not result from a failure to understand the task demands, since she performs well on visual lexical decision. Further, she does not seem merely biased to respond "word" secondary to a pathologically shifted criterion of acceptability. Although when instructed she can increase her successful rejection of nonwords, she remains very hesitant when doing so and apparently uses a criterion based on whether or not she can access the item's meaning. Finally, she has most difficulty rejecting nonwords that normals also find difficult to reject, showing a very exaggerated but normally distributed pattern of errors. These results support the paradoxical conclusion that EDE's access to real words within the lexicon is achieved normally. However, when nonwords are similar to words, she is virtually unable to decide that they are not words. One possible explanation for this unusual pattern is that the auditory (and perhaps very short-lived phonological) information that is available to EDE is sufficient to achieve a match with real lexical items. When only a partial match is made - as with wordlike nonwords - it may be necessary to check that partial activation against the stimulus word before rejecting it.

140

Berndt and Mitchum

In EDE's case, this checking process must be carried out against a suboptimal trace of the stimulus item, that is, one without fully specified phonemic information.

5.7. General discussion The general framework within which this report is placed is one in which immediate memory is conceived of as a multicapacity system consisting of transient representations of various codes generated during language-processing tasks. The performance of the immediate serial recall task can be viewed as systematically exploiting information extracted from the incoming signal and represented in auditory, phonological, lexical, and articulatory codes. There is evidence from studies of normal immediate serial recall that all of these sources of information contribute to performance (see Saffran, in press, for review). The model of immediate memory that has served as a basis for most of the recent neuropsychological investigations of memory deficits - the working memory model - has focused on the phonological and articulatory codes. "Higher-level" informational support for immediate repetition tasks (e.g., lexical and semantic information) has been assumed within the working memory model to involve access to permanently stored representations in long-term memory. Yet it is clear that these higher-level informational sources, once temporarily activated in the course of the recall task, contribute to performance in regular and consistent ways (Saffran, in press). For example, lexical frequency has been shown to exert its effects only on recall from early list positions (Watkins & Watkins, 1977). Nonetheless, lexical contributions to recall performance have received little attention from the developers of the working memory model, and only slightly more from investigators studying the effects of brain damage on immediate memory (but see Saffran & Martin, this volume, chapter 6). The role of an auditory, nonphonological code in the immediate recall task has been similarly neglected, perhaps because serious questions have been raised about the actual nature of such a code (e.g., Coltheart, 1984). Nonetheless, recent attempts to reformulate models of immediate memory to be more relevant to language-processing tasks such as comprehension have distinguished between information that merely persists for some period of time and information that has more permanent status. For example, Monsell (1984) distinguishes between two qualitatively different types of temporary storage, and reviews evidence for the distinction. A "Type I" record is the state of persisting activation that exists for some nontrivial time after information is activated. This type of record is very limited in its ability to encode order, but is, of course, sensitive to stimulus recency. A "Type II" register is necessary to represent novel structures, including those relational elements encoded by order. A "Type II" record may involve replicas of attributes copied from permanent memory structures into limited-capacity temporary storage, or it may involve the representation of novel structures by labeling existing units within permanent storage. A Type II record is


141

argued to exploit at least two types of codes, one phonological and the other lexical. For auditory input, this distinction points up the difference between a persisting auditory record of what is heard and a phonological buffer that is exploited in visual as well as auditory tasks. The Type II record may serve as an output as well as an input buffer for temporary storage of phonologically and lexically coded information. It is not clear from this distinction what the processing relation is between these two stores, that is, how information passes from one into the other. Barnard's (1985) more formal proposal along the same lines distinguishes several different types of "image records" that form an episodic memory representation for different kinds of information. An acoustic subsystem (AC), responsible for analyzing sounds and producing coded representations of their structure and content, is distinguished from a higher-level "morphonolexical" (MPL) code. Interfacing these two systems (AC -> MPL) is a set of processes that "mediates speech perception and effectively results in the replacement of an acoustic string with a morphonolexical string which reflects an initial constituent analysis of incoming speech. The MPL string is postcategorical, segmented and entails a loss of speech information" (p. 209). It has been stated throughout that the major deficit uncovered in this investigation of EDE is to the aspect of the system responsible for storage of phonologically coded information. Within the working memory model, this is the phonological short-term store; within Barnard's model it would presumably involve the MPL code. In the face of this deficit, performance must be based on other sources of information. Some of the results reported here can be accounted for by assuming that repetition of word lists is accomplished partially on the basis of information in a Type I persisting activation record. The recency effect, and especially the detrimental effect on recency of a suffix word and of lip-reading the stimuli, are consistent with this explanation. Other findings require additional assumptions about the nature of the processes that are impaired, and especially about the relationship between phonological and lexical codes. EDE's performance with single words is somewhat paradoxical: Performance on comprehension tasks is better than lexical decision, and it is unlikely that this discrepancy results from failure to understand the demands of the lexical decision task. One possible explanation for this finding is that lexical access, as well as semantic analysis and interpretation, is carried out imperfectly on the basis of the persisting activation record and the phonological code that it (fleetingly!) generates. Alternatively, as in Barnard's (1985) proposal that a code combining phonological, morphological, and lexical information is the basis of the Type II record, impaired phonological memory would necessarily involve problems with lexical and morphological information. The data do not distinguish between these two possibilities. Nonetheless, these considerations raise questions about the amount and nature of the auditory information that is required to gain access to lexical information, that is, about the necessity that it be phonologically interpreted. As discussed in the Introduction, this question has been

142

Berndt and Mitchum

widely debated by researchers interested in speech perception, and nothing resembling a consensus has emerged. Our results suggest that the information that EDE has available to her about the sounds of words she hears — and there is evidence that this consists of well-maintained auditory information, as well as poorly maintained phonological information — is not enough to support normal rejection of nonwords in the lexical decision task. In contrast, it appears that some degree of semantic information is made available by the same sources. This rinding is consistent with a recent study with normal subjects showing that semantic priming can be induced by nonwords that are phonetically similar to words semantically related to the target (Milberg, Blumstein, & Dworetzky, 1988). That is, an imperfect phonetic token of a word is apparently adequate to produce some level of semantic activation. The picture that emerges from these findings is one in which failure to maintain phonological information in immediate memory mandates reliance on other information sources for the performance of immediate serial recall tasks. Evidence was accumulated that one source of such information is an auditory code of some description that persists briefly, but is obliterated by further incoming auditory information. Lexical effects were less clearly evident. In fact, an impairment of lexical access mechanisms may represent a further deficit in this patient. Although it is possible that this might be an independent impairment, it is more likely to be related to the severe difficulty the patient has maintaining phonological information in memory. As EDE attempts to gain access to lexical items from auditory input, she may activate only partial phonological and lexical information. In addition, there are suggestions that such partial activation is primarily semantic, rather than strictly lexical, in nature. Clarification of the apparent implications of these findings for models of auditory word recognition and comprehension must await further research.

References Allport, D. A. (1984). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 313-325). Hillsdale, NJ; Erlbaum Aten, J. (1979). The Denver Auditory Phoneme Sequencing Test. Houston: College Hill Press. Baddeley, A. D:(1983). Working memory. Philosophical Transactions of the Royal Society of London, B302, 311-324. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 8, pp. 47-89). New York: Academic Press Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). London: Erlbaum. Benton, A. L., Hamsher, K. D., Varney, N. R., & Spreen, O. (1983). Contributions to neuropsychologic assessment. Oxford: Oxford University Press.


143

Campbell, R., & Dodd, B. (1984). Aspects of hearing by eye. In H. Bouma and D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 299-312). London: Erlbaum. Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-275. Coltheart, M. (1984). Sensory memory. In H. Bouma and D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 259-285). London: Erlbaum. Crowder, R. G., & Morton, J. (1969). Precategorical acoustic storage (PAS). Perception and Psychophysics, 5, 365-373. DeRenzi, E., & Vignolo, L. (1962). The Token Test: A sensitive test to detect receptive disturbances in aphasics. Brain, 85, 665-678. Drewnowski, A., & Murdock, B. B. (1980). The role of auditory features in memory span for words. Journal of Experimental Psychology: Human Learning and Memory, 6, 319-332. Friedrich, F., Glenn, G., & Marin, O. S. M. (1984). Interruption of phonological coding in conduction aphasia. Brain and Language, 22, 266-291. Gathercole, S. E. (1987). Lip-reading: Implications for theories of short-term memory. In B. Dodd & B. Campbell (Eds.), Hearing by eye: The psychology of lip-reading (pp. 227-241). London: Erlbaum. Goodglass, H., & Kaplan, E. (1972). The assessment of aphasia and related disorders. Philadelphia: Lea & Febiger. Klatt, D. H. (1979). Speech perception: A model of acoustic-phonetic analysis and lexical access. Journal of Phonetics, 7, 279-312. Kroll, J. F., & Mervis, J. S. (1986). Lexical access for concrete and abstract words. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 92-107. Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Lezak, M. D. (1983). Neurospychological assessment (2nd ed.). New York: Oxford University Press. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86. Milberg, W., Blumstein, S., & Dworetzky, B. (1988). Phonological factors in lexical access: Evidence from an auditory lexical decision task. Bulletin of the Psychonomic Society, 26, 305-308. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes, (pp. 327-350). Hillsdale, NJ, Erlbaum. Pisoni, D. B., & Luce, P. A. (1987). Acoustic-phonetic representations in word recognition. Cognition, 25, 21-52. Postman, L., & Phillips, L. W. (1965). Short-term temporal changes in free recall. Quarterly Journal of Experimental Psychology, 17, 132-138. Saffran, E. M. (in press). Short-term memory impairment and language processing. In A. Caramazza (Ed.), Advances in cognitive neuropsychology and neurolinguistics. Saffran, E. M., & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with deficient auditory short-term memory. Brain and Language, 2, 420-433. Salasoo, A., & Berndt, R. S. (1986). Morphemic structure and lexical processing: evidence from inflected words and pseudowords. Paper presented at the Midwestern Psychological Association, Chicago, Illinois. Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729-735. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261-273. Shallice, T., & Warrington, E. K. (1974). The dissociation between short-term retention of meaningful sounds and verbal material. Neuropsychologia, 12, 553-555.

144

Berndt and Mitchum

Vallar, G., & Baddeley, A. D. (1984). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Papagno, C. (1986). Phonological short-term store and the nature of the recency effect: Evidence from neuropsychology. Brain and Cognition, 5, 428-442. Watkins, O.C, & Watkins, M. J. (1977). Serial recall and the modality effect: Effects of word frequency. Journal of Experimental Psychology: Human Learning and Memory, 3, 712-7IS.

6. Neuropsychological evidence for lexical involvement in short-term memory ELEANOR M. SAFFRAN AND NADINE MARTIN

6.1. Introduction According to the standard approach to short-term memory (STM), set forth by Baddeley in chapter 2, verbal STM depends on a phonological store of limited capacity. It is not clear, however, how this capacity is to be defined. Thus, if we take span (number of items recalled in correct order) to be our index of STM capacity, it is necessary to deal with the fact that this number does not represent some fixed quantity. Rather, it varies substantially across material type: span for digits (7.98) is greater than span for familiar words (5.86), which is in turn greater than span for nonword materials (2.49) (Brener, 1940). Various manipulations also affect span for words and nonwords differently: Performance on word lists is more resistant to suffix effects (Salter, Springer, & Bolton, 1976), and less affected by such factors as presentation modality and phonemic similarity (Richardson, 1979). Lexicality is evidently a sustaining factor in STM. There has been little attempt to deal with these lexical influences, however, either from a theoretical perspective or as a matter for empirical study. Far from their being a focus of attention in STM research, experimenters have tended to minimize the effects of lexical variables by relying on digit materials or, when using words as stimuli, by sampling from a restricted item pool. It has sometimes been acknowledged that the information in the phonological store is "postexical," the implication being that the information represented in this store consists of phonological units filtered through the lexical system (e.g., Richardson, 1979). But there has been no effort to specify the representational characteristics of this store, an undertaking that would oblige the STM theorist to deal with such matters as the nature of the lexical-phonological units and the mechanisms that bind them into a linear arrray. In neuropsychological work on STM, which is grounded in this same theoretical framework (see Shallice and Vallar's discussion in chapter 1), the role of lexical factors in performance has been similarly neglected. Where lexical influences have been noted in This study was supported by Grant NS 18429 from the National Institutes of Health. We thank Graham Hitch, Randi Martin, and Tim Shallice for helpful comments on an earlier version of the manuscript.

145

146

Saffran and N. Martin

patients' performance, these effects are interpreted in terms of increased reliance on long-term memory (LTM) capacities and hence as not particularly germane to shortterm storage (e.g., Caramazza, Basili, Koller, & Berndt, 1981). In this chapter, we focus on these lexical effects, and attempt to come to grips with the issue of how lexical factors might be assimilated to models of STM function. The need to deal with this issue first confronted us in an earlier study that focused on the repetition performance of a patient (ST) with severely impaired lexical functions (Martin & Saffran, 1987). Clinically described as a "transcortical sensory aphasic," ST's spontaneous production was anomic, neologistic, and generally incoherent. Her output improved markedly in repetition tasks, where effects of word frequency indicated that her performance depended, at least in part, on lexical information. ST's ability to repeat word lists was, however, subject to stringent limitations. Although she repeated high-frequency three-word lists almost perfectly, early list items began to drop out with longer lists. Her performance on longer lists was also marked by a tendency to initiate recall with the last or next-to-last item. With low-frequency fouritem lists, neologisms intruded into her responses; of particular interest was the rinding that phonemes from the terminal item in the input string intruded into neologisms produced at the beginning of the output string. With six-word strings, ST was rarely able to retrieve any but the last two items. This set of observations suggested that ST's repetition performance relied heavily on a phonological trace, containing information only for the most recent items, that interacted with accessible (i.e., high-frequency) lexical units. In view of the fact that ST's production of specific word targets was very poor except when driven by immediate phonological (or orthographic) input, it was reasonable to interpret her failure to retrieve early list items as a further manifestation of her lexical impairment. The phenomena observed in ST prompted us to look more closely at lexical influences in the STM performance of other patients. In this chapter, we examine the performance of two patients with span limitations whose rather different serial recall patterns are interpreted to reflect different degrees of lexical support for the short-term retention of list information. The data are accounted for within a new theoretical framework for short-term information storage in which lexical structures figure prominently.

6.2. Subjects One of the patients, TI, is also the subject of another study (Saffran, 1985; Saffran & Martin, this volume, chapter 16) that focuses on his performance in sentenceprocessing tasks. The other, CN, has been a subject in studies of acquired dyslexia (Coslett, 1986). Characteristics of these patients are summarized in Table 6.1.

Lexical involvement in short-term memory

147

Table 6.1. Subject information

Age Sex Education Occupation Etiology Lesion

Language characteristics BDAE

Boston Naming Sentence comprehension"

CN

TI

56 Female Bachelor's degree Adult education teacher LCVA secondary to aneurysm, 1976 Patchy infarct involving inferior middle and superior portions of L temporal lobe with partial sparing of superior temporal gyrus

72 Male High school graduate Manager L and RCVAs, 1983

Moderately nonfluent, nonagrammatic Below aphasic mean only on Low Probability repetition subtest No cue: .75(45/60)correct Normal range: 46-60 correct

Fluent, occasional literal paraphasias Below aphasic mean only on comprehension of commands and High and Low Probability repetition subtests No cue: .78(47/60)correct Normal range 42-59 correct

Semantically reversible: Active: .92(22/24) correct Passive: .46(11/24) correct

Semantically reversible: Active: .67(16/24) correct Passive: .42(10/24) correct

Quiet: .90 correct (n = 30) 12th percentile

Quiet: .93 correct (n = 30) 50th percentile Noise: .50 correct (n = 30) 2nd percentile

Real words (n = 60): .95 correct Pseudowords (n = 60): .99 correct

Real words (n = 29): 1.00 correct Pseudowords (n = 31): .56 correct

Auditory presentation (words) Rhyme (n = 33): 1.00 correct No rhyme (n = 31): 1.00 correct Visual presentation (words) Rhyme (n = 33): .90 correct No rhyme (n = 31) .94 correct

Auditory presentation (words) Rhyme (n = 33): .91 correct No rhyme {n = 31): .90 correct

L posterior parietal and parietotemporal infarct; R inferior frontal infarct

Phonological processing

Discrimination^

Auditory lexical decision

Rhyme judgments

"Schwartz, Saffran, and Marin (1980). b Goldman-Fristoe-Woodcock Test of Auditory Discrimination.

Visual presentation (words) Rhyme (n = 33): .70 correct No Rhyme (n = 31): .52 correct

148


In most respects, the two patients appear quite similar. On clinical examination, they were primarily impaired in repetition tasks. On further testing, they both demonstrated the sentence comprehension impairment that is characteristic of patients with STM deficits (see part IV of this volume). Both appeared to be only mildly impaired in naming. The major differences between them occurred in phonological processing tasks: CN performed quite well on these tests (though she has been characterized as phonologically impaired in reading [Coslett, 1986]), while TI demonstrated significant impairment.

6.3. STM performance 6.3.1. Digit span Serial position curves for recorded digit strings presented at the rate of one per second are shown in Figure 6.1. Both patients have abnormally short spans, and both show loss of the normal recency effect with longer lists. The loss of recency is generally taken to reflect impairment of an auditory-phonological store (Shallice & Vallar, this volume, chapter 1). CN shows some erosion of the primacy effect with five-word lists. On an equivalent test in the visual modality, in which digit strings were presented item by item on a computer screen, neither patient showed the improvement with visual presentation that is typically found in STM patients (Shallice and Vallar, chapter 1).

6.3.2. Word span Recorded lists, comprising two to four concrete nouns, arranged in blocks of 10 of each length, were presented at a rate of one item per second. Results are shown in Figure 6.2. Although comparable in most respects on digit strings, the two subjects perform quite differently on word lists. TI shows essentially the same decrement across serial positions that he did with digit materials, while CN's serial position curve is marked by a loss of items from the beginning of the list, a pattern that emerges with lists as short as three items. With three- and four-word lists she generally did not initiate recall with the first position target, but rather with the penultimate or terminal list item. This tendency was observed on 40% of the three-word lists and on 90% of the four-word lists. The difference in CN's pattern across serial positions with digit and word lists suggested that lexical factors might figure importantly in her STM performance. The influence of lexical variables was examined systematically in the next set of studies.

Lexical involvement in short-term

memory

149 CN.

T.I.

1.0

1.0

0.9

0.9

0.8

0.8

0.7

0.7

Proportion of 0.6 digits repeated 0.5 accurately in serial order 0.4

0.6

0.3

0.3

0.2

0.2

0.1

0.1

0.5 0.4

0.0 •• 1

2

3

0

4

Proportion of digits repeated accurately in any order

1.0

1.0

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

2

3

4

0.3

0.3

0.2

0.2

0.1

0.1

o.oc

1

Position in input string


O.OC 1

2

3

4

0


••- Two digits (n = 20 strings) •o- Three digits (n * 20 strings) • - Four digits (n = 20 strings) • Q - Five digits (n=10 strings)

Figure 6.1. Performance of CN and TI on digit span tests.

1

2

3

4


5

150

Saffran and N. Martin CN.

T.I.

1.0

Proportion ol words repealed accurately in serial order

1.0

T

0.9

0 9

0.8

0.8

0.7

0.7

0.6

0.6

05

0 5

04

0.4

0.3

0.3

0.2

0.2

0.1

0.1

oo i

0.0 1

2

1

3

Proportion of words repeated accurately in any order

1 .0 •

1.0

0.9 •

09

0.8

08

07

0.7

°

6

2

3

4



0.6 05

Q4

0 4

0.3

03

0.2

0.2

01

0.1

0.0

00 1

2

3


• - Two words (n - 10 strings)

•°- True*i words (n - 10 strin gs) •-

Four words (n - 10 strin gs)

Figure 6.2. Performance of CN and TI on word lists.

1

2

3


4


151

6.4. Effects of lexical parameters on STM performance 6.4.1. Frequency and imageability Frequency and imageability are two parameters known to affect lexical processes such as word recognition and retrieval. Thus, frequency effects are demonstrable in lexical decision (e.g., Gordon, 1983), in object naming (e.g., Oldfield & Wingh'eld, 1965), and in a variety of other tasks , including list repetition (Watkins & Watkins, 1977). Imageability effects are observed in lexical decision (James, 1975) and memory tasks (Paivio, 1979) in normal subjects, and are very characteristic of the performance patterns of patients with left hemisphere brain damage (e.g., Saffran, 1982). In most instances, patients have greater difficulty with low- than high-imageability words, although occasional exceptions to this pattern have been reported (Warrington, 1975, 1981). If the STM patients are dependent on lexically-based storage capacities, these parameters should affect their performance on STM tasks.

Methods Sets of 60 words of the following types were prepared: high frequency (frequency greater than 35/million in the Kucera & Francis [1967] list)-high imagery (imagery value greater than 4.97 in the Paivio, Yuille, & Madigan [1968] list) (HiF-Hil); low frequency (less than 25/million)-high imagery) (LoF-Hil); high frequency-low imagery (imagery value less than 4.97) (HiF-Lol); and low frequency-low imagery (LoF-Lol).1 Each set contained 45 two-syllable and 15 three-syllable words. Fifteen 4-word strings were constructed from each set in semirandom fashion, with the constraint that each string contain a single trisyllabic word and that the trisyllabic items occupy the same serial positions across list types. The 60 strings were randomized in a single list and read to the subjects by one of the authors (NM) at a rate of one item per second. They were instructed to repeat the items in the order in which they were presented.

Results and discussion Two normal subjects aged 70 and 76 had little difficulty with this task; they achieved overall scores of 96 and 95%, respectively, and made no more than 10% errors on any condition. As expected from their performance on other four-item lists (Figures 6.1 and 6.2), both patients performed poorly. Scored with respect to serial position, CN produced 44% of the items correctly and TI, 50%. The data for items produced irrespective of order are presented in Figure 6.2>. As in the word list repetition task (see section 6.4.2.), the two patients differ markedly in their performance across serial positions. Combining across frequency levels (HiF-Hil + LoF-Hil vs. HiF-Lol 4- LoF-

152

Saffran and N. Martin C.N.

T.I.

1.0 0.9 0.8 0.7 0.6 Proportion of words repeated 0.5 accurately in any order 0.4 0.3 0.2 0.1 0.0 •• 0

1 2 3 Position in input string

1

2

3


"•" High frequency-High imagery •O- Low frequency-High imagery • " High frequency-Low imagery n

" Low frequency-Low imagery

Figure 63. Performance of CN and TI on word lists varied for frequency and imageability. Lol), TI showed a significant effect of imageability on total items recalled (^2[1] = 6.64, p < .01); combining across imageability levels (HiF-Hil 4- HiF-Lol vs. LoF-Hil + LoFLol), he also showed a marginally significant effect of frequency (#2[1] = 3.27, .05 < p < .10) on total items recalled. CN showed no effect of imageability (#2[1] = .02) and a marginally significant effect of frequency (/2[1] = 2.93, .05 < p < .10) on total items recalled. Although CN's total item score was not affected by imageability, her ability to retrieve the item from the initial list position was strongly influenced by this factor; combining data for first item recall across frequency levels, the effect of imageability is highly significant (#2[1] = 14.1, p < .001). With the low-imageability lists, she showed a marked tendency to respond by producing the third or fourth item first; this occurred on 60% of the Hil lists and on 93% of the Lol lists {f [1] = 6.67, p< .01). While TI's performance on initial items was also affected by imageability (x2 [1] = 18.3, p < .001),


153

Table 6.2. Analysis of intrusion errors

TI

CN

38

31

2 13 1

1 4 0

Present as target Present only as response

8 8

18 3

Extralist intrusions

6

5

Total intrusions

Relationship to target Semantic Phonological Semantic + phonological

Intrusions from earlier test trials

he very rarely initiated recall with an item from the latter part of the list; this occurred on only three trials, all of them in the HiF-Lol condition. Examining lexical influences across serial positions, it appears that frequency and imageability affect performance at different list locations. Inspection of Figure 6.3 indicates clearcut effects of imageability but no consistent effect of frequency at the first serial position (although it should be noted that ceiling [in TI] and floor [in CN] effects limit the observation to a single condition per subject). In the case of the terminal item, in contrast, performance is consistently better for the high-frequency items, while imageability appears to have little effect. Combining across imageability conditions, the effect of frequency on recall of the terminal item is significant in TI (#2[1] = 4.51, p < .05); CN shows a trend in the same direction (x2 [1] = 1.98, p > .10). These data implicate different capacities in the retention of early and late list items in these patients. The imageability effect, which emerges at early list positions, suggests that the maintenance of these items is supported, at least in part, by semantic structures. The influence of frequency on recall of items from the end of the list implicates a mechanism at the level of phonological form (e.g., Bock, 1987) in the retention of these items. These points will be addressed more fully in section 6.6. Additional clues to the storage capacities utilized by these patients were sought in an analysis of their substitution errors in the serial recall task. The error categories employed were as follows: semantic relationship between the error and any item in the target string; phonological relationship, defined by presence in the error of at least 50% of the phonemes of a target item; semantic and phonological relationship to a target item; intrusion of a target item from an earlier string; intrusion of a nontarget word that appeared as a response to an earlier string; and new extralist intrusion. The data are summarized in Table 6.2. TI's errors primarily involve substitutions that are phonologically related to the target and intrusions of items that occurred earlier in the list. In view of his relatively

154


poor performance on the phonological tasks summarized in Table 6.1, it is possible that the phonological errors represent perceptual confusions. The majority of CN's errors (58%) were intrusions of targets from earlier lists. Although both patients produced some semantic errors, these represent only 5.8% of the total intrusions. We do not have normative data for these particular lists, but the patients' semantic error rate is no greater, certainly, than the 9.5% semantic error rate reported by Drewnowski and Murdock (1980) in a study of normal (supraspan) serial recall. The relatively low rate of semantic substitutions suggests that the patients are unlikely to be generating responses on the basis of semantic representations of the target items. Semantic influences are implicated, however, by the effects of imageability in this task. An attempt to resolve these seemingly incompatible findings will be left for the General Discussion (section 6.6). For the moment, we need to turn our attention to a possible artifact in CN's data. We have taken CN's poor performance on items at the beginning of the list to indicate a selective difficulty in the retention of these items. It is necessary, however, to consider an alternative explanation. Suppose that her impairment is essentially the same as TI's, but that instead of complying with the serial recall instructions she has opted for the sort of response strategy that normals adopt in free recall (e.g., Baddeley, 1976) - a strategy of reporting the items from the end of the list immediately, to minimize the risk of losing them. The consequence of such a report strategy could well be to depress CN's performance on the initial items. To rule out report strategy as an explanation for her failure to reproduce items from the beginning of the list, CN was given a task in which she was required to report only a single word.

6.4.2. Probe test CN performed an immediate memory task in which a single serial position was probed on each trial. To examine lexical effects in probe recall, lists were again varied for frequency and imageability. For maximal contrast, the materials were drawn from the HiF-Hil and LoF-Lol conditions of the preceding study. From the 60 words in each of these conditions, four sets of 30 four-word strings, each string uniform in material type, were constructed. Each word appeared once in each block, in each instance at a different serial position. An equal number of HiF-Hil and LoF-Lol items was probed, 15 of each at each serial position; each word was probed only once. The blocks were administered over four successive weekly sessions. Presentation conditions were the same as in the serial recall study in section 6.4.1. Immediately upon termination of the list, the experimenter indicated the target by pointing to one of four cards, arrayed horizontally before the subject, which were numbered to correspond to the four positions in the string. The data are summarized in Table 6.3. As in the serial recall task, CN showed a


155

Table 6.3. Performance of CN on probe task A. Overall performance Position of item reported (proportion of responses) 1

2

4

3

Other

HiF-HiJ lists Position probed I(w = 15) 2(« = 15) 3(n = 15) 4(w = 15)

.27 .07

.20 .60 0 0

0 0

.20 .20 .27 .20

.07 .07 .27 .80

.47 .27 .27

.27 .33 .53

.27 .07 .27 0

LoF-LoI lists Position probed l(n = 15) 2(n = 15) 3(w = 15) 4(n = 15)

.07 .07 .07

.13 .13 .07 0

0

1.00

0

.07 .20 .07

0

B. Likelihood of late list items Jsubstituting for early list items Position of response iterri (number of responses)

Probe item: 1 or 2 HiF-Hil LoF-LoI

1 or 2

3 or 4

17 6

8 20

tendency to substitute items 3 and 4 for the earlier list items, particularly in the LoF-LoI condition. The effect of material type is clearly demonstrated in part B of this table, which contrasts performance on the two halves of the list; the difference in performance pattern as a function of list type is significant (x2 [1] = &.65, p < .01). Thus CN demonstrates essentially the same pattern on the probe task that she did in list repetition: Her responses are biased toward the most recent list items, particularly in the LoF-LoI condition. Replication of this pattern under minimal report conditions indicates that the difficulty in retrieving early list items in the serial recall task is not simply a consequence of choosing to report the terminal items first.

6.4.3. The effect of adding syntactic context What happens if CN is "forced" to report the items in correct serial order? It was not possible to enforce first item report in the serial recall task, but some approximation was achieved by having her repeat lists presented in sentence form.

156


Table 6.4. Repetition of unrelated words in list and sentence form: CN (proportion of words repeated correctly as a function of position in string; n = 24 words per cell) Position in input string Condition

1

2

3

.04 .42

.04 .13

0

.42 .42

25 .38

4

Total (n = 96)

In serial order

List Sentence

.13

0 .17

.02 .21

.83 .38

.88 A2

.59 .39

In any order

List Sentence

The same set of open class morphemes was presented as four-word lists and as sentences. In the sentence condition, stimulus words were inflected where appropriate (e.g., life, woman, state, school vs. The life and the woman are stating the school).2 Although

syntactically well formed, the sentences lacked semantic coherence. The materials, which consisted of two blocks of four-word lists and two blocks of matched sentences (N= 12/block), were presented over two sessions in an ABBA design. CN's performance is summarized in Table 6.4. Syntactic context does appear to have the intended effect, in that she was much more likely to initiate recall with the first item in the sentence condition. There is no indication, however, that reporting the initial item first improved performance on this item. Overall, the probability of recalling an openclass item, irrespective of position, was in fact higher in list than in sentence contexts (McNemar Test: x2 Ul = 9.03, p < .01). Although the greater length of the sentence strings might be a factor in these results, the pattern of performance across serial positions suggests that the superiority of the list condition is primarily due to better retention of the last two items, which were more likely to be reported first in list than in sentence contexts. But while delay in report may account for the loss of terminal items in the sentence condition, it does not explain the general difficulty with early list items; if delay were the problem, retrieval of the initial items should have been facilitated in the sentence condition, where they were more likely to be reported first. The results of this study therefore argue against the view that CN's failure to report items from the beginning of a list is an artifact of response strategy. They suggest, rather, a primary impairment in the retention of early list information.

6.4.4. Summary We have presented data from two patients with span limitations that indicate that their performance on STM tasks is subject to lexical influences, and, further, that the two


157

factors investigated - imageability and frequency -affect performance at different points in the list. The performance of one of the patients (CN) is also characterized by particular difficulty in the retrieval of items from the beginning of the list. In the remainder of the experimental section of this chapter, we focus on this aspect of CN's impairment.

6.5. Factors affecting CN's performance As we have noted, CN's performance across several different tasks is marked by a strong bias toward retrieval of items from the end of the list. In this respect, her performance pattern resembles that of the transcortical patient (ST) described in section 6.1 (Martin & Saffran, 1987). ST had general difficulty retrieving lexical information except in response to phonological or orthographic input; in STM tasks, this was manifested as a difficulty in recalling any but the last two items. Although CN is less impaired in lexical tasks, her performance pattern in list repetition tasks is similar to ST's. On the hypothesis that CN's difficulty with early list items is also the reflection of a deficit involving lexical storage capacity, we examined her performance across a set of tasks in which lexical support would be expected to vary. In the first study, the lexical contribution was minimized by the use of nonwords. In other studies, we introduced manipulations that would be expected to increase support from lexical structures, namely, semantic similarity and repeated presentation.

6.5.1. Repetition of nonwords CN was asked to repeat strings of nonwords ranging from one to four items in length. The nonwords were constructed by changing 50% of the phonemes in the materials from the word list repetition task in 6.5.1. (e.g., brush lamp —• bleesh sump). The lists were blocked for string length and taped for presentation at the standard rate of one item per second. CN's performance is summarized in Table 6.5. While CN performed fairly well on one- and two-item strings, she broke down completely on the three-item lists; on the four-item strings she apparently gave up entirely on the early items and initiated recall with the terminal item, which she reported correctly on 8/10 trials. Normal subjects have been observed to adopt a similar strategy with nonword lists of four or more items (Watkins, 1914). If we look at CN's performance across four-item lists of words and nonwords, a clear pattern emerges. With nonword strings, she reported only items from the terminal position; with low-imageability words, she reported approximately 60% of the words at each of the last two positions but no more than 15% of the items from the first half of the list; with high-imageability lists, she was able to add about 60% of the initial list items. It is evident from these data that CN's performance is based on both

158


Table 6.5. Repetition of nonwords and nonword strings: CN (proportion of items repeated correctly as a function of position in string; n = 10 words per cell) Position in input string String length In serial order 1 2 3 4 In any order 1 2 3 4

1

3

2

.90 .70

.10 .90 .70

0 0

Total

0

.90(n = .70(n = 0 (n = .03(n =

10) 20) 30) 40)

= .70(n = .I3(w = .28(n =

10) 20) 30) 40)

.70 0

0 0

4

0 0

.90(M

.70 .10 .10

.30 .20

.80

phonological and lexical information. In nonword tasks, where recall depends solely on phonological storage capacities, she is only able to retrieve the terminal item. The stronger the lexical influence, the greater the likelihood of retaining items from nonterminal list positons. Evidently, the strength of the lexical influence is a function of imageability.

6.5.2. Semantic similarity Three types of four-word strings were presented for immediate repetition: strings in which all the items were semantically related; strings in which all the items were phonologically related (i.e. rhyming words); and strings of unrelated words drawn randomly from the other two lists. The lists were matched for syllable length, but could only be roughly equated for frequency (mean Kucera-Francis frequencies: semantic = 31; phonological = 53; unrelated = 36). The lists were read to CN at the standard rate of one per second. She was instructed to attempt to reproduce the items in the order in which they were presented. The data are summarized in Table 6.6. CN again demonstrates her characteristic pattern on the unrelated lists, performing poorly on all but the terminal item. Semantic similarity markedly increased her ability to retrieve items from earlier positions. Although guessing on the basis of semantic constraints cannot be ruled out as a factor in these results, it seems likely that the improvement on early list items derives at least in part from the strengthening of lexical traces due to semantic priming.


159

Table 6.6. Repetition of semantic, phonological, and unrelated four-word strings: CN (proportion of words correct irrespective of order as a function of position in string; n = 20 words per cell) Position in string String condition

1

2

3

4

Total (n = 80)

Semantic Phonological Unrelated

.80 .40 .20

.65 .30 .15

.80 .45 .20

.85 .90 .85

.78 .51 35

Table 6.7 Effect of repeated presentation on list repetition: CN (proportion of items correct at each serial position) Position in input string 1

2

Unrepeated strings (n — 80) HiF-Hil LoF-Hil HiF-Lol LoF-Lol

.38 .38 .06 .13

Total Repeated strings (n = 40) HiF-Hil LoF-Hil HiF-Lol LoF-Lol Total

4

3

Total

5 .94 .88 .88

.06 .06

.25 .19 .13 .06

.63 .63 .63 .56

1.00

.24

.03

.16

.61

.93

.13 .50 .25

1.00

.22

.53

0

0 0

.50 .38 .25

.63

.75

.63

1.00

.25

1.00 1.00 1.00

.22

.94

.88

0 0

.88

1.00

.45 .43 .34 .35

.50 .63 .55 .55

6.5.3. Effect of repetition The Hebb (1961) paradigm, in which the same list of items recurs on every third trial, was used to examine effects of repeated presentation (repetition priming). Strings of five words generated from the items used in the frequency-imageability study in 6.4.1 were the stimuli in this task. The number of syllables per string was controlled across list types. The lists were organized in four blocks made up of 24 lists each; in each block, the same list appeared on every third trial. The recurring string in each block was drawn from a different frequency-imageability condition. The remaining lists were randomly interspersed among the four blocks. The task was

160


administered in four sessions, one block per session, using the same presentation conditions as described earlier. The results are summarized in Table 6.7. With respect to total items recalled, performance is clearly superior on the repeated lists (/ 2 [1] = 12.96, p < .001). Detailed quantitative comparisons are problematic, since there was only a single list of each type in the repeated condition. It is evident from the table, however, that CN's improvement on the repeated lists reflects better performance on early list items. The effect of repetition priming is presumably analogous to that of semantic priming: Repeated presentation of list items strengthens lexical traces associated with those items, increasing the likelihood that they can be retrieved on subsequent trials.3 The mechanisms underlying these effects will be discussed in more detail in the following section.

6.6. General discussion The performance patterns of two patients with limited spans provide evidence for lexical influences in STM. Both patients show effects of lexical parameters, particularly of imageability, that vary across list positions, and one of them (CN) demonstrates a pattern that is interpreted to reflect decreased levels of support from lexical structures. Let us consider, first, how these phenomena might be dealt with on the standard model of STM (e.g., Baddeley, this volume, chapter 2), which accounts for span performance in terms of a phonological store, coupled with an articulatory loop that allows it to be refreshed by means of subvocal rehearsal. On the basis of their reduced digit spans, and the suggestion of information loss at recency positions (see Figure 6.1), CN and TI qualify as deficient in phonological storage capacity on this model. Adopting the standard argument (as,- for example, in Butterworth, Shallice, & Watson, this volume, chapter 8), their performance even on short lists of words must therefore depend on LTM. Since LTM is known to be sensitive to semantic influences, such as concreteness - imageability, it is not at all surprising that these parameters affect patients' performance on STM tasks. Effects of semantic similarity and repeated presentation are also consistent with performance based on LTM. The differences that emerged in the patients' ability to recall early list items can be dealt with, on this model, by assuming that CN's phonological capacities are even more limited than TI's, or that she has an additional difficulty involving long-term storage. There is nothing in the evidence we have presented that would refute such an account. We do not, however, find it a particularly cogent explanation of the data. Attributing these effects to LTM does not elucidate the mechanisms underlying the patients' performance. Here, as in the case of phonological storage capacity, pointed out earlier, the model fails to address crucial representational issues. Elsewhere, one of us has argued for conceptualizing STM phenomena within the framework of a language-


161

processing model (Saffran, in press; see also Barnard, 1985) in which representational issues are a central concern. The arguments for a language-based STM model will not be recapitulated here. Our purpose in this discussion is to illustrate how the data from the present studies might be accommodated within a model of this type. A basic assumption of this approach is that performance on span-type tasks depends on capacities for information storage that are utilized in the normal course of comprehending spoken language and producing it. A further assumption is that the system is highly interactive, so that the various representations contacted in the course of repeating a list of words - in perceiving the input string and reproducing it - are mutually reinforcing. The phonological store that has been implicated in STM performance is construed, on this model, not as a unitary record of the input string but as a system in which a phonological record is supported by feedback from lexical units. We assume that item and order information are represented at the phonological level in the form of a sequence of segmentally and prosodically specified units, which are interpretable as words by virtue of their associations with lexical units; item information is also represented at the lexical level, while order is not. In addition to receiving support from the lexical level, the phonological representation can be further reinforced by information that is maintained at an auditory level of representation; and - to continue the chain of interacting units - just as lexical activation helps to stabilize the phonological record, lexical traces are, in turn, supported by semantic information (see Saffran, in press, for further discussion). The multilevel, interactive framework proposed here is not simply a terminological variant of more traditional models, in which STM is viewed as a phonological store and semantic effects are attributed to LTM (as, e.g., in Shallice, 1975). We expect that, as does recent work on speech production (Dell, 1986), the provision for interaction among these levels of representation will have explanatory and predictive value. Similar approaches to modeling short-term storage have been advocated by Allport (1986) and McClelland and Elman (1986). How do we account for the present data within this framework? Consider, first, the effect of frequency on recall of terminal list items, demonstrated most convincingly in TI's data. Frequency is thought to affect lexical processes at the level of phonological form, in word production (Bock, 1987) as well as in word recognition (Forster, 1976). In serial recall tasks, frequency comes into play when the phonological trace is weak; under these conditions, lexical involvement is critical for maintenance of the item in memory, and frequency becomes an important determinant of lexical activation. This would account for the pattern of frequency effects in normal serial recall. Frequency affects normal performance on supraspan (eight-item) lists, at all but the last two serial positions (Watkins & Watkins, 1977), where the information is presumably adequately specified by the auditory-phonological record; at earlier list positions, where the phonological record is degraded, access to lexical information is more critical. The situation is different in STM patients, who are deficient in

162


phonological storage capacity; they require more lexical support for terminal list information and would therefore be expected to demonstrate frequency effects for terminal list items. This effect was indeed demonstrated in TI. The frequency effect was less reliable in CN's data; however, as she tended to produce the last two items first, it is likely that these items were more fully specified by the auditory—phonological record. The second major feature of our data was the imageability effect, which was demonstrable at early list positions in both subjects. This effect implicates semantic factors in the retention of early list information. Effects of imageability and concreteness are, in general, poorly understood, in normals (e.g., Paivio, 1979) as well as in patients (e.g., Saffran, 1982). Consider, however, the following possibility: Assume, as suggested earlier, that feedback from activated representations at a semantic level helps to stabilize units activated at a lexical level. If so, the degree of stabilization provided is likely to depend on the number of activated semantic nodes. It has frequently been suggested that the differences that emerge with concrete and abstract words reflect differences in the quality of their associative networks (e.g., Schwanenflugel & Shoben, 1983; Jones, 1985). For example, Jones (1985) found imageability ratings to be highly correlated with ratings of "ease of predication," that is, the judgment of how easy it would be tp generate factual statements for a given lexical item; ease of predication is presumably a reflection of the accessibility of semantic information pertaining to the item. Following the line of argument we have been developing, it seems conceivable that the increased support provided for concrete words at a semantic level helps to maintain activation at a lexical level, which, in turn, helps to stabilize the phonological trace. Evidence for concreteness effects in the performance of normal subjects on the Brown-Peterson task is consistent with this proposal (Borkowski & Eisner, 1968). To recapitulate our discussion thus far: We have suggested that semantic information feeds back, by means of lexical nodes, to support the activation of a phonological trace that normally increases in strength across serial positions, and that the support provided by semantic representations is greater for high-imagery words. We assume that these semantic influences are not ordinarily detectable because the interaction of lexical and phonological representations is sufficient to support normal span performance. Thus, Brener's (1940) data indicate only an approximately 5% advantage for concrete words in normal span (although it is interesting to note that the difference increases to 9% with visual presentation, where there is less support from auditory—phonological information). When there is less phonological support, as in the Brown-Peterson task (Borkowski & Eisner, 1968), or as a consequence of brain damage in these STM patients, the semantic effects are unmasked. But why do the patients only show these effects at early list locations? One possibility is that there are differences in the way attentional capacity is deployed across serial positions. Because there are fewer items to keep in mind, attention is likely to be more narrowly focused on an incoming item at the beginning of the list than as list presentation progresses. Focused attention could


163

facilitate the processing of early list items, increasing the extent of activation at a semantic level. This is, of course, highly speculative, but not dissimilar to accounts of the role of attention in the processing of visual information (e.g., Treisman & Gormican, 1988). Finally, we need to consider the matter of the differences between our two subjects. Although TI did show an imageability effect at early list positions, he had no difficulty at all reporting the initial item in high-imageability lists. In contrast, CN had difficulty even with high-imageability items, and her performance, in general, was biased toward report of items from the end of the list. The likeliest explanation of the difference, in terms of the account we have been developing, is that CN receives less support than TI from lexical representations. If so, one might expect to find CN deficient, relative to TI, in other tasks involving lexical processes. Unfortunately, TI was not available for further study, and examination of CN's performance alone was not very revealing. CN performed reasonably well in object naming tests (Table 6.1), although her spontaneous production pattern was characterized by hesitancies and circumlocutions, suggesting that latency measures might reveal significant abnormalities. She had particular difficulty retrieving low-imageability words in these STM tasks, but it is not clear whether this pattern reflects differential impairment of this class of lexical items; TI showed the same pattern, and, as, we have noted, on some tasks, normal subjects do, too. CN did not, in any case, show comparable difficulty with abstract words on other tasks, such as lexical decision and oral reading with tachistoscopic presentation. In a word association task, she did appear to have more difficulty generating associates for abstract than for concrete words, but we lack relevant normative data for comparison. The question of the nature of CN's lexical impairment, if any, must therefore be left unresolved. It is worth considering, however, whether her deficiency might reflect dynamic properties of the lexical system that figure importantly in STM performance but that are not, in general, sensitively addressed by the tasks that we use to assess lexical capacities in patients. The approach we have taken here seems a good fit to the kinds of phenomena that have emerged in studying STM patients, and may turn out to be particularly useful in accounting for the range of deficits found in this population. It also has much in common with interactive models of language production, and like them (e.g., Dell, 1986), should lend itself well to computer simulation. What is perhaps less obvious is that this will prove to be a useful framework for the study of normal STM. It could be argued, for example, that too much is being made of performance patterns that reflect patients' reliance on auxiliary mechanisms that normally contribute little to STM. There is no doubt that normals rely heavily on phonological information in these tasks, and that these patients are less well served by phonology. But it is undeniable that lexical structures contribute to normal performance on STM tasks, for if normals were reading off a purely phonological record, span for nonword lists should not be approximately

164


half as long as span for familiar lexical items (Brener, 1940). The study of brain-damaged subjects can unmask influences that may be easy to overlook in circumstances where performance is overdetermined. In this instance, the neuropsychological data draw attention to lexical effects in STM that have, in our view, too long been neglected.

Notes 1. Due to the constraints on item selection, it was necessary to include some words for which frequency or imagery values were not available. There were 5 words used that were not included in the Kucera and Francis (1967) frequency lists. Imagery ratings were not available in Paivio et al. (1968) for 5 words in the HiF-Hil condition, 19 words in the LoF-Lol condition, and 15 words in the HiF-Lol condition. To justify the inclusion of these words in the set of test stimuli, we obtained imagery ratings of the entire set of words using the same procedure as Paivio et al. (1968). The set of 240 words was rated by each of 17 subjects. The overall mean rating was 4.80; listed here are the mean imagery values of test items by condition as well as the mean frequency values of those items listed in Kucera and Francis (1967).

Test condition

Frequency

Imagery

HiF-Hil LoF-Hil HiF-Lol LoF-Lol

112.4 8.7 113.8 11.7

6.3 6.7 3.3 3.0

2. We thank Karen Nolan for making this task available to us. 3. The interpretation of repetition priming effects is controversial. Some have argued that these effects reflect recovery of traces entered in an episodic memory system (e.g., Jacoby, 1983), while others account for them in terms of changes in lexical structures (e.g., Monsell, 1985), as we do here.

References Allport, D. A. (1986). Distributed memory, modular subsystems and dysphasia. In S. Newman & R. Epstein (Eds.), Current perspectives in dysphasia (pp. 32-57). New York: Churchill Livingstone. Baddeley, A. (1976). The psychology of memory. New York: Basic Books. Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). London: Erlbaum. Bock, J. K. (1987). Co-ordinating words and syntax in speech plans. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 3, pp. 337-390). London: Erlbaum. Borkowski, J. G., & Eisner, H. C. (1968). Meaningfulness and abstractness in short-term memory. Journal of Experimental Psychology, 76, 57-61. Brener, R. (1940). An experimental investigation of memory span. Journal of Experimental Psychology, 26, 467-482. Caramazza, A., Basili, A. G., Koller, J., & Berndt, R. S. (1981). An investigation of repetititon and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271.


165

Coslett, H. B. (1986). Dissociation between reading of derivational and inflectional suffixes in two phonological dyslexics. Paper presented to the Academy of Aphasia, Nashville, TN, October 1986. Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93, 283-321. Drewnowski, A., & Murdock, B. B. (1980). The role of auditory features in memory span for words, journal of Experimental Psychology: Human Learning and Memory, 6, 319-332. Forster, K. I. (1976). Accessing the mental lexicon. In R. J. Wales & E. Walker (Eds.), New Approaches to Language Mechanisms (pp. 257-287). Amsterdam: North Holland. Gordon, B. (1983). Lexical access and lexical decision: Mechanisms of frequency sensitivity. Journal of Verbal Learning and Verbal Behavior, 22, 24-44. Hebb, D. O. (1961). Distinctive features of learning in the higher animal. In J. F. Delafresnaye (Ed.), Brain mechanisms and learning: A symposium (pp. 37-51). Oxford: Blackwell Scientific Publications. Jacoby, L. L. (1983). Perceptual enhancement: Persistent effects of an experience. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38. James, C. T. (1975). The role of semantic information in lexical decisions. Journal of Experimental Psychology: Human Perception and Performance, 1, 130-136. Jones, G. V. (1985). Deep dyslexia, imageability and ease of predication. Brain and Language, 24, 1-19. Kucera, H., & Francis, W. N. (1967). Computational analysis of present day English. Providence, RI: Brown University Press. Martin, N. & Saffran, E. M. (1987). Factors underlying repetition performance in a transcortical sensory aphasic. Unpublished manuscript. McClelland, J., & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86. Monsell, S. (1985). Repetition and the lexicon. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 147-195) London: Erlbaum. Oldfield, R. C, & Wingfield, A. (1965) Response latencies in naming objects. Quarterly Journal of Experimental Psychology, 17, 273-281. Paivio, A. (1979). Imagery and verbal processes. Hillsdale, NJ: Erlbaum. Paivio, A., Yuille, J., & Madigan, S. (1968). Concreteness, imagery, and meaningfulness values for 925 nouns. Journal of Experimental Psychology Monograph, 76, (1, Pt. 2). Richardson, J. T. E. (1979). Precategorical acoustic storage and postcategorical lexical storage. Cognitive Psychology, 11, 265-286. Saffran, E. M. (1982). Neuropsychological approaches to the study of language. British Journal of Psychology, 73, 317-337. Saffran, E. M. (1985). Short-term memory impairment and language comprehension: Specifying the nature of the interaction. Paper presented at the Second Venice Conference on Cognitive Neuropsychology, Venice, March 1985. Saffran, E. M. (in press). Short-term memory impairment and language processing. In A. Caramazza (Ed.) Advances in cognitive neuropsychology and neurolinguistics. Hillsdale, NJ: Erlbaum. Salter, D., Springer, G., & Bolton, L. (1976). Semantic coding versus the stimulus suffix,. British Journal of Psychology, 67, 339-351. Schwanenflugel, P. J., & Shoben, E. J. (1983). Differential context effects in the comprehension of abstract and concrete verbal materials. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 315-328. Schwartz, M. F., Saffran, E. M., & Marin, O. S. M. (1980). The word order problem in agrammatism I: Comprehension. Brain and Language, 10, 249-262. Shallice, T. (1975). On the contents of primary memory. In S. Dornic and P. Rabbitt (Eds.), Attention and performance V (pp. 269-280). New York: Academic Press.

166

Saffron and N. Martin

Treisman, A. M, & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review, 95, 15—48. Warrington, E. K. (1975). The selective impairment of semantic memory. Quarterly Journal of Experimental Psychology, 27, 635-657. Warrington, E. K. (1981). Concrete word dyslexia. British Journal of Psychology, 72, 175-196. Watkins, M.}., & Watkins, O. C. (1977). Serial recall and the modality effect: Effects of word frequency. Journal of Experimental Psychology: Human Learning and Memory, 6, 712-7IS. Watkins, S. H. (1914). Immediate memory and its evaluation. Journal of Psychology, 7, 319-348.

7. Auditory-verbal span of apprehension: a phenomenon in search of a function? ROSALEEN A. McCARTHY AND ELIZABETH K. WARRINGTON

7.1. Introduction Memory span for spoken lists of random digits, letters, and words has been the subject of numerous experimental investigations in normal subjects. The basic phenomena of the span task have been well established. It is known that only a limited number of items can be retained, that storage appears to be based on phonological representations, and that in the absence of rehearsal these items are susceptible to very rapid forgetting. However, the functional significance of this short-term representation, the auditory-verbal span of apprehension, remains somewhat mysterious. Span for random lists of spoken material may be gravely and very selectively impaired in patients with brain damage. Analysis of preserved and impaired skills in these cases has been used as a means of investigating the normal functional role of the short-term representation that is measured by span. Thus, by establishing what other abilities are preserved, and what abilities are impaired, we can go some way towards a specification of the types of information processing that require the integrity of this level of representation. Such an approach is not without its difficulties; for example, failure on a task may be attributable to associated disorders that happen to arise as a consequence of damage to areas that are functionally independent but anatomically close together. This means that it has been easier to establish independence of this type of representation from other forms of processing (or processing systems), rather than their necessary relationship. Thus in the 1960s a widely held model was that short-term representations were a necessary precursor for long-term memory. Early studies of "span-impaired" cases showed that performance on span tasks was independent of longer-term retention for the same type of material (Warrington & Shallice, 1969; Shallice & Warrington, Rosaleen McCarthy's research is funded by the University of Cambridge. Further support is provided by the Charles Slater and Grindley funds. We wish to thank Dr. P. Rudge and Dr. J.N. Blau for allowing us to work with the patients under their care. Much of the material reported in this chapter is based on McCarthy and Warrington (1987a, b). We are grateful to Brain for permission to reproduce Figure 7.1, which was initially published in McCarthy and Warrington (1987a).

167

168

McCarthy and Warrington

1970). These studies permitted a fractionation of short- and long-term memory processes; they are now largely considered to be independent. However, having contributed to the elimination of one theory of the functional role of this phenomenon, further investigations have failed to provide a satisfactory alternative. The most frequently expressed opinion with regard to the role of the level of processing measured by span is that it may be directly involved in some aspect of language comprehension. It is widely accepted that patients with an impaired span do not appear to have any difficulty in understanding or participating in everyday conversation. They respond appropriately to questions, and appear to cope with the maintenance of discourse topics, and the comprehension (or production) of topic shifts, quite normally. The relative integrity of their language comprehension abilities is in part confirmed by evidence that span-impaired patients may score within the normal range on tasks requiring them to name items from lengthy oral descriptions (e.g., What is the name of the thin grey dust that remains after something is burned such as a cigarettel).

Nevertheless, and without exception, patients with an impaired auditory span have been described as showing deficits when they are assessed on certain more subtle tasks that have been designed to assess sentence processing (e.g., the Token Test). However, there is no consensus as to the precise nature of the linguistic operations for which an adequate span of apprehension is either useful or necessary. A very broad subdivision can be drawn between current theoretical accounts in terms of their emphasis on the role of short-term representations in the contemporaneous (or "on-line") analysis of spoken language. Two major perspectives can be identified, specifically: 1. Short-term representations are implicated in the operation of a crucial component of the "online" language processor. They may be of particular importance in parsing the spoken utterance prior to further linguistic analyses (e.g., Heilman & Scholes, 1976; Frazier & Fodor, 1978; Hitch, 1980; Marcus, 1980; Ellis & Beattie, 1986). 2. Short-term memory is a back-up resource that is required in those situations in which on-line contemporaneous linguistic processing is inadequate for comprehension. Under such conditions a verbatim record may be used to backtrack over spoken input and provide a basis on which a sentence can be reanalysed for comprehension (e.g., Shallice & Warrington, 1970; Saffran & Marin, 1975; Baddeley, 1976; Caramazza & Zurif, 1976). These perspectives should probably not be viewed either as competing hypotheses or, more critically, as complementary descriptions of a single flexible processing system. It may be inappropriate to view the representations measured by span as the only form of auditory-verbal short-term memory. There may be multiple levels of "short-term" representation that are implicated in language processing, only one of which is measured by performance on list retention tasks. In this chapter we shall present evidence that we believe requires the integration of both of these positions within a single multicomponent model. We will review a series of experiments previously reported by us (McCarthy & Warrington, 1987a, b). We were led to undertake these investigations by an anomaly

A phenomenon in search, of a function?

169

in the language-processing skills of span-impaired patients: We observed two cases whose span was maximally two items but who were nevertheless able to repeat novel sentences with an unexpected degree of competence (McCarthy & Warrington, 1984). It was far from obvious why there should be relative preservation of sentence repetition if these patients had a fundamental impairment in their "on-line" sentence-processing abilities.

7.2. Experimental section 7.2.1. Case reports The case reports of the patients are summarized here. (Further details are available in McCarthy and Warrington, 1987a.)

Case 1

RAN was a 54-year-old (date of birth: January 16, 1932) plumber who sustained an intracerebral haematoma in the parietal region of the left hemisphere that was treated conservatively. He had mild speech production difficulties and was severely impaired on span tests. His span was short, and only reliable for one item (see Table 7.1). He also forgot information exceptionally quickly, scoring 4/20 correct in recalling a single auditory letter following 5-sees distraction. His language skills were otherwise relatively well preserved. In particular his comprehension both for single spoken words and for many spoken sentences was considered to be consistent with his premorbid level of competence. Thus on the Peabody Picture Vocabulary he obtained an average score, and he performed at the level of controls on Lesser's (1974) syntax test. Despite his good single word comprehension and excellent performance on the Lesser test, his performance on the revised Token Test (De Renzi & Faglioni, 1978) was impaired: He scored 22/36.

Case 2

NHA was a 42-year-old (date of birth: November 21, 1944) right-handed executive who sustained a subarachnoid haemorrhage from a left middle cerebral aneurysm in 1983. The aneurysm was clipped. His major residual deficit was a severe impairment on tests of span (see Table 7.1). Not only was his span abnormally short but he also showed abnormally fast forgetting for even a single letter. Following 5-sees distraction, he scored only 9/20 correct. Other aspects of his language skills, apart from some hesitancy in his spontaneous speech, were relatively well preserved. In particular, his comprehension of single spoken words (Peabody Picture Vocabulary Test) was

170

McCarthy and Warrington Table 7.1. Span for digits and words (% correct for three-item lists n = 30)

Digit Words

Case 1

Case 2

Case 3

51 44

53 39

100 81

Table 7.2. Case 3: definition Frequency

of words

in three frequency ilands (in %)

High ( > 50)

Medium (25-50)

Low «25)

61 n = 36

46 n = 28

26 n = 153

superior, and his comprehension of spoken sentences (Lesser, 1974) was entirely satisfactory. He did, however, have marked difficulties on a shortened version of the Token Test (Coughlan & Warrington, 1978), scoring 3/15.

Case 3

NHB was a 62-year-old (date of birth: September 12,1924) right-handed mechanic who was investigated because of his progressive impairment of memory and language functions. His CT scan demonstrated focal widening of the sulci of the left temporal lobe and widening of the left lateral ventricle. His verbal comprehension at the singleword level was gravely impaired. He was asked to define 217 words from the Snodgrass and Vanderwart (1980) pool. His scores for three frequency bands are given in Table 7.2. In contrast to his poor single-word comprehension his repetition span for digits and words fell at the lower limits of normal for his age (see Table 7.1). The patient's deterioration in language abilities appeared to be highly selective; thus although his WAIS verbal \Q was reduced to 71 (due to very defective performance on the vocabulary and similarities tests), his performance IQ was 118.

7.2.2. Experiment 1 Our aim in this experiment was to make a direct comparison between list and sentence repetition using matched vocabularies for the two tasks. Ten sentence frames were generated such that there were two vacant noun slots in each. The six- and seven-word sentences were completed using either an abstract or a concrete word vocabulary, matched for frequency (e.g., He took care with the law; He took salt with the meal).

A phenomenon in search of a function?

171

Table 73. Sentence and list repetition {% correct; n = 20)

Complete sentence Complete list

Case 1

Case 2

Case 3

55 15

60 40

30 75

Matched three-word lists were derived from the sentences using the verb and two nouns (e.g., took, care, law, took, salt, meal). The patients were simply required to repeat either the whole sentence or the three-word list verbatim. (For a detailed description of this experiment see McCarthy and Warrington, 1987a.)

Results The patients' responses were scored with respect to both order and items. The results for each of the three patients on the sentence and list repetition tasks are shown in Table 73. There was evidence of a clear sentence advantage effect in the span-impaired patients. They were significantly better at repeating the complete sentence material as compared with the complete list. The converse result was obtained for Case 3. The error patterns were very different in the two types of patient. In repeating lists the two spanimpaired cases made errors of order and/or of omission. In repeating sentences Cases 1 and 2 made a few minor paraphrases, and occasional omissions of function words. For example, Case 1 repeated The plane can land and fly as The plane can fly and land; Case 2 repeated He had that coat for school as He had this coat for school. Case 3 tended to omit the final words from the stimulus sentence and also made errors of phonemic transposition. For example, he repeated Her cat likes to have milk as Her, like and He spent time on his art as He spent time hon his heart.

Comment This experiment provided evidence for a dissociation between list and sentence repetition. A relative sentence advantage for the span-impaired cases and a list advantage for the span-preserved case has been documented. This would not be expected if sentence repetition was directly related to span in any simple manner.

7.2.3. Experiment 2 In this experiment we investigated the processes involved in sentence repetition in greater detail by contrasting the patients' performance on repeating complete and incomplete sentences. The stimuli, derived from Bloom and Fischler (1980), comprised

172

McCarthy and Warrington Table 7.4. Complete and incomplete sentence repetition (% correct; n = 48)

Complete sentence Incomplete sentence

Case 1

Case 2

Case 3

54 25

43 10

37 42

48 sentences with a high-probability final word (e.g., London is a very busy city). In one condition the patients were required to repeat the complete sentence, and in another, the sentence was presented for verbatim repetition without the final word. In this condition the incompleteness of the sentence fragment was indicated by vocal pitch contour. (For a more detailed account of this experiment see McCarthy and Warrington, 1987a.)

Results

In scoring the patients' responses, only verbatim reproductions of the sentence, or of the incomplete sentence, were considered as correct. The percentage correct for the two conditions for each case is shown in Table 7.4. Qualitatively it was observed that the errors made by the span-impaired cases on the incomplete sentences consisted of (a) production of the complete sentence and (b) omission of penultimate elements of the sentence fragment. For example, London is a very busy was repeated as London is a very busy place. Their errors on the complete sentences were predominantly minor paraphrases. For example, The train was still on time was repeated as The trains are still on time. In contrast, the errors made by Case 3 were similar in both conditions; they consisted of (a) omissions of the later parts of the stimulus and (b) phonemic transposition errors. For example, Storms make the air damp and cold was repeated as Snail murks the air danad.

Comment

The span-impaired cases appeared to be deficient in the ability to "check" their anticipatory processing when incomplete sentences were presented, thus resulting in sentence completions and erroneous reconstructions.

7.2.4. Experiment 3 Case 3 was the focus of this experiment. Our aim was to investigate the role of lexical-semantic processing in sentence and list repetition tasks. In the course of our clinical investigation of Case 3 we were able to establish that he had a small but highly


173

Table 7.5. List vs. sentence repetition for known and

unknown

vocabularies {% correct)

List (n = 30) Sentence (n = 23)

Known

Unknown

90 61

90 17

stable vocabulary of concrete nouns. We contrasted his repetition of this known vocabulary (established on the basis of repeated clinical testing) with his repetition of words that he had "forgotten." Repetition was tested in two conditions: (a) in lists of three words and (b) in six- and seven-word sentences containing at least one item that we had independently determined was either known or unknown to him (see section 7.2.1). (The status of the remaining words in these sentences was not assessed.) As a baseline, his ability to repeat lists of three nonwords was also assessed.

Results

The percentage correct for the known and unknown word vocabularies in the sentences and list conditions is shown in Table 7.5. Although there was no effect of the known or unknown vocabularies in the list repetition condition, there was a highly significant effect of vocabulary type in the sentence repetition condition. His repetition of nonword lists was significantly worse than either of the word lists: He scored only 2/20 lists correct. Although there is an effect of lexicality in list repetition, semantic knowledge does not appear to be a significant variable. This was not the case in sentence repetition where semantic knowledge appeared to be critical (see Table 7.5). Case 3's attempts at sentence repetition in the "unknown words" condition resulted in numerous errors of phonemic transposition and omission. For example, the sentence The flag was coloured bright red (in which flag was the unknown word) was reproduced as The blag was fullered with a right breg. Performance on the sentence repetition task was also scored in terms of the individual words that were reproduced correctly for each serial position of the sentence target in the known and unknown words conditions (without regard to the actual serial position of the item in the patient's response). There was a massive serial position gradient in the unknown words condition (see Figure 7.1). Case 3 scored perfectly on the initial item but showed an increasing and progressive decrement in his scores over the next five or six positions. There was no serial position effect on the "known" vocabulary sentences. Comparable data for the two spanimpaired cases are also included to show that the pattern of Case 3's performance is not attributable to any nonspecific artefact of task difficulty.

174

McCarthy and Warrington 100 90 80

I7600 « 50

f

£ 40 30 20 10 1 2

3 4 5 Serial position

6

7

Figure 7.1. Repetition of sentences containing a "known" and "unknown" vocabulary. Percentage of words correct at each serial position for six- and seven-word sentences. A (Case 1); D (Case 2); • (Case 3); known (—); unknown ( ). Comment

This experiment replicated previous observations that there is no effect of a "known" versus "unknown" vocabulary in a list repetition task but that there is an effect of lexicality (Warrington, 1975). However, there was a dramatic effect of a single word drawn from an unkown vocabulary on the sentence repetition task. These findings not only substantiated the findings from the previous experiments in showing a dissociation between list and sentence repetition tasks; they are also telling with respect to the critical processes implicated in sentence repetition. First, the effect of vocabulary type indicates that adequate verbal-semantic knowledge is critical for sentence repetition. Second, the effects of serial position effectively rule out any major residual contribution from syntax in the absence of lexical-semantic knowledge.

7.2.5. Experiment 4 In this experiment we evaluated the patients' ability both to repeat a list of words and also to comprehend the core concept conveyed by the list. Method Fifty triplets of words were constructed such that they formed an abbreviated "naming from description" task for which there was a high-probability response that could be used to establish comprehension (e.g., small yellow bird). In one condition the patients were required to repeat the word triplets; in the other they were asked to name, or


175

Table 7.6. List repetition and list comprehension (% correct; n = 48)

Target response Repetition

Case 1

Case 2

Case 3

76 26

91 37

26 68

indicate by gesture, that they had comprehended the implied question. The same triplets were tested in both conditions using an ABBA design (50 stimuli were presented to Cases 1 and 3; a subset of 43 was presented to Case 2). A more detailed account of this experiment is given in McCarthy and Warrington (1987a).

Results The percentage correct for each patient in the two conditions is given in Table 7.6 There was a highly significant effect of conditions for each patient. The effect was, however, in the opposite direction for the span-impaired and the span-preserved cases. To consider the data from Case 3: There was evidence of satisfactory list repetition, but not surprisingly his poor comprehension gave rise to many failures in the naming task; although his commonest response was "I don't know," other errors were clearly failures to comprehend part of the word triplet (e.g., bull fighting country -» war; number legs duck->four). In contrast, the span-impaired cases gave prompt and appropriate answers in the naming conditions; they tended, however, to incorporate a naming response into their attempt to repeat the triplet (e.g., colour summer sky -* summer blue sky; number legs horse - • number horse four). In other instances the triplet was incomplete or misordered.

Comment The most striking finding in this experiment was the difficulty Cases 1 and 2 had in repeating highly meaningful triplets that conveyed an implicit question; this contrasted with their ability to repeat longer, and propositionally more complex, sentences. We have suggested that their performance on the repetition condition was strikingly poor; indeed, it was almost at the level of random word triplets. In contrast, their ability to comprehend these same word triplets was good, and far better than would be expected from their performance on other list or span tasks. Thus although they were unable to retain or repeat three-word lists, they were able to understand the message conveyed by a three-word list. We have argued that this experiment provides further evidence for a dynamic and integrative memory system that is in some sense independent of that employed in list retention (See also section 7.3.1).

176


7.2.6. Experiment 5 It has previously been argued that span impairment results in a major difficulty in the processing of sentences in which the subject and object roles of the constituents are indicated by word order (Saffran & Marin, 1975; Vallar & Baddeley, 1984). In this experiment we investigated the patients ability to comprehend sentences in which word order is critical.

Method The material was an expanded version of that described by Schwartz, Saffran, and Marin (1980) (kindly made available to us by E. Saffran). The task requires the subject to match a sentence spoken by the examiner to one of a pair of pictures: Each picture of a pair shows the same two characters engaged in a plausibly reversible transitive action. The sentences and pictures are constructed so that pragmatic biases are minimized. For example: 1. The dancer applauds the clown. 2. The clown is applauded by the dancer. The test uses an equal number of active and passive constructions, and comprises 78 sentence—picture items. (As an aside it should be noted that Case 3 was able to indicate knowledge of the critical items from this simple high-frequency vocabulary by matching the spoken word to the appropriate target in the test pictures.)

Results

Case 1 scored 65/78 on this task (five errors on the active voice sentences and eight on the passive). Case 2 scored 73/78 correct; Case 3 scored 72/78.

Comment Case l's performance, though somewhat weaker than normal, was creditable, considering his very severe impairment in retaining word or digit lists. More impressive was the performance of Cases 2 and 3, whose error rate was minimal. Overall, these findings suggested that the resources required for span tasks were not necessarily required for the comprehension of sentences in which word order is critical. We also established that, within the constraints of the vocabulary used in this test, Case 3 was capable of processing a range of syntactic constructions. His difficulty with other sentenceprocessing tasks could not therefore be attributed to "agrammatic" difficulties. (In this context it is of note that agrammatic patients show marked impairments on this test [Schwartz et al., 1980; McCarthy & Warrington, 1985].)


177

7.2.7. Experiment 6 The previous experiments have established that the span-impaired patients could comprehend verbal descriptive phrases and order-dependent sentences that exceeded their span of apprehension for lists by a considerable degree. The results of our experiments on their sentence repetition skills suggested that their language comprehension deficits might become evident in tasks in which sentences were contextually pragmatically implausible. The experiment used a technique adapted from Huttenlocher and her colleagues (e.g., Huttenlocher, Eisenberg, & Strauss, 1968). This technique enables one to vary the "realworld" salience of items that are referred to by the subject and object constituents in a prepositional phrase. Thus in the phrase a is above b, a is the logical subject of the statement (the thing that is being above) and b is the logical object (the thing that a is being above). In processing such sentences there are extralinguistic assumptions as to what the appropriate subject and object of the sentence should be. Given two movable tokens, and asked to place one above the other, children and adults tend to initiate action using the sentence subject. If one of the tokens is fixed in position, the sentence can be comprehended easily if the fixed token is the object of the sentence. If, however, the fixed token is referred to in subject position, then the sentence becomes less easy to comprehend. Huttenlocher et al. suggested that comprehension in this situation required an additional mental transformation of the examiner's statement so that the mobile token became the logical actor referred to by the statement. Thus, given a mobile green block and a fixed red block and the statement The red block is on top of the green block, some children said, "Oh, you mean the green block goes under the red one."

Method The procedure was a modified version of the task described by Huttenlocher et al. (1968). Two sets of locative expressions were investigated: (a) above and below and (b) in front of and behind. The subject was asked to place small models of a man (wearing black) and a woman (wearing red) on the desk in accordance with a spoken instruction. For the first condition a three-step "staircase" was constructed on which the models could be placed. Three arrangements of the arrays were used; in one both models were movable, and in the other arrangement either the model man or the model woman was fixed in position. In the concrete condition the patients were asked to "place the man/woman above or below the woman/man"; all permutations were tested equally often. Similarly, when testing in front of and behind all permutations of the instruction to "place the man/woman in front of or behind the woman/man" were tested in both the movable and the fixed conditions. For the more abstract, colour name conditions, the terms black and red were substituted for man and woman. Movable and fixed conditions

178

McCarthy and Warrington Table 7.7. Huttenlocher comprehension task (% correct; n = 64)

Unconventional: subject fixed Conventional: object fixed Control: neither fixed

Case 1

Case 2

53

59

84

78

89

83

were ordered in an ABBA design using 8 trials per block, and locative types were presented in blocks of 32 trials. Abstract and concrete conditions were alternated. An abbreviated 16-trial version of this task was given to case 3. (For further details see McCarthy and Warrington, 1987b.)

Results

There was no effect of "abstractness," but there was a significant effect of "conventionality" of the form of reference. The percentage correct for each condition is shown in Table 7.7. Both Cases 1 and 2 were consistently poor on the condition in which conventional conversational usage was violated; indeed, their scores were not reliably above chance on this condition. In contrast, when conventional subject-object usage was maintained there was evidence of satisfactory ability to process the same order-dependent sentences. Case 3 performed the abbreviated version of this task effortlessly.

Comment

Although both of the span-impaired patients were capable of comprehending reversible locative expressions, their comprehension was affected by the situation in which it must be demonstrated. When the conventional form of reference to subject and object in the prepositional expression was not adhered to, the patients were impaired. Huttenlocher has argued that under such conditions an additional cognitive operation is necessary for successful comprehension, that difficulty in performing this task is not due to the demands of linguistic processing per se, but rather to the requirement to perform additional restructuring operations on such linguistic information. 7.2.8. Experiment 7 In this experiment we explored the hypothesis that the span-impaired patients' deficit in sentence comprehension was independent of any form of word order processing


179

difficulty, but rather that it was attributable to the requirement to utilize spoken information in additional constructive and reconstructive operations. An adequate test of such a hypothesis requires an elementary task that cannot be solved directly on the basis of a propositional representation constructed on-line, and for which a veridical representation of the spoken information is therefore required. Simple comparative judgments would appear to meet these requirements in that they place a significant load on the construction and reconstruction of novel cognitive representations, and a minimal load on other aspects of processing. This applies whether the comparison is based on a relatively intrinsic feature such as colour, or a more arbitrary extrinsic feature such as size. Method A set of five pairs of words referring to items varying in colour and size were assembled (e.g., poppy, lettuce). These were placed in question frames, requiring a comparison in terms of either colour or size. The same item pairs were used in both types of comparison. The questions were framed so that maintenance of noun order was irrelevant (e.g., Which is red, a poppy or a lettuce?). The order of items was factorially combined so that the same items appeared in the initial position of half the sentences and in the final position for the remainder. This yielded a 2(colour) x 2(size) x 2(order) set of comparisons for each pair of items, and 40 trials for the block. A second set of items was constructed from five pairs of animal names that could be contrasted either in terms of the intrinsic attribute dangerous/tame or in terms of size. As in the first part of the experiment the names were placed in a non-order-dependent sentence frame for the comparison task (e.g., Which is more dangerous, a scorpion or a Iambi Which is larger, a scorpion or a Iambi). The same item pairs were used in each comparison type. The factorial combination of dangerous/tame by size and by order gave a 40-trial set of comparisons for this block also. In both blocks items in the more "arbitrary" size comparison were selected on the basis of pilot investigations so that they were not in some sense "marked" as large or small (e.g., elephant, mouse) but rather so that they were (as far as possible) unambiguously larger or smaller than the other item in the comparison pair. The comparison questions were presented in two sets of 20 trials per block and the colour-size comparison items were presented before the tameness-size comparisons. The extrinsic and intrinsic comparison types were randomized within blocks. (For further details see McCarthy and Warrington, 1987b.) Results Both span-impaired patients showed marked impairments in performing the intuitively simple intrinsic comparison task (Case 3 was tested on a version of this task tailored to

180

McCarthy and Warrington Table 7.S. Comparative judgments (% correct; n = 10)

Intrinsic attributes Arbitrary attributes

Case 1

Case 2

75 73

53 44

his "known" vocabulary and his performance was without error). There was no significant difference between the patients' ability to perform intrinsic or extrinsic comparisons. The percentage correct for each patient for the intrinsic and extrinsic conditions (summing across comparison types) is shown in Table 7.8. Qualitatively it was observed that the patients were slow and laboured in their performance, and they claimed that they had marked difficulty in retaining the comparison constituents. A number of their errors were intrusions from previous comparison questions within the same block of trials (e.g., Which is green, a fox or a gooseberry! -> poppy). Case 2 attempted a variety of strategies for coping with the questions, including using his hands as a means of retaining the comparative adjective.

Comment The findings from this experiment supported the hypothesis that the patients' difficulties were attributable to the requirement to use linguistic information in a supplementary set of cognitive operations. They could not in any way be attributed to an impairment in the processing of syntactic information, since this was of limited relevance in this task. Indeed, for Case 2 there appeared to be a clear dissociation between his good ability to process order-dependent sentences and his inability to perform a cognitive operation on the basis of nonredundant spoken information. There was also no question that this difficulty with comparative judgments reflected a failure to comprehend the individual word, since Case 1 had an average, and Case 2 a superior, receptive vocabulary (Peabody Picture Vocabulary).

7.3. Discussion This series of experiments documented the repetition and comprehension skills of two patients with an impaired span for lists of words but normal single-word comprehension and one case whose span was preserved but whose comprehension of the single word was gravely impaired. In this discussion of the data we will consider two main issues. First, the significance of a dissociation between the ability to repeat lists and the ability to repeat sentences, and second, the functional basis of the language comprehension deficits observed in the span-impaired cases. To anticipate, we shall argue:


181

1. That there are multiple short-term representations of spoken input. The span-impaired patients have relative preservation of an on-line active and dynamic memory system that mediates sentence repetition and can sustain the comprehension of (at least some) orderdependent constructions. Auditory-verbal span and many aspects of sentence processing therefore dissociate. 2. For a specific functional role for "span" in normal information processing. The short-term representation measured by list span tasks is required for backtracking operations when comprehension necessitates more than a superficial linguistic or propositional analysis of the input.

The paradox that the span-impaired cases were able to repeat some sentence material at a better level than the span-preserved case was a surprising and hitherto undocumented pattern of results. Consider first the findings from the span-impaired subjects: Despite having reliable digit spans of only one item, they were nevertheless frequently capable of repeating relatively long sentences absolutely verbatim. At the same time they were impaired in repeating lists of key content words taken from the same sentences. On meaningful three-word lists that contained an implicit question (e.g., small yellow bird), their performance deteriorated to that observed for random lists; this deficit did not reflect a failure in comprehension since they had no difficulties in answering the question. If the verbal stimulus formed an incomplete sentence they had more difficulty with verbatim recall than with repetition of a whole sentence. In experiments 2 and 4 it was noted that they had marked impairment in suppressing the final word or the "answer" in their recall. In this respect they appear to be analogous to normal subjects when performing the Stroop task - namely, the most automatic or available response was difficult to suppress. Turning now to Case 3, his performance on repetition tasks contrasted directly with that of Cases 1 and 2. On span tasks for digits and words his performance was at the lower limits of normal. He did not, however, show the normal pattern of "gain" for sentence context that was evident in the span-impaired cases. If anything, there was a negative, rather than a positive, effect of sentence context in that he was able to repeat lists of words more adequately than sentences containing the same vocabulary. In sentences containing one word that he did not comprehend, his performance deteriorated markedly. Furthermore, he showed a serial position effect in repeating those sentences that contained at least one unknown word (Figure 7.1), tending to omit the final words from the target. By contrast with the span-impaired cases, there was no decrement in his sentence repetition when the final word was omitted.

7.3.1. Multiple short-term memory systems? The findings on repetition tasks indicated that there are dissociable processes implicated in the retention of lists and the retention of sentences. We have gone somewhat further and argued that there are dissociable short-term memory systems (for related accounts see Monsell, 1984, Barnard, 1985). We suggested that the system the span-impaired

182


cases were using for sentence repetition has all the characteristics of a dynamic and integrative language processor operating in an anticipatory mode. It would seem that this is a system that operates at a level of processing complexity beyond that of the single word. We have argued that we could also account for certain findings from the span-preserved case in terms of an inability to use this system effectively. More specifically our suggestion was that his lexical comprehension deficit underpins his difficulty with sentence repetition. Clearly a system that operates in a dynamic, and, more importantly, an anticipatory, mode must have a high degree of lexical-semantic knowledge available to it. If such knowledge is degraded, then it would be predicted that the operation of such a system would be severely compromised. In the extreme case, the patients' level of performance would be expected to be similar to that for random word lists. For our span-preserved case this interaction was most clearly demonstrated in the experiment that compared his performance on repeating sentences containing a single item drawn from a "known" or an "unknown' word vocabulary. Those sentences containing at least one unknown word were particularly bad. In the case of the ''unknown" word sentences we have interpreted his repetition as being based on those procedures that underpin list repetition. However, his auditory-verbal list span was insufficient to cope with these longer strings of words. This interpretation was corroborated by qualitative evidence from the patients' errors in all sentence repetition tasks. The span-impaired cases made minor semantic paraphrases when repeating sentences, indicating that they were processing them for meaning (e.g., Case 1 repeated His work was his joy as Ms work or his job was his joy). By contrast, the errors made by the span-preserved case were phonological transpositions such that many of his utterances were neologistic (e.g., he repeated She takes lemon in her tea as She lakes temmon in her tea).

We have argued that there is a dynamic, integrative memory system that underpins sentence repetition which is preserved in the span-impaired patients. If this were the case, then there clearly must be an alternative system that mediates word list repetition. This system was largely preserved in Case 3. It appears to be based on phonological information and to be sensitive to lexicality. It may be independent of semantic knowledge, since in a list repetition condition it did not matter whether the words were "known" or "unknown" to the patient.

7.3.2. Function of auditory—verbal "span" What then is the normal functional role of those processes measured by list retention tasks? We have already suggested that the dynamic, anticipatory processing that was intact in the span-impaired cases had many of the properties required for comprehension of normal conversation or running speech. However, this level of processing


183

appeared to be insufficient on its own for modulating language processing in the face of anomalous or low-probability spoken input. Cases 1 and 2 appeared to be unable to correct or modify the initial set of interpretive hypotheses on which their recall of input was based. It appeared that they were specifically impaired in their ability to employ those systems that are required for the monitoring and control of language processing when procedures based on anticipatory hypotheses are inadequate: They were unable to modulate and control the operation of a dynamic and forward-looking language processor. Specifically, we suggested that the processing system that subserves list retention has the function of a backup resource for noncontemporaneous or "off-line" language processing. The question arises as to when such a backup resource is required. To consider the empirical data: Despite having reliable digit and word spans of only one item, there was evidence that both Case 1 and Case 2 were capable of comprehending a variety of sentence types, and in particular there was no evidence of specific comprehension difficulties on sentences whose interpretation was dependent on order-dependent processing. Thus they were able to perform satisfactorily on a test of sentence comprehension in which noun phrases were plausibly reversible around either an active or passive verb phrase (Experiment 4). Similarly, good levels of comprehension were documented using reversible locative constructions (e.g., above, below). However, the patients had specific difficulties when normal conversational conventions were contravened and when the verbal information had to be used in a subsequent set of cognitive operations (Experiments 6 and 7). We have argued that these findings are consistent with the view that Cases 1 and 2 have relatively intact on-line language comprehension skills, but that they are impaired in those operations that require backtracking over spoken input. To elaborate on the conditions in which such backtracking operations are likely to be necessary: In essence, we have suggested that backtracking procedures are required when an appropriate central cognitive representation cannot be constructed on-line or contemporaneously with spoken linguistic input. By the term central cognitive representation we refer to a level of processing that is not only based on analysis of the spoken utterance but that also incorporates those aspects of real-world knowledge and expectancies that underpin understanding. It can be considered as a level of representation in which there is an interaction between a spoken phrase and other salient sources of evidence such as preceding or anticipated speech, events, or other contextual information (see e.g., Webber, 1981; Garnham, 1985; JohnsonLaird, 1983). Backtracking resources are likely to be required under those circumstances in which the results of processing auditory-verbal information cannot be immediately transcoded into such central cognitive representations (McCarthy & Warrington 1987a, b).

184


We have suggested that the following three conditions are likely to impose constraints on such transcoding operations and therefore on the requirements to resort to backtracking operations: 1. When the rate of information presentation is too great for the development of a sufficiently unambiguous central cognitive representation. 2. When extralinguistic assumptions bias the interpretation of the spoken message. 3. When the achievement of an adequate central cognitive representation requires supplementary cognitive operations to be performed on the spoken input. These three conditions are unlikely to be mutually independent, or even exhaustive. They are merely highlighted in this context as characteristic of situations in which backtracking operations appear to be necessary. They are required so that a spoken message can be understood at a level that goes beyond linguistic analysis. Thus a central cognitive representation can provide a framework that subserves action and reaction to an utterance. Let us consider these three conditions in the light of the evidence from the spanimpaired cases. First, performance on tasks such as the Token Test requires the transcoding of low-redundancy information into an adequate central representation and so in the present terms would place a considerable load on backtracking operations. The span-impaired cases performed very poorly on this task. Second, the importance of extralinguistic assumptions was clear in the pattern of performance on the Huttenlocher et al. (1968) task. The span-impaired cases were able to perform the task when conventional forms of reference were used, but were impaired in the pragmatically anomalous condition. Third, the requirement to perform additional cognitive operations on the spoken message may lead to impairment as was evident in the comparative judgment task. In all three conditions precise verbatim information is required in order to construct a central representation that is adequate for these particular tasks. Backtracking procedures provide a means by which a verbatim record can be reanalysed in the absence of continued external auditory input. In summary, we have argued that the on-line language-processing system preserved in these patients is adequate for them to perform a range of linguistic operations, including syntactic processing. When additional operations are necessary in order to transcode between the auditory-verbal message and a developing central cognitive representation, then a backtracking facility may be utilized that has recourse to a verbatim record that can be "replayed." The processes involved in understanding a sentence so that it can be used in action, rather than in comprehending it as a linguistic structure, appear to reflect an interaction between linguistic and nonlinguistic or pragmatic levels of representation. As with other cognitive skills, sentence comprehension would appear to be a complex and multifaced process, the various subcomponents of which can potentially break down highly selectively in a variety of aphasic syndromes.

A phenomenon in search, of a function?

185

References Baddeley, A. D. (1976). The psychology of memory. New York: Basic Books. Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short term memory. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). London: Erlbaum. Bloom, P. A., & Fischler, I. (1980). Completion norms for 329 sentence contexts. Memory and Cognition, 8, 631-642. Caramazza, A., & Zurif, E. B. (1976). Dissociation of alogrithmic and heuristic processes in language comprehension: Evidence from aphasia. Brain and Language, 3, 572-5S2. Coughlan, A. K., & Warrington, E. K. (1978). Word comprehension and word retrieval in patients with localised cerebral lesions. Brain, 101, 163-185. De Renzi, E., & Faglioni, P. (1978). Normative data and screening power of a shortened version of the Token Test. Cortex, 14, 41-49. Ellis, A., & Beattie, G. (1986). The psychology of language and communication. London: Weidenfield & Nicholson. Frazier, L., & Fodor, J. D. (1978). The sausage machine: A new two stage parsing model. Cognition, 6, 291-325. Garnham, A. (1985), Psycholinguistics: Central topics. London: Methuen. Heilman, K. M , & Scholes, R. J. (1976). The nature of comprehension errors in Broca's, conduction and Wernicke's aphasics. Cortex, 12, 258-265. Hitch, G. (1980). Developing the concept of working memory. In G. Claxton (Ed.), Cognitive psychology: New directions (pp. 154-196). London: Routledge & Kegan Paul. Huttenlocher, J., Eisenberg, K., & Strauss, S. (1968). Comprehension: Relation between perceived actor and logical subject. Journal of Verbal Learning and Verbal Behavior, 7, 527-530. Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, inference and consciousness. Cambridge: Cambridge University Press. Lesser, R. (1974). Verbal comprehension in aphasia: An English version of three Italian tests. Cortex. 10, 247-263. McCarthy, R. A., & Warrington, E. K. (1984). A two route model of speech production: Evidence from aphasia. Brain, 107, 463-485. McCarthy, R. A., & Warrington, E. K. (1985). Category specificity in an agrammatic patient: The relative impairment of verb retrieval and comprehension. Neuropsychologia, 23, 709-727. McCarthy, R. A., & Warrington, E. K. (1987a). The double dissociation of short-term memory for lists and sentences: Evidence from aphasia. Brain, 110, 1545-1563. McCarthy, R. A., & Warrington, E. K. (1987b). Understanding: A function of short-term memory? Brain, 110, 1565-1578. Marcus, M. P. (1980). A theory of syntactic recognition for natural language. Cambridge, MA: MIT Press. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and Performance X (pp. 327-350). London: Erlbaum. Saffran, E., & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with a deficient auditory short-term memory. Brain and Language, 2, 420-433. Schwartz, M. F., Saffran, E., & Marin, O. S. M. (1980). The word order problem in agrammatism I: Comprehension. Brain and Language, 10, 249—262. Shallice, T., & Warrington, E. K. (1970). Independent functioning of the verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261-273. Snodgrass, J. G., & Vanderwart, M. (1980). A standardised set of 260 pictures: Norms for name agreement, familiarity and visual complexity. Journal of Experimental Psychology, Human Learning and Memory, 6, 174-215.

186 McCarthy and Warrington Vallar, G., & Baddeley, A. D. (1984). Phonological short-term store, phonological processing and sentence comprehension. Cognitive Neuropsychology, 1, 121—141. Warrington, E. K. (1975). The selective impairment of semantic memory. Quarterly Journal of Experimental Psychology, 27, 635-657. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory verbal short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176-189. Webber, B. L. (1981). Discourse model syntheses: Preliminaries to reference. In A. K. Koshi, B. L. Webber, & I. A. Sag (Eds.), Elements of Discourse Understanding. Cambridge: Cambridge University Press.

8. Short-term retention without short-term memory BRIAN BUTTERWORTH, TIM SHALLICE, AND FRANCES L. WATSON

8.1. Introduction Immediate memory capacity for lists is usually estimated at around 5—6 words, whereas for sentences, Miller and Selfridge (1950) found that 20-word sentences can be produced with nearly 100% recall, a finding replicated by Craik and Masani (1969). In a more recent study, Butterworth, Campbell and Howard (1986) presented subjects with 40 sentences 15-21 words long for immediate recall. Undergraduate subjects recalled about 25 of them perfectly. Most of the errors were omissions and word substitutions, not word order errors (less than 3% of all errors). There were significant serial position effects, with the first two words and the last word recalled better than the others. "Running memory span" tasks (where subjects are allowed to choose the segment to report) show accurate recall for segments up to and exceeding 20 words; in general, about 88% of the words are recalled (discounting order) irrespective of the segment length (Wingfield & Butterworth, 1984, Experiment 1). Why is immediate serial recall of sentences better than lists? A number of probable solutions will doubtless spring to mind. Sentences encourage chunking; the meaning of a sentence can be stored in long-term memory (LTM) and thence retrieved in recall; sentences are meaningful; sentences have structure; sentence materials utilize quite different memorial systems; and so on. Various authors have made suggestions along one or other of these lines (e.g., Miller & Selfridge, 1950; Craik, 1971; Shallice, 1979; Butterworth et al., 1986; McCarthy & Warrington, 1987a). It has frequently been argued that a verbatim record of sentence input is held in some limited-capacity store while the construction of a higher-level representation is carried out. If you test well beyond the end of the clause, gist may be retained, but the exact

We are grateful to David Howard for handcrafting a Jonckheere Test of Trend for us, and to Paul Burgess for his help in analysing JB's data. Participants at the meeting made many useful comments, and we would especially like to thank Graham Hitch for discreetly saving us from a serious error. We would also like to thank Elizabeth Warrington for providing facilities that enabled the research to be carried out.

187

188

Butterworth, Shallice, and Watson

words and their order are likely to be lost (Sachs, 1967; Johnson-Laird & Stevenson, 1970). The short-term memory patients provide good evidence that for list retention, a limited-capacity short-term store (STS) is indeed used (see Shallice & Vallar, this volume, chapter 1). The inability to reproduce the full surface representation of a sentence when gist can be retrieved suggests, if less firmly, that the same system is also used in sentence repetition. There are three obvious questions that may be asked about systems of this type. First, what are the contents of the limited-capacity store? Second, what is its capacity? Third, is retrieval from the store independent of retrieval from other systems? Our primary concern in this chapter is the third of these questions, but an answer to the second is used as an intermediate step. Answers to both these questions, however, are clearly dependent on the answer given to the first.

8.2. The contents of the input store If we assume some kind of speech input store of limited capacity or duration, what does it contain? Short-term memory experiments point strongly to a single store's being primarily responsible for span (see Shallice & Vallar, this volume) and for its being phonological in nature (e.g., Conrad, 1964; Baddeley, 1966; Craik, 1968a; Kintsch & Buschke, 1969; Shallice, 1975; Campbell & Dodd, 1984). As we have noted, verbatim memory is typically short-lived, although some information about meaning lasts longer; nevertheless, semantic errors are found even in immediate recall for sentences. This has led Clark and Clark to suggest that "both verbatim wording and semantic interpretations are retained in short-term memory" (1977, p. 141). Similar arguments were made in the memory literature (e.g., Shulman, 1970), but there is no good evidence that the semantic effects did not arise from a longer-duration memory system (Shallice, 1975). The idea that some intermediate workings of the comprehension system use the same limited-capacity STS has appealed to a number of other investigators. Savin and Perchonock (1965) hypothesized that syntactic features, like PASSIVE, NEGATIVE, QUESTION, EMPHATIC, WH-, are "encoded separately and therefore each occupies a characteristic amount of space in immediate memory The additional space can be measured by seeing how much additional material can be remembered along with the sentence" (pp. 349-350). Their task required subjects to remember a sentence plus a list of unrelated words for immediate recall. Immediate memory will, according to their view, contain three kinds of thing: a "kernel sentence" (presumably in words, e.g., The boy hit the ball), some abstract grammatical features (like QUESTION, NEGATIVE), plus some additional and unrelated words (e.g., tree, cow, bus, hour, [chair, rain, hat, red]. So, in immediate recall, the subject would say "Hasn't the boy hit the ball? Tree, cow, bus, hour," with the remaining words unavailable through lack of space. And indeed the


189

results of their experiment are impressive support for the hypothesis. Fewer words are recalled following complex sentences, even those that are the same length as the simplest kernel, for example, Who has hit the balll and Has the boy hit the balll The more

grammatical features to be held, the fewer additional words are reproduced. However, the word list recall differences could be simply explained through difficulties of retrieving and producing the various sentence forms, producing differential Brown-Peterson interference on the word lists, rather than through the transformation and word list occupying a common system (see Boakes & Lodwick, 1971). Moreover, it would appear on this view that an STM patient should never be able to reproduce a sentence containing more content words than he can repeat as a list (but see McCarthy & Warrington, 1987a, and this volume, chapter 7). There are also good computational grounds for finding Savin and Perchonock's view implausible. Computational parsing systems have to be explicit as to the contents of temporary stores at each time step in the parse. Recent parsing models typically operate with two working memories - a "stack" that holds the intermediate parsing results, and a register that holds the word string to be parsed. In "shift-reduce" parsers (e.g., Pereira, 1985), the next word in the register is "shifted" to the top of the stack and then "reduced" by assigning it a grammatical category by means of grammatical rules held in LTM. Complex w-tuples of categories can be further reduced by rule (e.g., Det N - • NP, NP VP —• S). Since there is usually more than one next move — a shift or a reduce, or two or more possible reductions — another temporary memory is needed to store the outcomes of lookahead, or other top—down procedures, for resolving these conflicts. From a programming point of view, the three stores are independent, with material transferred with strict directionality among them. Why should not this be the case for human processing? We will therefore assume that one possible source of information that can be used in the immediate recall of sentences is a phonological buffer. What capacity would such a system have? The standard position of early STM theorists was based on the recency effect in free recall. Thus, as Craik (1971) claimed in his pre-levels-of-processing days, "Most workers now accept that PM [primary memory] can hold between 2.5 and 3.5 words. This estimate may strike readers as low in view of the fact that word span for the same type of subject is 5-6 words. My conclusion is that the traditional span measure of STM includes a SM [secondary memory] component" (p. 223). One argument he gave for this conclusion was that "semantic and associative factors apparently play no part in PM ... while patently such factors can affect span - sentence span, for example, is around 20 words for student subjects" (p. 223). This implies that when a 20-word sentence is recalled verbatim, only 2 or 3 words are coming from a phonological buffer. A very different position was developed in early psycholinguistic studies. Jarvella (1971) found that for multiclause sentences there is evidence that the last clause is better recalled than previous clauses, indicating early syntactic processing of heard material.

190


One implication of this position is that complex material, or clauses that require further clauses for a full interpretation, should be retained longer because work on them takes longer; and there is some evidence that this is the case. Thus sentences with SUBORDINATE-MAIN clause structure are recalled more accurately than those with MAIN-SUBORDINATE structure because the subordinate clause must be held until the main is heard for full interpretation (Jarvella & Herman, 1972); and more generally, there is an effect of clause "completeness" on a variety of tasks, which points to holding the low completeness clauses longer (see Flores d'Arcais & Schreuder, 1983, for a review of relevant studies). This suggests that the store holds at least a clause. How are these two estimates to be reconciled? There would appear to be three possibilities. One possibility that has been suggested by Butterworth et al. (1986) and McCarthy and Warrington (1987a) is that different systems may be involved for lists and for sentences. Butterworth et al. (1986) report a subject, RE, whose performance on three-digit lists was only 80%, and for four-digit lists, 40%, yet her processing of sentences was unaffected. So, for example, she was able to score 100% for spoken repetition of the Token Test (matched controls, 97-98%). A second possibility is that psycholinguistic studies overestimate the phonological level capacity. Thus, the extra work done on subordinate clause/main clauses in the Jarvella and Herman experiment could lead to their being semantically better coded. The opposite possibility, explored by Shallice (1975, 1979), is that free recall and even list span tasks seriously underestimate the phonological buffer capacity because they are tasks that do not make effective use of its speech-specific characteristics. We will return to this issue in the Discussion. The final issue concerns the relation between retrieval from the phonological buffer and from more long lasting stores. The classic position in memory is that retrieval from different-level stores is independent. This assumption allows findings from many experimental paradigms to be explained from the same model (Waugh & Norman, 1965; Craik, 1968b). (Levy & Craik, 1975, provide a formulation in terms of codes that is essentially equivalent.) Interactive models widely applied to perceptual analysis have also been suggested to account for short-term memory phenomena (e.g., McClelland & Elman, 1986). If there is an interactive relation between representations in different stores, then it would be predicted that the presence of subthreshold information in one of the stores would facilitate the retrieval of subthreshold material in the other. Yet in the few experiments that have looked at this issue in list memory situations (e.g., Craik & Levy, 1970), the only deviations from independence reported have been failures of an additional source of information to provide any advantage at all in immediate recall (Smith, Barresi, & Gross, 1971). However, none of these studies has used sentential material. The present study tackles this issue directly. We use a procedure that derives from Savin and Perchonock (1965) to test whether the capacity of a short-term memory buffer used in list retention is the same as that used in sentence retention. However,


191

instead of using their technique of presenting a sentence followed by a list of unrelated words, sentences containing lists are employed. This overcomes the problem alluded to earlier, namely, that the recall of the sentence may affect the recall of the following list. It further permits a direct comparison of recall of list elements presented separately with recall of equivalent elements embedded in a sentence, thereby enabling us to assess more straightforwardly the contribution of sentence-forming elements to memory. Our concern in this study will be to assess the relative contributions of a phonological store, A, and a higher-level store, B, that holds syntactic and semantic information, and to determine whether these contributions are independent. To get the argument off the ground, three assumptions are made: Assumption I. Span for sentences uses information from both stores. Assumption II. The contribution of B is the same whether retrieval is immediate or if retrieval occurs after a period of filled delay (FD), say 20 sec; this implies that B is not a short-term store. Assumption III. After a 20-sec FD, the contribution of A is zero. Our basic evidence is derived from the study of normal subjects. However, we test our assumptions with a well-known STM patient, JB (Warrington, Logue, & Pratt, 1971; Shallice & Butterworth, 1977). We then assess the model we produce on further findings from the patient and from the normal subjects.

8.3. Immediate memory for lists and sentences: the basic argument In our first set of experiments, we derive data that allow us to assess the contribution of A and B to recall. We estimate the contribution of B first, from the performance of JB, a patient with a known deficit on STM tasks, whom we assume had very limited A capacity; and second, from the performance of normal subjects, as well as JB, on recall following a filled delay, which we assume prevents rehearsal (Brown, 1958; Peterson & Peterson, 1959). We can then estimate the contribution of A by subtracting estimates of B from total recall; further, comparisons of list recall, where higher-level representations will be minimal, with recall for sentences containing listlike elements allow us to assess whether the contributions of A and B are independent.

Subject

JB, a secretary born in 1935, had a meningioma removed from the neighbourhood of the angular gyrus in the left hemisphere at the age of 23. She was initially aphasic but her language functions recovered very satisfactorily, except for a dense STM impairment that has been extensively studied (see Warrington et al., 1971, and Shallice & Butterworth, 1977, for basic data; see also Shallice & Warrington, 1977; Allport, 1984). Her spontaneous speech is normal in fluency rate and sentence structure; apart from a

192


small increase in function word errors, her rate of production of a number of types of speech errors (including phonemic and verbal paraphasias) was well inside the normal range (Shallice & Butterworth, 1977).

Control subjects

Two groups of normal subjects were women from the Applied Psychology Unit subject panel, roughly matched for age with JB.

8.3.1. Experiment 1. Lists and sentences, with and without filled delay: the basic phenomena This was a serial recall task using two types of materials - lists and sentences containing list items. We tested JB at the limit of her span on four-item lists and on sentences containing four listlike items. She performed the test twice, using complementary versions of the sentences in the two sessions. We compared her performance with that of 6 normal subjects. A second study was used to estimate the limits of normal retention, using 12 new subjects on six-item lists and on sentences containing six listlike items; sentence structure was otherwise identical. In both studies we compared immediate serial recall with recall after a 20-sec delay occupied by forced speeded addition by 1 from numbers presented on cards. The rate used was titrated to be just slower than that at which the subject breaks down.

Materials Lists were composed of four (or six) words in the same semantic category, to eliminate the possibility that the sentence advantage consisted solely in providing cues to the type of material to be recalled. For example, 1. tablecloth towel curtain duster bull donkey duck rabbit EXNK Sentences were composed of similar items to the lists, but with additional words to turn them into sentences. Two basic types of construction were used. In the first, all the list items were dominated by a single phrase (2); in the second, the list items were divided into two phrases (3). (The effect of this manipulation is discussed in section 83.5.) In every sentence, as well as function words, there were either two or three additional content words. For example, 2. The removal firm took [a bed, a cabinet, a wardrobe, and a chair]. The new visitors were called [Patrick, Anthony, Mark, and Jean]. 3. Wash [the sheet and the bedspread] at the same time as [the pillowcase and the napkins]. In the alphabet, [G and L] come before [T and P].

Short-term retention without short-term

memory

193

Table 8.1. Lists and sentences recalled by JB and six control subjects. Mean percent completely correct {and range, in parentheses) Filled delay

Immediate

Sentences Lists

JB

Controls

JB

Controls

0 5

85(60-100) 100

2.5 5

7(0-20) 10(0-20)

Note: JB's results are averaged from two sessions with the same materials.

Table 8.2. Recall of four-item lists and sentences by ]B and controls, immediately and following 20-sec FD

FD

Immediate

JB Lists Mean list items correct (max = 4)

3.0

Sentences Mean list items 2.65 correct (max = 4) Mean percent of 79 other content words correct Mean percent of 98 function words correct: List linked3 Other 3 91

Control

JB

Control

4.0

2.5

2.67(1.7-3.6)

3.88

2.35

2.67(2.0-3.2)

100

77

83

(75-92)

100

95

94

(91-100)

100

89

78

(67-100)

3

List-linked function words are those directly associated with list items - e.g., determiners and conjunctions. Maxima are calculated according to the mean number of list items attempted by subjects; proportions correct are derived from these figures.

Results

We present first data from both normals and JB on four-item lists and sentences with immediate recall and after a 20-sec FD. Overall performance is given in Table 8.1. Taking just the proportion of strings completely correct (including order), controls were dramatically superior on immediate as opposed to delayed recall, whereas JB was near or at floor in both conditions. The sentence performance was further analysed in terms of (a) the list items recalled, (b) the other content words recalled, (c) function words recalled, and (d) order errors. The results are given in Table 8.2.

194


Table S3.

Recall of six-item lists and sentences by normal subjects, immediately and

following 20-sec of FD

FD

Immediate Mean no. items

Proportion of each item

Mean no. items

Proportion of each item

Lists

List items (max = 6)

5.16

.85

2.72

.45

4.5 2.21

.75 .94

3.07 1.83

.51 .78

3.33

.97

2.24

.95

1.67 11.71

.95

1.38 8.52

.79

Sentences

List items (max = 6) Other content words (max = 2.35) Function words List linked (max, imm = 3.42; FD = 2.37)a Other (max = 1.75) Total items a

List-linked function words are those directly associated with list items - e.g., determiners and conjunctions. Maxima are calculated according to the mean number of list items attempted by subjects; proportions correct are derived from these figures.

It is clear from Table 8.2 that JB's performance is unaffected by filled delay, even though she reported that the task was demanding for her, since she was unable to rehearse during the filled interval. The performance of controls, on the other hand, suffers a considerable decline. In fact, after FD, the controls are equivalent ko JB on both lists and sentences, and JB falls within the normal range for most measures. In other words, on this task FD turns normal subjects into STM patients. In a further study, 12 control subjects carried out the same task but with 10 six-word lists, and 10 sentences containing six list-type words. The results are summarized in Table S3 Again it is clear that preventing rehearsal has a dramatic effect. On all measures, all subjects perform worse after FD. In fact, no six-item lists were reproduced accurately after FD, though with immediate recall, subjects managed on average 3 lists out of 10 completely correctly. The other striking feature about these data is that far more words are recalled from sentences than from lists. For six-item materials, controls recalled about 5 words from a list, and on average 11.7 words from sentences. JB also recalls more from sentences in both conditions. If we consider only content words (list plus other), the same effect occurs; in six-item strings, 5.16 are recalled immediately from lists, yet 6.71 are recalled from sentences; even after FD, 2.72 content words are recalled from lists, and 4.9 are


195

recalled from sentences. Does this advantage just reflect the independent contribution of higher-level syntactic and semantic stores, which can be used more effectively for sentential material?

8.3.2. Does sentence retention involve two independent stores? On the four-item sentences, the controls' immediate recall of list items is close to perfect (97% correct). However, this is far from the case in delayed recall, where 33>% of the list items cannot be retrieved. This means that retrieval from Store B is far from sufficient to support perfect performance. Moreover, this is true for all four serial positions, and especially when the list items are all in the same phrase. Here performance after FD ranges from 46% to 76% correct serial positions. Thus if the probability of retrieving an item from Store A is independent of the chance of obtaining it from Store B (and if the contents of A and B are not negatively correlated - see later in this section), then all four items must be available from A to give perfect recall. After FD, order errors frequently occur (40% of all lists), indicating that retrieval from B does not provide satisfactory information on order. In the immediate recall of sentences, order errors were very infrequent (5% of list items). This further supports the conclusion that all four list items are available to be retrieved from Store A in immediate recall (which is consistent with the claim that the phonological record contains order information). The other content words and nonlist-linked function words were also perfectly retrieved in immediate recall, but not in delayed recall, so they too must presumably be available for recall from Store A. This implies that at least eight words are being held in A, without taking into account list-linked function words. This is considerably greater than span for lists, which is less than six items. The conclusion is that retrieval from the two stores is not independent and subthreshold information in the two stores can summate in some way, or the capacity of A is greater for continuous speech. The findings for six-item lists and sentences are broadly similar. Only 0.66 fewer list items can be immediately retrieved in the sentence condition than in the list condition, indicating relatively little difference in the number of list items retrievable from A. Recall of other content words and the nonlist function words is at the 94-95% level in the immediate condition, compared with less than 80% after FD, so that most of these words must also be available in A in the immediate condition. We shall argue later that Assumptions I—III are valid. Thus either the assumption of independence or the assumption of equivalent capacity is at risk of rejection. There are other explanations that may preserve all five assumptions. The first is that the observed differences between lists and sentences are a simple artefact of the length of strings to be recalled. Now, given that sentences containing six list items are about 8 words longer than 6-word lists, then, if p(A) = 0.73 irrespective of

196


the number of items to be recalled, we could expect 0.73 x 8 = 5.84 extra words to be recalled. This is not too dissimilar from the observed value of 4.86, if p (item recall) remains at around the list value. However, the probability of recall from A for lists is not independent of the number of items to be recalled: When span is exceeded, the normal forgetting curve shows a decreasing probability of recall as the number of items increases. We would therefore expect p(A) to be lower where there are eight extra words. Indeed, it is standardly assumed that there is a drastic falloff in the probability of recall for lists greater than span, and it is a basic tenet of work in this area that capacity is limited. Evidence from other studies supports the idea that p(A) for lists declines sharply beyond span (Craik, 1968b). Thus, from our own two studies, we can see that for immediate recall p4 (item recall) = 1, whereas p6 (item recall) = 0.85. Second, it is logically possible that although retrieval of an item from A itself reduces the probability of the retrieval of other items from A, retrieval of an item from B does not reduce the probability of subsequent retrieval of other items from A. One can imagine a mechanism for A that would have this property — for example, where only the retrieval of an item from that store had the effect of reducing the activation or remaining items. In an arrangement of this kind, outputs from A and B would appear to interact, although in fact retrieval processes would be quite independent. However, such a supposition runs counter to findings of Craik and Levy (1970): In ordinary free recall, when unrelated words were followed by semantically related words, which were mainly retrieved from B, the unrelated words, were no better recalled than when followed by other unrelated words which were mainly retrieved from A.1 Finally, it is also logically possible that what is stored in A and B is negatively correlated. That is, the contents of A and B are complementary prior to retrieval: Gaps in A are filled by items in B, and vice versa. However, the wide range of sentence content for which retrieval from B is not perfect means that any such strategy would have to be highly "intelligent", with the use of A specifically tailored to the characteristics of the individual sentence. This makes the explanation both ad hoc and a priori implausible. Thus none of the ways by which the two final assumptions can be preserved seems plausible. If these arguments are accepted, and assuming that list and sentence processing do not use entirely separate systems, we are left with two main candidate explanations of the results. The first is that retrieval from the two stores is not independent with subthreshold information in the two stores summating in some way. The other is that the capacity of the short-term store is greater for continuous speech than for lists. These inferences, however, depend on our three assumptions. Assumption I - that both Stores A and B are involved in sentence span — is, we trust, entirely uncontroversial. Assumption II says that retrieval from B is constant over 20 sec, and Assumption III


197

holds that with filled delay the contribution of A to recall is zero after 20 sec. We can justify both of these by considering the performance of JB. She is normal on all measures after a 20-sec FD (see Table 8.4). That her performance is normal after a 20-sec FD implies both that JB is normal on B and that there is no contribution of A after a 20-sec FD. Moreover, in the sentence condition, JB performs nearly as well after a 20-sec FD as on immediate recall, with the differences in all cases being insignificant. This implies, finally, that B shows virtually no decline after a 20-sec FD. Therefore, from the analysis of JB's performance the two assumptions appear satisfactory. Our two main candidate explanations are in turn indirectly supported. Experiment 1 has shown that a filled delay dramatically reduces immediate recall of both lists and sentences in normal subjects; JB, however, is unaffected by this manipulation. This result supports the traditional distinction between short-term and longer-term memory systems, which we have designated A and B; it also supports our contention that JB lacks a usable store A. Experiment 1 also demonstrates that subjects can recall more words from sentences than would be predicted from their list performance. This means either that the capacity of A can be expanded to take in more sentence items, which are held and retrieved in a listlike manner, or that A and B are not independent and higher-level elements in B somehow support items in A.

8.4. STM and higher-order processing: the contents of B and their relation to A 8.4.1. Experiment 2. Grammaticality judgments The major possibility that remains is that when sentences are heard, comprehension processes construct a representation that is held in Store B, and these representations in Store B are not independent of the phonological representations in A. We now consider in more detail what kinds of representations might be held in B and how they may be related to phonological items in A. Our method here is to make further studies of the performance of JB, in whom, we have argued, the contents of B are normal, but the phonological trace in A is virtually absent. This should enable us to distinguish the processes that require the maintenance of a phonological trace from those that do not. As we have pointed out in the Introduction, many authors have claimed that maintenance of the phonological trace is necessary for the syntactic analysis of at least complex sentences. If this were the case, then JB should be impaired on grammaticality judgment tasks. We therefore tested JB on long (14-21 word) sentences, all of which required the accurate syntactic interpretation of grammatical affixes, function words, or word order, including sentences requiring accurate analysis of long-distance

198

Butterworth, Shallice, and Watson Table 8.4. Grammaticality judgments of JB and controls on long strings, given as d measures (0.00 is chance, 4.66 is perfect performance)

Functors deleted Functors transposed Wrong functor Suffix deleted Wrong suffix Wrong tag Wrong reflexive Wrong voice Overall d'

JB

Controls

3.17 1.68 1.09 1.09 3.17 2.08 3.17 2.58

2.09 3.5$ 1.94 2.29 2.76 1.95 2.52 3.05 2.86

2.1

Range - 0.59-4.66 1.09-4.66 0.59-3.17 0.00-4.66 0.00-4.66 0.59-3.17 0.50-4.66 0.59-4.66 1.66-3.56

Note: The materials and the details of the control group are given in Butterworth et al. (1986). dependencies. Details of these materials are given in Butterworth et al. (1986) and are summarized in Table 8.4. Out of 80 strings, JB made eight misses and eight false positive responses, giving an overall d of 2.1. Looking at each type of ungrammaticality, it can be seen from the d scores that JB's ability to discriminate grammatical from matched ungrammatical strings is always within the normal range, and is usually close to the mean for undergraduate controls. (Similarly good performance was obtained on the grammaticality judgment sentences of Linebarger, Schwartz, & Saffran, 1983.) There is thus no reason to assume that a severe impairment of A would have affected JB's ability to construct or evaluate grammatical representations.

8.4.2. Experiment 3. Retention of syntactically well-formed meaningless sentences In an attempt to identify an effect of syntactic processes separate from other higherorder processes in the construction of representations stored in B, we presented JB and six control subjects with syntactically well-formed but meaningless sentences. This also enables us to assess whether Store A might have a greater capacity in continuous speech than in staccato list presentation mode. The strings each contained five content and three function words. Examples are given in (4): 4. Rapid bouquets are often deterred by sudden nightmares, h/lany funny jewellers created distressed chimneys of them.

Each subject heard 10 sentences of this sort and was asked for immediate recall. The results are given in Table S.5. JB's are the mean of two tests carried out several months apart, with performance being virtually identical on the two occasions.


199

Table S.5. Immediate recall of meaningless eight-word sentences by ]B and six control subjects, given as percent correct

JB Controls range

Sentences correct

Words in sentences correct

Random words correct

0 S3 (60-100)

54 97 (92.5-100)

35 —

Note: JB's performance on random lists of eight words taken randomly from the sentences is also given. In her performance, errors on affixes were ignored.

The control subjects performed near or at ceiling, whereas JB was poor at this task. The experiments carried out with JB contained two sets of 5 eight-word strings presented together with the tests of 10 meaningless sentences in an ABBA design. The eight-word strings consisted of 10 random selections (without replacement) from the SO words used in the meaningless sentences. JB performed significantly more poorly on the random word strings than on the meaningless sentences (Mann-Whitney U[10,10] = 21.5, p < .05 averaging performance on individual lists-sentences across the two test sessions and ignoring affix errors). This means that there is some advantage in having syntactic structure even if it fails to yield meaningful units, but far less than when it comprises a meaningful sentence. It appears that the use of continuous speech may be a factor but by no means the most critical one. Presentation in sentence mode does not in itself lead to a major increase in the capacity of Store A.

8.4.3. Experiment 4. Comprehension and retention of complex sentences: word order effects It is, of course, possible that representations that can support grammaticality judgments are nevertheless inadequate to support comprehension (see Linebarger et al., 1983). One way of assessing this is to analyse recall of complex sentences to see whether gist has been extracted and retained, since this is presumably possible only if comprehension has been achieved. Since JB is unable to maintain a phonological record, it will be possible to estimate from her performance the kinds of information and processing that depend on trace maintenance by breaking down the elements recalled into gist and other components. Now, we do not exclude the possibility that some trace is available at immediate recall; however, this should be completely wiped out following a 20-sec FD. In this experiment we used 20 from Saffran and Marin's (1975) set of sentences, which were designed principally to test subjects' ability to derive syntactic analyses from word order. Examples of these sentences are given in (5).

200

Butterworth, Shallice, and Watson Table 8.6. Recall by JB of Saffran and Marin (1975) sentences, immediately and following a 20-sec FD

Gist preserved Gist half preserved" Gist wrong Errors per sentence where gist preserved Content words Function words

Immediate

FD

16 1 1

12 3 3

0.37 0.37

1.06 0.39

"Where the sense of one phrase, but not more, is incorrectly retrieved.

5. The man the child hit carried the box. The soldiers knew pleasing women can be fun.

The results from immediate recall and following FD are given in Table 8.6. It can be seen that JB was able to recall gist very well, although perhaps not at normal levels, and this performance was not significantly affected by FD (Wilcoxon T = 0, n = 5). FD did, however, affect her ability to retain the exact wording of the sentences; content words were retained significantly worse in that condition (Wilcoxon T = 0, n = 16, p < .005). In the following three examples of her errors in the FD condition, JB had verbatim immediate recall. 6. peaceful neighbourhood-+quiet area searching everywhere —> searching for a long time narrowly avoided - • nearly collided with

In summary, lack of A, the STS, does not seriously compromise JB's ability to understand sentences where structure is critical, even following FD.

8.4.4. Experiment 5. Comprehension and recall of complex sentences: the garden path effect In 8.3.2 and 8.4.3, we examined the effects of a deficient PSTS on the processing of sentences where word order is critical and grammatical relations can span many intervening words and phrases. Sentences with ambiguous structures are also thought to need the maintenance of superficial information, since the hearer may be led up the garden path to an invalid interpretation, and will then have to backtrack and reanalyse the sentence elements. However, as we pointed out in the Introduction, the precise nature of the STM demands for a given type of sentence depends critically on certain theoretical assumptions about grammar and parsing.


201

In the next study, we examined JB's ability to recall a well-known type of garden path sentence. These contain an embedded subjective relative passive clause with the relative pronoun missing. The famous example is: 7. The horse raced past the barn fell. The idea here is that raced is interpreted as the matrix verb, with the horse as the matrix subject, so that when the hearer reaches a second main verb, there is no analysis possible for it. It cannot be another matrix verb, and there is no subordinate clause for it. However, with the relative pronoun and the passive auxiliary, the sentence beginning becomes unambiguous: 8. The horse that was raced past the barn fell. We compared JB's ability to recall both types of sentences. Notice that the garden path version is shorter than the other, and in this study all the sentences were well beyond JB's list span (7-11 words for the garden path versions, 9-13 for the full versions). There were two test sessions, separated by several months. The test sentences were mixed in with the sentences of other types. However, there are other grounds for expecting sentences with the relative pronouns to be easier to interpret because these explicitly mark grammatical relations (see Garrett, Bever, & Fodor, 1966). At the same time, it has been noticed that the hearer is more likely to be misled by a sentence in which the subject is a highly probable agent of the verb, as in (7), but less likely to be so where the subject is a probable (logical) object of the verb, as in 9. The student taught by the new method passed the test (see Crain & Steedman, 1985). We used both types of subject—verb relations to see whether any effects obtained are due simply to the presence of the relative pronoun, or to the effect of a presumed need to backtrack. If the latter, then we would expect an effect of plausibility. The point in the sentence where backtracking was syntactically forced varied from the fourth word to the eighth word. In (9), for example, the eighth, passed, is the earliest word to force backtracking, although the point where a current analysis is abandoned may well be later. To the extent that JB must hold the input prior to the forced word, then one might expect more confusion if backtracking is forced later in the sentence. The data are given in Table S.7. The interpretation of the critical NP-VP relation was easier with relative pronouns present, as one would expect whether or not the sentence as a whole was a garden path type. However, any effect of plausibility was too small to detect. Nor did the location of the word that forced backtracking appear to have an effect. Note that any effect of prosodic cues would serve to aid interpretation of garden path sentences, and hence work against the finding of differences between the two sentence types.

202

Butterworth, Shallice, and Watson Table 8.7. Proportion of correctly interpreted relations between the matrix NP and the embedded verb in garden path sentences without relative pronouns and in sentences with relative pronouns Test date Plausible

Garden path

Relative pronoun

50

84

1986 Implausible

50

Plausible

84

Implausible

67

92

1987

8.4.5. The effects of grammatical structure on recall: additional analyses of data from Experiment 1 Two types of sentences containing lists were employed in Experiment 1: In the first, all the list nouns fell under a single NP node, whereas in the second they fell under two NP nodes, often separated by more than one phrase (see Example in [3]). The idea here is that by dividing list items in this way, additional structural support from B is available for retrieving the items. If it is, then the effect should be more marked after FD when the phonological traces in A are largely lost. The results are presented in Table 8.8 according to the serial position of the item. For four-item lists, immediate recall was perfect, and so only the FD data were included. For six-item materials, we looked at effects of condition (immediate recall vs. FD), structure (one-phrase vs. two-phrase sentences), and position (six serial positions for list words). For four-item lists, subjects score 100% in the immediate condition, and performance was clearly superior to the FD condition. An analysis of variance on the FD condition showed significant main effects of structure — one- vs. two-phrase sentences (F[l, 5] = 17.5, p < .01) - and of serial position (F[3,15] = 3.4, p < .05). Inspection shows an advantage for the earlier items. On the six-item sentences, immediate recall is superior to FD by more than one list word per sentence. In the FD condition, there is no overall effect of structure, but a more detailed analysis reveals a significant advantage on the fourth list word when it occurs as the first list word in the second phrase, as compared with its occurrence as the fourth list word in the single phrase sentences (£[11] = 3.45, p < .01). No comparable differences were found in the immediate condition. This bears out the hypothesis of the effectiveness of structural support in recall when phonological information is less available. In particular, it shows that the probability of word recall depends on its position in the sentence - not just its serial position, but its position in the grammatical structure.


203

Table 8.8. Structure effects on sentence recall with and without filled delay; maximum number of list-type words correct per position was five (five subjects)

Items in two phrases

Four-item sentences

Six-item sentences

Phrase position

FD

Phrase position

Immediate

FD

XI X2

4.5 4.0

XI X2 X3

5.0 4.2 3.7

3.3

Yl Y2

2.8 3.2

Yl Y2 Y3

3.75 2.3 3.7

3.2 1.7 1.75

4.5

3.2

4.2 3.6 3.5 3.3 2.8 4.4

4.2 3.1 2.3 2.1 1.9 1.8

4.4

3.1

Total Items in one phrase

Total

3.6 Zl Z2 Z3 Z4

3.8 3.8 2.3 2.8

Zl Z2 Z3 Z4 Z5 Z6

3.0

3.75 2.6

Note: Immediate recall for four items was virtually 100%.

8.4.6. Experiment 6. Recall of complex sentences: structural effects If this last claim is correct, the probability of recall will depend critically on sentence structure when recovery of the phonological information is not possible. To explore this, we tested JB on a new set of sentences of different structural types, matched, as far as possible, on lexical content. Materials We constructed 50 sentences, each containing five content and four function words. The nouns were such as to be plausible candidates for any of the NP roles. There were 10 sentences in each of the following structural types. 10. Active E.g.: One of the departing businessmen insulted the gifted economist. 11. Passive E.g.: The miserable twins were heard by the frantic mother. 12. VP Coordination E.g.: An unarmed boxer will chase and fight the wrestler. 13. Subject Relative E.g.: The stag that startled the magnificent lion was beautiful 14. Object Relative E.g.: The truck that the blue van pulls is green.

204

Butterworth, Shallice, and Watson Table 8.9. Recall of complex sentences by JB Number of content words omitted

Sentence type Active Passive VP coordination Subject relative Object relative

0

1

2

3-f

Total

1 2 0 0 1

5 6 4 1 1

3 1 5 3 3

1 1 1 6 5

10 10 10 10 10

Note: Figures in cells designate number of sentences with content words omitted.

Retention was measured by the number of content words correctly recalled. These are given in Table 8.9. Collapsing columns 0 and 1, and 2 and 3, a chi-square test showed a significant effect of sentence type (%2 = 13.47, df= 4, p < .01). Assuming the order of difficulty of sentence types was as listed, a Jonckheere Test of Trend again showed a highly significant difference among sentence types (S = 364, z = 3,25, p < .001). Inspection reveals a major difference between Actives and Passives on the one hand, and relative-clause sentences on the other. Where only a single content word was omitted, on 13/16 occasions it was an adjective, significantly more frequently than would be expected by chance (#2 = 11.3, p < .001). These results, along with those in sections 8.4.4 and 8.4.5, show that in the absence of a phonological record, verbatim recall depends on grammatical structure. There is thus no simple "sentence span," analogous to list span. We do not yet have a comprehensive view of the structural factors that affect sentence recall, but certain features are clear: The availability of an unambiguous structure helps recall, and the presence of several identical structural elements (like NPs dominated by NPs) impairs recall. In Experiment 6, relative clauses also impair recall, although this does not appear to be due to the presence of a gap, or trace, in the grammatical structure, since these also appear in the passive and VP coordinate sentence, according to current theory.

S.5. Discussion Both immediate recall and recall after FD are better for sentences than for lists, and some kinds of sentences are better than others. This advantage is attributed to the additional syntactic and semantic information carried by sentences (but not by lists), which is encoded into a separate nonphonological memory that we have designated "B." Two techniques were employed to estimate the contribution of B to recall. The first used


205

filled delay, which presumably had the effect, in normal subjects, of degrading the phonological trace in the STS, which we have designated "A." The second technique used a subject, JB, who was known to have a severely impaired STM performance attributable to a deficit of A. The completely normal performance of JB, when tested after a filled delay, together with her intact spontaneous speech, makes it more plausible than for many other STM patients that her performance on tasks that require explicit or implicit repetition of the speech input is affected only by impairments to STS.2 Theoretically, the endogenous absence of phonological information should have the same consequences as its experimental elimination, unless this information is necessary for encoding into B, that is, for on-line comprehension. This is how it turned out: Following FD, recall by our normal controls became equivalent to that of JB. Thus, there were two equivalent ways of estimating the contribution of B, in the absence of A, to recall, and so drawing conclusions about the contribution of A itself. It was demonstrated that, given certain assumptions, the amount retrievable from A was greater in sentence than in list recall. The main assumptions are that the contribution of A diminishes to zero after a 20-sec FD, whereas the contribution of B is unaffected by this manipulation. In classic memory terms, A has short-term and B longterm memory characteristics. The performance of the STM patient, JB, supports both assumptions, corroborating the earlier studies with her (Warrington et al., 1971; Shallice & Butterworth, 1977).3 Since this finding implies that A and B cannot be assumed to be independent, we have argued that elements in A and B are mutually supportive. The data from our various sentence tasks allow us to be more specific as to how this support works. A syntactic-semantic (S-S) structure is formed at input that contains two kinds of elements: those adequately supported (and realized in gist) and those inadequately supported. Inadequately supported elements can be retrieved only with the aid of a phonological record, but themselves aid retrieval from the phonological record. Adequately supported elements can be retained and retrieved without the assistance of the phonological record (see Figure 8.1). Adequately supported information, on this account, can be distinguished operationally by 1. whether it can be retrieved after a 20-sec FD; and 2. whether it can be retrieved immediately by a pure STM patient with a total deficit, to which JB is a close approximation. On the basis of the studies reported here, we cannot give a full formal characterization of the kinds of material that will induce the construction of adequately and inadequately supported elements, nor the types of processing procedures involved. We do not view the concept of adequate support as all-or-none; rather, elements will be more or less adequately supported. In general terms, the degree of support among elements will depend on the kind of higher-level representation constructed: The higher the level, the better supported the elements will be.

206


Q'""0 O "0 ""0

6 6 6 6 6 List Memory

oooooo Sentence Memory Figure 8.1. Representation of a list and sentence containing a number of elements supported by only a single node (X) in Store B. Full lines indicate support adequate for retrieval from Store B alone.

Can the present findings be used to provide further information on the situations in which the support obtained is relatively inadequate? This depends on whether it is appropriate to use satisfactory reproduction of gist as a measure of sentence comprehension. Clearly any test of comprehension introduces extra processing demands specific to {he task; indeed, the concept of a task-independent comprehension is only a convenient abstraction. Reproduction of gist can overestimate comprehension if the phonological record contributes additional information; this is clearly not the case in an STM patient who performs very similarly in immediate reproduction and after a 20-sec FD. Is it plausible that the measure underestimates comprehension? For particular subjects who perform sentence repetition more poorly than some other measure of comprehension, then it is likely that factors specific to those subjects are affecting the measure.4 The following conclusions about the adequacy of structure can be derived from our findings:


207

1. If it is not possible to construct a semantic interpretation for an input string, the elements will be much less well supported than if it is. Evidence for this comes from JB's poor recall of meaningless sentences. The internal organization of S-S, and higher, structures will also determine the degree of support among elements; in particular, 2. Multiple elements having equivalent roles in the overall S-S structure will not be adequately supported. Lists, and lists within sentences, are paradigm examples. Each list item will be a noun phrase dominated by an NP. (Notice that recall of list elements depends on structural features, so that the first element in an NP is better recalled than items in other serial positions.) We predict that this generalizes to all kinds of coordinate structures, including verb phrase coordination. In Experiment 6, the one condition for which JB, without the benefit of Store A, performed as badly with verbs as with adjectives, was that involving VP conjuncts. One reason for the lack of support of multiple elements may lie in their relative lack of distinctiveness in memory - equivalent elements in a substructure will all have very similar representations. This is an analogous principle to the "cue overload" hypothesis of Mueller and Watkins (1977). A second possibility lies in what one might characterize intuitively as the obligatoriness of elements in the structure. Presumably, there will be some advantage in trying to retrieve obligatory elements because they will be wholly or partly reconstructible from other elements. Intuitively, only one element in a multiple-element substructure will be obligatory; for example, only one NP will be required following a transitive verb, and even if S-S structure includes a head node dominating coordinate NPs, the grammatical information will indicate merely that two or more NPs are needed. Thus, 3. With only grammatical information available, elaborations on obligatory structures will be inadequately supported. We found in Experiment 6, for example, that adjectives and relative clauses, which intuitively are both elaborations of basic NP structure, are vulnerable to loss. We would predict however, that suitable contextual conditions could induce structures to support these elements and improve recall. For example, semantic or discourse information may specify the number of coordinates, as in ''The three greatest evils of our day are sex, drugs, and " Our studies have used only isolated sentences with no specified discourse function. Text materials, and ordinary conversations, presumably induce higher levels of structure, with words and S-S elements linked to mental model representations

208


(Stenning, 1978; Johnson-Laird, 1983). Such materials, we predict, would be better recalled than isolated, meaningful sentences. (For some evidence in support of this, see Wingfield and Butterworth, 1984.) The familiarity or accessibility of mental model representations would influence the adequacy of the structures formed. Thus, texts of familiar situations, like shopping and getting to work, should, ceteris paribus, give rise to adequate memorial structures. Pragmatic factors, like plausibility and the correlation of text order with event order, should also affect adequacy (see McCarthy & Warrington, 1987b, and this volume, chapter 7). Of the factors suggested by Shallice (1979) for the superiority of sentence over list utilization of STS, the first factor — retrieval facilitation from semantic and syntactic information retained separately from the phonological representation — is supported. The lack of a decisive advantage for meaningless sentences over lists, and especially JB's poor performance on this task, suggests that S-S structures are primarily semantic and that the prosodic and other phonological cues to structure are not a critical difference between lists and sentences; thus, little support has been obtained for Shallice's second factor. The longer sentence materials used in these studies are beyond the list span of the subjects. This means that good recall depends on the construction of S—S representations in B by comprehension processes. Although loss of the phonological record through FD, or endogenously, affects recall dramatically (and in specified ways, as our theoretical accounts predict), it is not clear that it has any effect on on-line comprehension processes. In fact, JB performed normally on grammaticality judgments, where the construction and examination of at least a syntactic analysis seems required. On this position, the loss of the phonological record will have no effect on comprehension where this corresponds to construction and maintenance of an adequately supported structure. Only when the test of comprehension requires the accessing of elements in an inadequately supported structure will lack of the phonological record be a problem. Examples of this would be sentences with a number of structurally interchangeable elements, or where the interpretation of an adjective is required, like the Token Test.5 The standard position has been that an auditory-verbal STS (Store A) is necessary for speech comprehension, because it holds word and order information while a syntactic representation is being worked out; hence the observed association in patients of both STM and comprehension deficits, particularly for long or complex sentences, where the exact order or relation among words is critical (e.g., Saffran & Marin, 1975; Shallice, 1979; Vallar & Baddeley, 1984). Recently, however, this position has been challenged by Butterworth et al. (1986) and by Caplan, Vanier, and Baker (1986) who have found that some subjects with STM deficits have no special difficulty comprehending syntactically complex material. If construction of the S—S structure is unaffected by loss of the phonological record,


209

the maintenance of a phonological record cannot be necessary for on-line interpretation of syntactic structure, and hence it cannot explain the observed association between STM and comprehension deficits. One way of saving the standard position might be to argue that a phonological record is held until the end of sentence, during which time the construction of an S-S structure can be carried out before FD begins. It may be possible for normal subjects to operate in this way, but it seems unlikely for patients with endogenous STM deficits, like IL (Saffran & Marin, 1975) or PV (Vallar & Baddeley, 1984), that a sentence's worth of material can be retained during input. Indeed, in probe recognition tests (with lists) STM patients still show severely impaired performance (Shallice & Warrington, 1970, 1977). A second manoeuvre might be to assume that on-line comprehension processes require a phonological record of far fewer words than was originally supposed, say, two or three. Although this preserves the form of the standard position, it certainly reduces the importance of the phonological record in comprehension, and leads to other consequences. For example, long sentences in themselves would not pose a special difficulty, provided that they could be correctly parsed using a lookahead confined to the next two or three words; but comprehension difficulties would be predicted for just that class of sentences that need long lookahead for correct interpretation. In a transformational grammar, transformations move elements around, and the treatment of many sentence types requires the linking of elements to the locations from which they have been moved. Since elements and their original locations can be arbitrarily far apart in surface structure, a natural implementation of this grammar entails a (phonological) record of many elements awaiting the assignment of syntactic interpretation (as in Wanner & Maratsos, 1978). Hence, these sentence types would be predicted to cause comprehension difficulty. However, in earlier investigations, neither RE nor TI (Butterworth et al. 1986; Saffran and Martin, this volume, chapter 16) was below control levels for grammatical judgments on long sentences of these types - for example, those containing agreement or unbounded dependency between items separated by 10 or more words. The same is true for JB. Yet all these subjects appear to operate with virtually no phonological record. Reduced dependence on a phonological record will thus be more plausible where the parser implements a grammar that does not move elements around but generates them directly in their final locations. Several of these grammars are currently under investigation. In these, an element that is part of an unbounded dependency construction (e.g., WH- constructions) can normally be analysed with reference to just the next constituent. The shift-reduce parser of Pereira (1985), described in the introduction, is one example. It is perhaps significant that this is what the most successful computer parsers do (see Gazdar & Pullum, 1985, for a review). However, as Gazdar and Mellish (1987) point out, an endemic property of language, and a central

210


problem for parsers, is the analysis of ambiguous constructions. Typically these can be resolved locally, but not always: Garden path sentences are an instance in which resolution needs relatively long lookahead, or relatively long backtracking - both of which would appear to entail some record of several input words and their order. In our findings, JB had more difficulty with this type of sentence, although it is not clear that this was because resolution needed to refer to distant constitutents. Since competing parsers make different assumptions about lookahead (or backtracking) for different sentence types, patients like JB would appear to provide a good test bed for different parsing theories: Sentences requiring long lookahead for successful resolution should be exactly those on which JB should fail, but the competing theories will differ on which those sentences should be. To maintain a role for a reduced phonological record in the construction of an S-S structure appears to entail abandoning transformational grammar in favour of a mathematically more restrictive type of grammar. Moreover, if a reduced record - of, say, three words - is sufficient for normal comprehension, proponents of the standard view still have to explain their claimed association between STM and comprehension deficits. Given the findings reported here, it is more perspicuous to assume that for texts that yield adequate representations there is no role in comprehension for a phonological record, and to attribute comprehension difficulties for such tests, in those patients who show them, to an additional deficit.

Notes 1. This formulation does not require that the identical form of an item (a word) is in A or B, or both. Indeed, it implies that a word in A is represented phonologically and in B more abstractly. Rather, it requires that both representations denote the same item, although by different means, just as the same house may be designated either by an address or a map reference. 2. Allport (1984) has suggested that JB is impaired on word reproduction tasks. He states: "Phonemic paraphasias also frequently occurred when JB was requested to repeat single words from the same low-frequency and low-imageability range." In fact, in reproducing one from a set of 240 words of one, two, and three syllables of high and low imageability and in three frequency ranges, she was over 90% correct and showed no effects of length, imageability, or frequency (Shallice, unpublished). Errors were indeed mainly literal paraphasias but these were even rarer in reproduction of short sentences. Allport's stimulus set may have contained items not in her speech vocabulary. 3. Allport (1984) has argued that JB's impaired short-term retention is secondary to phonological processing difficulties. He produced a number of arguments, for instance, that she had a higher error rate in auditory lexical decision than did normal subjects, particularly for the distractors (20% vs. 10%). His argument has been disputed (see, e.g., Vallar & Baddeley, 1984; Shallice, 1988, who have held that the tasks that Allport discusses also have a short-term memory loading). In fact, JB scored 96/100 on CV-CV matching with minimal pairs (Shallice, unpublished). The present experiments clearly indicate that any minimal phonological processing difficulties she might have do not lead to impaired on-line comprehension or storage difficulties in Store B. 4. Thus EA (see Friedrich, chapter 3), who performed much better on picture-sentence matching


211

of reversible sentences than on repeating them, had some difficulties in speech production. RE, who performed better on the Token Test than in repeating its stimulus sentences, may well have coded a visuomotor plan on hearing the sentence when she had to carry out the instruction physically; in sentence repetition this possibility was not available to her. JB's speech production system was basically intact (Shallice & Butterworth, 1977), so there is no reason to assume that the additional factors involved in gist reproduction make it a less adequate measure than, say, sentence-picture matching or acting out. 5. JB's relatively poor performance with complex relative clause sentences in Experiment 6 suggests that in certain situations relative clause interpretation may present difficulties on-line.

References Allport, D. A. (1984). Auditory-verbal short-term memory and aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 318-325). London: Erlbaum. Baddeley, A. D. (1966). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experimental Psychology, 18, 302-309. Boakes, R. A., & Lodwick, B. (1971). Short-term retention of sentences. Quarterly Journal of Experimental Psychology, 23, 399-409. Brown, J. (1958). Some tests of the decay theory of immediate memory. Quarterly Journal of Experimental Psychology, 10, 12-21. Butterworth, B. L, Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A, 705-737. Campbell, R., & Dodd, B. (1984). Aspects of hearing by eye. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 299-352). London: Erlbaum. Caplan, D., Vanier, N., & Baker, C. (1986). A case study of reproduction conduction aphasia II: Sentence comprehension. Cognitive Neuropsychology, 3, 129—146. Clark, H. H., & Clark, E. V. (1977). Psychology and language. New York: Harcourt Brace Jovanovich. Conrad, R. (1964). Acoustic confusion in immediate memory. British Journal of Psychology, 55, 75-84. Craik, F. M. (1968a). Types of error in free recall. Psychonomic Science, 10, 353-354. Craik, F. M. (1968b). Two components in free recall. Journal of Verbal Learning and Verbal Behavior, 7, 996-1004. Craik, F. M. (1971). Primary memory. British Medical Bulletin, 27, 232-236. Craik, F. M., & Levy, B. A. (1970). Semantic and acoustic information in primary memory. Journal of Experimental Psychology, 86, 77-82. Craik, F. M , & Masani, P. (1969). Age and intelligence differences in coding and retrieval of word lists. British Journal of Psychology, 60, 315-319. Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by the psychological syntax processor. In D. Dowry, L. Karttunen, & A. Zwicky (Eds.), Natural language parsing (pp. 320-358). Cambridge: Cambridge University Press. Flores d'Arcais, G. B., & Schreuder, R. (1983). The process of language understanding: A few issues in contemporary psycholinguistics. In G. B. Flores d'Arcais & R. Jarvella (Eds.), The process of language understanding (pp. 1-42). New York: Wiley. Garrett, M. F., Bever, T. G., & Fodor, J. A. (1966). The active use of grammar in speech perception. Perception and Psychophysics, 1, 30—32. Gazdar, G., & Mellish, C. (1987). Computational linguistics. In J. Lyons, R. Coates, M.Deuchar, & G. Gazdar (Eds.), New Horizons in linguistics II. Harmondsworth: Penguin.

212


Gazdar, G., & Pullum, G. (1985). Computationally relevant properties of natural languages and their grammars. New Generation Computing 3, 273-306. Jarvella, R. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Behavior, 10, 409-416. Jarvella, R., & Herman, S. J. (1972). Clause structure of sentences and speech processing. Perception and Psychophysics, 11, 381-384. Johnson-Laird, P. (1983). Mental models. Cambridge: Cambridge University Press. Johnson-Laird, P., & Stevenson, R. (1970). Memory for syntax. Nature 227, 412. Kintsch, W., & Buschke, H. (1969). Homophones and synonyms in short-term memory. Journal of Experimental Psychology, 80, 403-407. Levy, B. A., & Craik, F. M. (1975). The co-ordination of codes in short-term retention. Quarterly Journal of Experimental Psychology, 27, 33-46. Linebarger, M. G, Schwartz, M. F., & Saffran, E. M. (1983). Sensitivity to grammatical structure in so-called 'agrammatic' aphasics. Cognition, 13, 361—392. McCarthy, R., & Warrington, E. K. (1987a). The double dissociation of short-term memory for lists and sentences: Evidence from aphasia. Brain, 110, 1545-1563. McCarthy, R. A., & Warrington, E. K. (1987b). Understanding: A function of short-term memory. Brain, 110, 1565-1578. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86. Miller, G., & Selfridge, J. (1950). Verbal context and the recall of meaningful material. American Journal of Psychology, 63, 176-85. Mueller, C W., & Watkins, M. J. (1977). Inhibition from post-set cuing: A cue-overload interpretation. Journal of Verbal Learning and Verbal Behavior, 16, 699—709. Pereira, F. G N. (1985). A new characterisation of attachment preferences. In D. Dowty, L. Karttunen, & A. M. Zwicky (Eds.), Natural language parsing (pp. 307-319). Cambridge: Cambridge University Press. Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193-198. Sachs, J. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception and Psychophysics, 2, 437-442. Saffran, E., & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with deficient auditory short-term memory. Brain and Language, 2, 420-433. Savin, H., & Perchonock, E. (1965). Grammatical structure and immediate recall of sentences. Journal of Verbal Learning and Verbal Behavior, 9, 348—353. Shallice, T. (1975). On the contents of primary memory. In P. M. A. Rabbitt, & S. Dornic (Eds.), Attention and performance (Vol. 5, pp. 269-280). London: Academic Press. Shallice, T. (1979). Neuropsychological research and the fractionation of memory systems. In L. J. Nillson (Ed.), Perspectives on memory research (pp. 157-277). Hillsdale, NJ: Erlbaum. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, T., & Butterworth, B. L. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia. 15, 729-735. Shallice, T., & Warrington, E. K. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261—273. Shallice, T., & Warrington, E. (1977). Auditory-verbal short-term memory and conduction aphasia. Brain and Language, 4, 479-491. Shulman, H. G. (1970). Encoding and retention of semantic and phonemic information in shortterm memory. Journal of Verbal Learning and Verbal Behavior, 9, 499-508. Smith, E. E., Barresi, J., & Gross, A. E. (1971). Imaginal versus verbal coding and the primary-secondary memory distinction. Journal of Verbal Learning and Verbal Behavior, 10, 597-603.


213

Stenning, K. (1978). Anaphora as an approach to pragmatics. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality. Cambridge, MA: MIT Press. Vallar, G., & Baddeley, A. D. (1984). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Wanner, E., & Maratsos, M. (1978). An ATN approach to comprehension. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality. Cambridge, MA: MIT Press. Warrington, E. K., Logue, V., & Pratt, R. T. C. (1971). The anatomical localisation of selective impairment of auditory verbal short-term memory. Neuropsychologia, 9, 377-387. Waugh, N., & Norman, D. (1965). Primary memory. Psychological Review, 72, 89-104. Wingfield, A., & Butterworth, B. (1984). Running memory for sentences and parts of sentences: Syntactic parsing as a control function in working memory. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 351-364) London: Erlbaum.

Part III. Short-term memory studies in different populations (children, elderly, amnesics) and of different short-term memory systems

The behaviour of "short-term memory" patients on short-term memory tasks is of interest because their performance contrasts so strongly with those of normal adults. The approach of contrasting normal adult behaviour with that of other groups can be extended to other populations whose performance differs markedly from the normal adult pattern. This part includes discussions of short-term memory performance in children (Hitch, chapter 9); in the elderly (Craik, Morris, & Gick, chapter 10); in the deaf - or rather the procedure, lipreading, they use (Campbell, chapter 11) and in two types of neurological patients (Howard & Franklin, chapter 12; Kinsbourne & Hicks, chapter 13). Hitch (chapter 9) reports a number of studies concerning the development of working memory in children of various ages. He suggests that developmental studies may usefully complement neuropsychological research in advancing our understanding of normal cognitive processes, such as short-term memory, since both can be based on a "fractionation" methodology. The neuropsychological fractionation method currently used in patients with acquired brain lesions capitalizes on the presumed more or less complete damage of specific functional component(s), for example, the phonological short-term store, to investigate the functional architecture of aspects of the cognitive system. In the case of normal children the fractionation approach advocated by Hitch assumes that the normal development of cognitive abilities may be characterized by the addition of subsystems, which previously were relatively nonoperative. If this is the case, the study of children of different ages should produce results complementary to those obtained with brain-damaged patients. A given subcomponent (e.g., some involved in rehearsal) may not be active in younger children, who would then be expected to have a pattern of performance broadly parallel to that of patients with an acquired deficit of this subsystem. As Hitch points out, there are of course other developmental possibilities (e.g., deletion of subsystems, radical reorganization of the functional architecture at a specific age) that could make difficult, if not impossible, the utilization of the developmental fractionation method. Furthermore, other possible factors such as age-related differences in the efficiency of the system under investigation 215

216

Part III

could make it inappropriate to draw direct inferences from developmental data to normal adult function. While it is then entirely appropriate to adopt a cautious attitude, the empirical data provide a clear illustration of the potential of this type of fractionation, at least in the case of short-term memory. Hitch, Halliday, and their coworkers have adopted a multistore working memory framework of the type reviewed by Baddeley (chapter 2) in their investigation of the developmental course of some aspects of verbal short-term memory. They have shown that younger children are able to code phonologically and rehearse auditorily presented verbal material, but that this does not occur for visual input. In the case of older children, however, phonological coding and rehearsal are available for both input modalities. This developmental dissociation has close parallels in both normal subjects and brain-damaged patients (see Shallice & Vallar, this volume, chapter 1). Most of the preceding chapters are concerned with the functional properties and patterns of impairment of a relatively passive phonological short-term storage subcomponent of verbal memory and of a closely connected rehearsal process, and do not address the problem of any putative role of more central components in immediate retention (see Shallice & Vallar, chapter 1; but for an exception, see McCarthy & Warrington, chapter 7). Craik et al.'s investigation of immediate memory performance in a population of young and old normal subjects (chapter 10) is directly relevant to this issue. They draw an operational distinction between "primary memory tasks" (e.g., span), which require little manipulation, translation, and recoding of the material between input and output, and "working memory tasks," which call for a more or less substantial reorganization of the material held. It should be noted here that the control process component of the verbal working memory system of Crain, Shankweiler, Macaruso, and Bar-Shalom (chapter 18) is also involved in the translation of information between different levels of processing, although its role is strictly confined to speech comprehension. Craik et al. use dual task paradigms, where a decision-making and reasoning task (sentence verification) is carried out with an immediate memory task. Since age differences in primary memory tasks are comparatively minor, Craik et al. interpret the clear impairment of their old subjects in terms of declining efficiency of a central executive component of working memory (see Baddeley, 1986). Both primary and working memory tasks are likely to involve storage components, such as the phonological short-term store and the process of rehearsal. These tasks may, however, differ in the relative role of the central component. Immediate memory span, which involves one set of operations on one set of material, would mainly represent the output of a rather peripheral component such as the phonological short-term store (see also Shallice & Vallar, chapter 1), requiring a comparatively minor contribution from the central executive. This in turn would be involved in the dual tasks used by Craik et al. in which subjects are required to retain some words while making decisions about additional verbal material. Craik et al.'s studies provide an instance of the fractionation

Studies in different populations and of different systems

217

of short-term memory. Within a multistore framework, some of the functional properties of the unitary short-term store of Atkinson and Shiffrin (1971), such as decision making and passive store, are here attributed to different subcomponents of the memory system (see also McCarthy & Warrington, chapter 7; Crain et al., chapter 18). Two chapters discuss the fractionation of the phonological system along the input—output dimension. On the basis of normal and neuropsychological data concerning a specific phenomenon, lipreading, Campbell (chapter 11) puts forward a model of speech perception and phonological memory that contains input and output subcomponents. She suggests a distinction between input and output phonetic-phonological subsystems, each having both phonetic (acoustic and lipread) and abstract phonemic units. Heard and seen (lipread) speech have access to the input subsystem, whereas spoken or mouthed speech is produced by the output subsystem. The functional architecture of these subsystems is conceived in terms of the fully interconnected components of the TRACE model (see McClelland & Elman, 1986; see also Friedrich, chapter 3, for a related account of the properties of the phonological code). It does not include distinctions between processing and storage subcomponents (see Shallice & Vallar, chapter 1; Baddeley, chapter 2). In Campbell's view a number of characteristics of normal memory performance (i.e., modality, recency, and suffix effects in immediate recall) may be interpreted in terms of differential patterns of activation of input and output phonological components. At the neurological level, relying on findings in unilateral brain-damaged patients, such as word deaf patients, and from laterality and neurophysiological experiments in normal subjects, Campbell suggests that there is some contribution from the right hemisphere to the phonetic analysis of lipread speech. Although the input and output phonetic-phonological processors are held to be located in the left hemisphere, dominant for language (see also Shallice & Vallar, chapter 1), the right hemisphere would contribute to the phonetic analysis of seen place of articulation (see Campbell, chapter 11, Figure 11.1). Howard and Franklin's paper (chapter 12), which is also concerned with the fractionation of the phonological code into input and output subcomponents, may be located within a multistore approach to short-term memory (see Shallice & Vallar, chapter 1; Baddeley, chapter 2). Howard and Franklin distinguish an auditory-phonological input short-term store, output phonological representations, and a process (rehearsal) that converts input phonology into output phonology, and vice versa. They report the case of a patient, MK, whose pattern of phonological impairment is interpreted as a selective rehearsal disorder. Interpretation of the case is, however, complex, as he suffers from a co-occurring deficit of an auditory input lexicon component and has a severe sentence comprehension impairment. Howard and Franklin suggest that the patient has a problem in input-to-output conversion: Repetition of auditory nonwords is impaired and in immediate memory tasks auditory material is not

218

Part III

rehearsed. Output-to-input conversion is also impaired: In span tasks visual material does not gain access to the phonological store. Both input and output phonology, on the other hand, are argued to be unimpaired because of the relatively perserved performance in auditory—verbal discrimination and nonword reading tasks. On this analysis MK has an isolated phonological short-term store, which is not supported either by rehearsal or by lexical information. Seen in this perspective, Howard and Franklin's study complements those by Saffran and Martin (chapter 6), McCarthy and Warrington (chapter 7), and Butterworth, Shallice, and Watson (chapter 8), which investigate the role of nonphonological components in immediate retention. Finally, in addition to immediate memory performance, Howard and Franklin assess the role of input and output phonological subcomponents in nonmemory tasks such as rhyme and homophone judgments with auditory and visual input. MK's selective pattern of impairment, a deficit in rhyme judgments with written presentation, provides further evidence for a fractionation of the phonological code. The studies by Craik et al. (chapter 10) relate to memory systems different from those discussed in earlier chapters. They are, though, entirely compatible with the approach of the earlier analyses. The study of short-term memory by Kinsbourne and Hicks (chapter 13) has a very different framework. They investigate time estimation abilities in alcoholic Korsakoff amnesics, patients who typically suffer from a more or less selective impairment of episodic long-term memory (see, e.g., Cermak, 1982). Their amnesics have a preserved performance on the task up to a time period of about 20 sec, followed by a disproportionate underestimation of passing time for longer periods. In a second experiment in normal subjects, Kinsbourne and Hicks contrast prospective and retrospective time judgments, obtaining results that support the neuropsychological dissociation. They interpret their findings with reference to William James's (1890) notion of primary memory as the psychological present, being distinguished from secondary memory, which concerns the psychological past. Within a classical shortterm/long-term memory framework, their results extend to time estimation the notion that short-term memory is spared in amnesia and are in good agreement with the classical data from this type of patient (see, e.g., Baddeley & Warrington, 1970): Aspects of long-term memory are severely impaired, independent of input modality, whereas short-term memory is substantially preserved. This pattern is clearly different from the modality-specific (auditory-verbal) short-term memory deficit discussed in the majority of chapters of this book. In fact, as Kinsbourne and Hicks point out, no shortterm memory patient with a deficit complementary to amnesia (i.e., a short-term memory impairment nonspecific to a single modality or code) has yet been described. The short-term retention investigated by Kinsbourne and Hicks can hardly be based on the phonological short-term storage system involved in immediate verbal memory span. It is also worth noticing that brain-damaged patients with selective deficits of auditory-verbal span do not show any clinical evidence of temporal disorientation (see

Studies in different populations and of different systems

219

Shallice & Vallar, chapter 1), even though they have not been formally tested with Kinsbourne and Hicks's paradigm. Like the study by Craik et al. (chapter 10), Kinsbourne and Hicks's experiments suggest that a number of short-term memory components exist, these being involved in different cognitive activities.

References Atkinson, R. C , & Shiffrin, R. M. (1971). The control of short-term memory. Scientific American 225, 82-90. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and the distinction between long- and short-term memory. Journal of Verbal Learning and Verbal Behavior, 9, 176—189 Cermak, L. S. (1982). Human memory and amnesia. Hillsdale, NJ: Erlbaum. James, W. (1890). The principles of psychology. New York: Holt. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86

9. Developmental fractionation of working memory GRAHAM J. HITCH

9.1. Introduction In recent years, neuropsychological evidence has proved to be of considerable value in advancing our understanding of cognitive processes and their organization. A particularly powerful methodology is that of "neuropsychological fractionation" (see, e.g., Shallice, 1979a), which attempts to interpret highly selective neuropathological deficits in cognitive abilities in terms of models of the intact cognitive system in which one assumes damage to specific components. In this chapter I wish to discuss a somewhat analogous method of "developmental fractionation," which involves studying the cognitive abilities of normal children and which I believe can usefully complement neuropsychological evidence in constraining models of adult function. Since developmental fractionation is not, to my knowledge, particularly widely used, I shall begin by describing what it involves and the general rationale behind it. I shall then go on and describe some research on the development of "working memory" (Baddeley & Hitch, 1974; Baddeley, 1986), which I think illustrates the potential of the method. This work bears directly on the fractionation of visual and phonological components of working memory and on the infrastructure of these components. I conclude by considering some of the mutual implications of developmental and neuropsychological evidence about working memory, and the ways in which they can usefully complement one another.

9.2. What is developmental fractionation? Although currently rather little used, developmental fractionation has, by psychological standards at least, a long history. Baldwin (1894) took the view that the complex cognitive abilities of adults could be regarded as being "the union of simpler elements" Many of the ideas in this paper were developed jointly with Sebastian Halliday in a research project supported by the Economic and Social Research Council. Janet Littler assisted in carrying out the experiments on presentation rate. I am grateful to Ruth Campbell, Sebastian Halliday, Peter Meudell, Tim Shallice, and Don Shankweiler for their critical comments on earlier versions of this chapter.

221

222

Hitch

that are assembled during childhood development. He suggested that if one wanted to identify these elements and study them, childhood is a very good place to start looking. Developmental fractionation is essentially no more than the attempt to separate out components of the adult cognitive system by examining normal children at different ages. In this way it can provide a useful way of testing theories about the adult system. It is interesting to consider the plausibility of Baldwin's suggestion. His first assumption concerns the possibility that the adult cognitive system is made up of simpler elements. Put so generally, this is perhaps uncontroversial. In modern terms, it is closely related to the stronger claim that such elements consist of functionally separate, independent subsystems for processing and storing information. This claim is often referred to as the "modularity hypothesis" (Fodor, 1983). There are, however, several alternatives to Fodor's specific suggestion about the nature of modules (see, e.g., Shallice, 1984). Currently, several information-processing theories of memory incorporate this kind of assumption (e.g., Broadbent, 1984), and it is common in modelling other aspects of cognition too, for example, reading and related skills (e.g., Morton & Patterson, 1980). An important distinction within such models is between the "functional architecture" they propose, the subsystems and their permanent interconnections, and the dynamic processes that control the flow of information between subsystems. This distinction has its roots in Atkinson and Shiffrin's (1968) separation of variable control processes from the fixed structure of the human memory system. The modularity hypothesis is also a common assumption in modern cognitive neuropsychology. Here, considerable success has been attained in identifying patients with highly selective cognitive deficits and interpreting their performance in terms of damage to specific subsystems in models of normal function. This has been especially true for deficits of memory (Shallice, 1979a) and reading (Coltheart, 1985), where suitable, reasonably well worked out models are available. The question of how a modular information processing system might develop is actually quite open and has been relatively little discussed. There are a number of interesting possibilities. The simplest is the "preformist" notion that the functional architecture of the system remains constant right from the start of development, with developmental differences being confined to changes within subsystems. This is similar to Fodor's (1983) view that an independent developmental course is a defining property of a module. An analogy can be drawn with the physical development of the human body, where parts such as the limbs and the trunk can grow at different rates, but the number of parts and their interrelationships remain the same. A second, more complex possibility is that the number of information-processing subsystems also changes, but not so radically that the functional architecture of the system as a whole is unrecognizably altered. On this view, developmental change would be characterized by the addition and possibly even deletion of subsystems. For an analogy based on physical development, we might take an example such as the growth and shedding of


223

antlers in the life of the stag. Yet a third, more dramatic possibility is that both the number of information-processing subsystems and their functional organization change radically. This might be likened to the metamorphosis in physical growth whereby the caterpillar is transformed into a butterfly. Here, the change may be so great that the immature structure bears very little resemblance to the mature state. To complicate matters further, there are also likely to be changes to the dynamic processes that control the flow of information among subsystems, over and above whatever changes there are to the functional architecture itself. Granted such a range of possibilities, we can see that it is by no means inevitable that the study of cognitive function in children will provide useful clues about the components of the adult system. In particular, the possibility of developmental fractionation seems to be critically dependent on the existence of relatively simple, highly specific differences between either the functional architecture or the control processes or both of children of a given age (which we will call X) and adults. For this to be at all probable, it is important that the cognitive system does not undergo any radical reorganization after age X. If it does, then models of adult function are clearly going to be inapplicable to the analysis of the children's performance. The possibility of some sort of cognitive reorganization during development is actually quite commonly entertained, most frequently within the context of Piaget's views. However, Piaget was concerned with logical aspects of cognitive representations and operations rather than with the information-processing system that would support them. Finally, it is important to emphasize here that developmental fractionation does not depend on assumptions about how the cognitive system came to have its functional architecture at age X. Thus it is quite compatible with the possibility of radical reorganization prior to age X, and is neutral with regard to the question of whether the structure of the system is preformed. In the absence of any clear consensus about the nature of cognitive development, Sebastian Halliday and I came to the conclusion that the possibility of developmental fractionation should not be prejudged but should instead be regarded as an empirical question. Indeed, it seemed to us probable that fractionation of this sort might be possible for some aspects of cognitive function and not others. In the specific context of short-term memory abilities, we were encouraged by the existence of a number of apparent continuities during development, such as the existence of a limit on memory span (Dempster, 1981) and a recency effect in immediate free recall (Cole, Frankel, & Sharp, 1971; Thurm & Glanzer, 1971). We took these and other similarities as suggesting that the system does not undergo a truly radical reorganization over a fairly wide period of development. We were further encouraged by evidence for a discontinuity in the development of the particular control process of rehearsal, which is widely believed to emerge at around ages 5-6 (see, e.g., Kail, 1984). However, we were aware that most of this research had been undertaken with the goal of understanding

224

Hitch

developmental processes, rather than that of obtaining developmental fractionations of the adult short-term memory system. In our own recent research, summarized in Halliday and Hitch (1988), we have begun to explore the fractionation approach. We adopted the working memory model of adult function (Baddeley & Hitch, 1974; Baddeley, 1986) as our initial theoretical framework because of its success in accounting for a range of empirical phenomena using a reasonably small number of assumptions. Our aim was to see whether the performance of children at different ages could be explained in terms of this model minus some specific subsystems, transmission routes, or control processes. It will perhaps be helpful if I summarize the discussion briefly at this point. I started with the assumption that the functional architecture of the adult cognitive system comprises a set of separate but interconnected information-processing subsystems. Developmental fractionation may be said to occur when the cognitive system of the child at a particular age mirrors that of the adult except in certain highly specific respects, such as the absence of a particular subsystem or interconnection. This would be theoretically interesting, primarily as a way of testing and exploring models of the adult cognitive system. I suggested that the possibility of observing developmental fractionations should be regarded as an empirical matter, ultimately dependent on as yet unresolved questions about the nature of cognitive development. Short-term memory abilities seem a promising candidate for analysis in this way. In passing, I noted that obtaining a developmental fractionation would obviously say nothing about the nature of developmental processes, especially those whereby the cognitive system of the child came into being. However, such a fractionation could of course be a useful guide to the sorts of developmental processes that might be entertained for the period from childhood to adulthood.

9.3. The working memory model In common with many current approaches to adult short-term memory (e.g., Broadbent, 1984; Monsell, 1984; Barnard, 1985), the working memory model assumes a set of separate but interacting subsystems, each performing a particular function such as storing a special type of information or executing some specific process. The chief area of disagreement concerns the number and nature of these subsystems and their interconnections. According to the working memory model, short-term memory is seen as a multipurpose system made up of the limited-capacity subsystems shown in Figure 9.1. Baddeley (1986, this volume, chapter 2) has documented the empirical evidence supporting the model and has-described its detailed operation. For our present purposes it is necessary to draw attention to only some of its features. The subsystems of working memory comprise the "articulatory loop" and the "visuospatial scratch pad," which are both under the overall control of an attentional

Developmental fradionation

of working

memory

225

visuo-spatial scratch-pad

articulatory loop Figure 9.1. Diagram of the working memory model (from Baddeley, 1986, with permission of the author and the Royal Society London).

"central executive." The articulatory loop consists of a phonological store, in which memory traces decay with the passage of time, and an active process of subvocal rehearsal, which can be used to alter the contents of the store. When subjects rehearse in experiments on short-term memory for verbal materials, they are assumed to be using this process to refresh fading traces in the phonological store. A number of short-term memory phenomena are explained in terms of the operation of the articulatory loop. These include the effects of word length and phonemic similarity of the materials, irrelevant articulation (or "articulatory suppression"), and exposure to irrelevant speech. An important feature of the model is that access to the articulatory loop is thought to depend on the way materials are presented. Speech inputs are thought to feed directly and automatically into the phonological store, whereas visually presented verbal materials have first to be recoded by the optional control process of subvocalization. This distinction accounts for experimental evidence that adult subjects' tendency to use the articulatory loop for visually presented materials is highly susceptible to disruption by suppression and irrelevant speech. A related concept has emerged from studies of divided attention. On the basis of patterns of dual-task interference for different task combinations, McLeod and Posner (1984) have argued that the immediate verbatim repetition of heard words, unlike other stimulus-response mappings, involves a "privileged loop" (see also Friedrich, this volume, chapter 3). They suggest that the

226

Hitch

privileged loop is a separate structure and operates without drawing on generalpurpose attentional respurces. We note here that this concept is consistent with the idea that access to the articulatory loop depends on the way materials are presented. There is, however, an important difference. According to the current model of the articulatory loop, speech inputs gain automatic access only as far as the phonological store, with access to articulatory programmes (as in subvocal rehearsal) dependent on optional control processes. If, however, McLeod and Posner are correct, automatic access goes further than this model maintains. We shall see later that the question of how far speech inputs are automatically processed in the articulatory loop is of central concern for the interpretation of young children's memory abilities. The second subsystem of working memory, the visuospatial scratch pad, is thought to have a structure analogous to the articulatory loop. Thus it also consists of a specialpurpose store and a process for changing the contents of the store. In this case, however, the traces represent visuospatial information and the process corresponds to imagery or visualization. The role of the scratch-pad in short-term memory experiments has been much less thoroughly investigated than that of the articulatory loop (of which more later). The final component of the model, the limited-capacity central executive, is held to be responsible for control processes such as directing the flow of information between subsystems and is also thought to provide more abstract (i.e., nonphonological, nonvisuospatial) temporary information storage.

9.4. Applicability of the working memory model to children We saw earlier that developmental fractionation of a modular system is likely to be informative only when the system does not undergo any radical reorganization during the period of development of interest. With respect to the working memory model, data on the development of the articulatory loop suggest that this sybsystem behaves in a similar, though, as we shall see, interestingly different, way in children. Thus, studies of the word length effect show that it is present in children from age 4 upwards (the youngest it has proved possible to test) in the case of auditorily presented materials (Hulme, Thomson, Muir, & Lawrence, 1984), and from a slightly older age for visually presented words (Nicolson, 1981) or pictures (Hitch & Halliday, 1983). These experiments also show a surprisingly detailed correspondence between children and adults concerning the size of the word length effect. In adults, it is well known that the size of the effect varies in proportion to the rate at which subjects can articulate the materials (Baddeley, Thomson, & Buchanan, 1975). It turns out that the size of the word length effect in children is also proportional to articulation rate, and that the constant of proportionality is, so far as can be ascertained, the same. Figure 9.2 illustrates this with data from one of our own studies (Hitch, Halliday, & Littler, unpublished). According to the working memory model, this constant, corresponding to the slope of the linear

Developmental fmctionation of working memory

127

6 -

5 -

H • n

5 yr olds 8 yr olds 11 yr olds

o 4 -

31.0

1.5

2.0

2.5

3.0

articulation rate (items/sec)

Figure 9.2. Relation between memory span for auditorily presented words and the rate at which they can be articulated in 5-, 8- and 11-year-old children. The three data points for each age group correspond to span for one-, two-, and three-syllable words as indicated (data from Hitch, et al, 1989).

function that is plotted, reflects the decay rate for traces held in the phonological store. Hence, there is strong evidence for continuity in the development of this aspect of the articulatory loop. However, since the development of other subsystems of working memory has yet to be studied in equivalent detail, the applicability of the full model to children remains to be demonstrated.

9.5. An experimental dissociation in young children Early in our experiments on the word length effect in immediate recall, we had unexpected difficulty in obtaining the effect in young children. We knew that Hulme et al. (1984) had obtained the effect in children as young as 4, but we ourselves could not find an effect in 6-year-olds. We now believe that the crucial difference was that Hulme et al. had used spoken presentation and spoken recall, whereas we were presenting the materials as nameable pictures and asking for spoken recall. Indeed, had we been aware of a study by Allik and Siegel (1976), we might have been able to spot the importance of this discrepancy earlier. Allik and Siegel compared immediate memory for pictures of animals and objects whose names were either one or two syllables long. They found a word length effect in children aged 8 and 11, but not in children aged 4, 5, and 6. In an experiment in which we directly compared visual and auditory presentation in 6-year-olds (Hitch & Halliday, 1983), we confirmed that there was indeed an interaction between presentation mode and word length (see Figure 9.3). We also confirmed that when 10-year-old children are tested, there is a clear word length effect with both visual and auditory presentation (see Figure 9.3). The older

228

Hitch o.u -

6 yr olds 2.5-

2.0-

1.5-

Q •

spoken pictures

1.0-

1

2

3

word length (no syllables)

4 -

10 yr olds 3 -

E

2 Q spoken • pictures 1 1

2

3

word length (no syllables)

Figure 9.3. Effect of word length on recall following visual and auditory presentation in 6and 10-year-olds (data from study reported in Hitch & Halliday, 1983). children seem to behave like adults, who also show the word length effect for both auditory and visual presentation, whether the latter involves either nameable pictures (Schiano & Watkins, 1981) or written words (Baddeley et al., 1975). We have subsequently repeated our basic observation on younger children, and we have confirmed the difference between modalities. We do find, however, that there is sometimes a small visual word length effect in 6-year-olds and even, on one occasion, in 5-year-olds. Thus, although we are confident that the word length effect emerges more rapidly for spoken than visual materials, the precise chronology and its possible dependence on other factors remain to be determined. An analogous dissociation by modality in young children can be obtained in relation to the phonemic similarity effect. In an unpublished experiment carried out by Alma


of working memory

229

Table 9.1. Effects of phonemic similarity and overt naming at presentation on mean number of pictures recalled for 5- and 11-year-old children Materials

Control

Phonemically similar

5-year-olds (max = 3) Silent (n = 18) Aloud (n = 18)

2.16 2.66

2.03 2.06

11-year-olds (max = 5) Silent (n = 18) Aloud (n = 18)

3.78 3.77

2.67 2.33

Source: Hitch, Halliday, Schaafstal, and Heffernan, unpublished data. Schaafstal at Manchester, visual presentation involved showing a series of simple pictures, and auditory presentation was achieved by having the child label the pictures overtly at presentation. The materials had either phonologically similar names (e.g., rat, hat, cat) or dissimilar names (e.g., girl, horse, clock). Older children were presented with sequences of five items, younger children with three. Immediate, spoken serial recall was required in all conditions. The results are shown in Table 9.1. The older children show equal sensitivity to phonological similarity in the two presentation conditions; the younger children are sensitive only when there is an auditory input. The developmental emergence of the phonemic similarity effect for visually presented materials has been known ever since Conrad (1971) demonstrated that it appeared only after age 5. More recently, Hulme (1987) has shown that the effect is quite clearly present in children as young as 4 for auditorily presented materials, supporting the present suggestion of the importance of modality. Hulme (1987) also reported obtaining a phonemic similarity effect in young children using visual presentation, but in his visual presentation procedure the experimenter named the items for the child, thus providing spoken (and therefore also heard) input. The present claim for a developmental dissociation by input modality is of course critically dependent on the use of silent visual presentation. It seems that the working memory model can offer a ready interpretation of the effects associated with presentation method in young children. All that is needed is to suppose that the separate processes whereby auditory and visual materials gain access to the loop develop at different rates, such that the one that is obligatory in adults emerges before the one that is under optional control (cf. Hitch & Halliday, 1983). If so, then there should be a point in development where spoken materials are held in the articulatory loop while visually presented materials are stored elsewhere in the system, as indeed is the case. We can therefore regard the present contrast between older and

230

Hitch

younger children as an example of a developmental fractionation, according to which younger children operate as if a particular component of working memory is not fully functional. The deficiency is associated not with the absence of an entire subsystem but with the availability of a specific process within the articulatory loop. We turn now to consider young children's use of the articulatory loop in greater detail.

9.6. The articulatory loop in young children: verbal rehearsal Interpreting young children's immediate memory for spoken materials in terms of the articulatory loop raises an important further question. According to the working memory model, the effects of phonemic similarity and word length are explained in different ways. This is motivated by some of the adult data on the sensitivity of these effects to articulatory suppression (Baddeley, Lewis, & Vallar, 1984). Specifically, phonemic similarity effects are thought to arise from problems of discriminating confusable traces within the phonological store, whereas word length effects are attributed to the slower rate at which traces of long words can be refreshed by the control process of subvocal rehearsal. According to a strict application of the working memory model, therefore, the presence of a word length effect in young children implies that they actively rehearse spoken materials. However, according to standard accounts of the development of short-term memory strategies, rehearsal emerges only rather later in development (see, e.g., Kail, 1984). It is evidently important to try to resolve this apparent discrepancy. Interestingly, most of the evidence on the development of rehearsal strategies in young children derives from studies that have used visually presented materials (see, e.g., Hagen & Stanovich, 1977, p. 96). It could be, therefore, that the standard account has been accepted because insufficient attention has been paid to the role of presentation modality. We have consequently begun to look for evidence that younger children do actively rehearse spoken inputs in short-term memory tasks. If they do no rehearse, and yet show a word length effect, it will be necessary to revise an important aspect of the working memory model. If, on the contrary, they do rehearse, then conventional views of the development of rehearsal are incorrect. Before summarizing some of the evidence we have obtained, it is necessary to emphasize that our investigations are still in progress and will be reported more formally when they are completed. We have so far looked at three aspects of young children's performance that might indicate whether they actively rehearse auditorily presented materials: effects of rate of presentation, serial position, and articulatory suppression. Given that there is trace decay in the phonological store, one might expect faster presentation rates to lead to better recall. However, slower rates have the advantage of allowing more time for rehearsal, which would tend to offset any effects of decay. In a

Developmental fradionation of working memory

231

Table 9.2. Effect of presentation rate on mean number of auditorily presented words recalled for 5- and 11-year-old children Presentation rate

5-year-olds (max = 5) (n = 18) 11-year-olds (max = 6) (n = 18)

Slow (1 word/1.5 sec)

Fast (1 word/0.75 sec)

1.42

1.82

2.77

2.93

Source: Hitch, Halliday, and Littler, unpublished data.

study of adult subjects, Baddeley and Lewis (1984, Experiment 1) obtained results consistent with this analysis: Fast rates of presentation were advantageous only when subjects were prevented from rehearsal by articulatory suppression. In an experiment carried out by Janet Littler, Sebastian Halliday, and myself, we reasoned that if younger children were not rehearsing spoken materials, they should show clear effects of trace decay and therefore perform better at fast presentation rates. Older children, on the other hand, should be able to offset the effects of decay by rehearsal, and should therefore show a smaller or even opposite effect of rate. Groups of 5- and 11-year-olds ( N = 1 8 ) performed immediate spoken recall of auditory word lists presented at a fast rate of one item every 0.75 sec or a slow rate of one item every 1.5 sec. The young children were presented with lists of five items and the older children six items in an attempt to avoid floor and ceiling effects. They were required to recall each list in strict serial order starting from the beginning and guessing when necessary. Credit was given for each item recalled in the correct serial position. The results (see Table 9.2) confirmed our prediction in that the young children performed better at the faster rate and the older children showed only a nonsignificant difference. We have subsequently confirmed these age differences in a number of experiments, one of which is described later in this section. However, although we thought initially that our experiment would be definitive, this is unfortunately not so. Our results do support the weaker claim that young children are not rehearsing as effectively as older ones; however, they do not show that such children do not actively rehearse at all. The serial position curves reinforce the need for cautious interpretation (see Figure 9.4). It is clear that there are strong primacy effects in both age groups of children. It is interesting to note the contrast with serial position curves for visually presented materials. Figure 9.5 presents data taken from two conditions of one of our earlier studies (Hitch, Halliday, Schaafstal & Schraagen, 1988, Experiment 2), in which children were presented with series of line drawings of common objects at a rate of one

232

Hitch 5 yr olds 80 -

S

Q fast • slow

\

60-

vs.

z J

40 -

SL 20 0 2

3

4

serial position

11 yrolds 80 |

60-

fc

40 -

\

V

Q.

20 -

Q fast • slow

0 1

2

3

4

5

6

serial position

Figure 9.4. Serial position curves for 5- and 11-year-olds' recall of auditorily presented words (unpublished data).

item every 2 sec and then recalled them immediately in the manner described earlier. It is clear that older children show a strong primacy effect for these materials but younger children do not. Given that the lack of primacy with visual presentation goes with the absence of rehearsal (as revealed by the absence of word length and phonemic similarity effects described earlier), one might interpret the presence of primacy in the case of auditory presentation as a sign that some rehearsal is taking place. In a further study carried out with Janet Littler, we examined the effect of articulatory suppression on auditory digit span in 5- and 11-year-olds. Suppression is thought to disrupt active rehearsal but not the entry of spoken material into the phonological store in adult subjects. We reasoned that if younger children are not rehearsing, they should be insensitive to this type of interference. Indeed, in a previous study (see Hitch &


233

uu 80 60 -

2.

>•

•

40 Q •

20 -

5yrolds 10yrolds

01

2

3

4

5

serial position

Figure 9.5. Serial position curves for 5- and 10-year-olds' recall of lists of three and five pictures respectively (data from Hitch et al. 1988).

Table 9.3. Effects of presentation rate and articulatory suppression on auditory digit span for 5- and 11-year olds Presentation rate Slow (1/2 sec)

Fast (2/sec)

3.81 2.74

4.74 3.50

5.70 4.46

6.09 5.02

5-year-olds

Control (w = 18) Suppress (n = 18) 11-year-olds

Control (n = 18) Suppress (n = 18)

Source: Hitch, Halliday, and Littler, unpublished data.

Halliday, 1983), we found that with visual presentation, young children were unaffected by suppression, in contrast to older children. In the present experiment, children suppressed during auditory presentation of the digits by repeating the phrase happy birthday and then recalled orally. The slow presentation rate was one digit every 2 sec; the fast rate was two digits every second. The results are shown in Table 9.3. Suppression had a clear detrimental effect on span in both the 5-year-olds and the 11year-olds. Although we repeated our earlier finding of a differential effect of rate of presentation in these two age groups, there was no suggestion that the effect of suppression was smaller in the younger children or at faster presentation rates. Indeed, the trends were in the opposite direction.

234

Hitch

The results of our investigations of 5-year-old children's immediate recall of spoken sequences can be summarized as follows: 1. Recall is disrupted by articulatory suppression. 2. The serial position curves show a clear primacy effect. 3. Recall is better at faster presentation rates.

All these effects are also found in older children, except that older children show a reduced effect of rate of presentation. The simplest interpretation, therefore, is that 5year-olds do not behave radically differently from older children. This would imply that 5-year-olds engage in the active rehearsal of spoken materials. The effects of presentation rate could then be interpreted as suggesting that their rehearsal is merely less effective in offsetting the effects of trace decay than that of older children. This conclusion would, of course, contradict the standard view that active rehearsal develops at a somewhat later age. It would, however, imply that the word length and phonemic similarity effects in young children can be adequately explained in terms of the articulatory loop subsystem of the adult working memory model. Nevertheless, there are good reasons for adopting a somewhat cautious attitude to this interpretation. Specifically, all our evidence about rehearsal is indirect and is open to alternative interpretations. Thus, primacy could be caused by factors such as the buildup of proactive interference or even simple %trace decay (if, as might have been the case, children recalled items more slowly than they were presented). Furthermore, articulatory suppression might disrupt recall through dividing attention rather than disrupting rehearsal itself. Alternative interpretations such as these evidently require careful checking. We are currently tackling this question by examining other methods of assessing rehearsal such as monitoring lip movements and looking at the effects of an unfilled postlist delay on recall. For these reasons, we must regard an interpretation of our 5-year-olds' data in terms of the active rehearsal of material in the articulatory loop as only tentative. Indeed, it is extraordinarily difficult to discard the strong intuition that such young children do not in fact rehearse actively in a similar way to older children. It is therefore of interest to think through what this alternative possibility might entail. Earlier on, I drew attention to McLeod and Posner's (1984) concept of the privileged loop whereby speech inputs gain automatic access to corresponding articulatory motor programmes. I contrasted this with the working memory model, according to which speech inputs gain automatic access only to the phonological store. To carry the discussion further, it will be useful to make a distinction between the automatic triggering of articulatory motor programmes by speech inputs and the voluntary repetition of such programmes as in subvocal rehearsal. This would allow us to consider revising the concept of the articulatory loop such that its passive component comprises both a phonological store and a closely linked set of articulatory motor programmes, which in turn feed an active subvocal rehearsal process. Five-year-olds' immediate recall


235

of spoken inputs could then be interpreted as being based on the privileged loop, the redefined passive component of the articulatory loop. Subsequent developmental change in the articulatory loop would then be explained in terms of the acquisition of active rehearsal strategies. On this interpretation, the passive component of the articulatory loop would be responsible for the effects of word length, phonemic similarity, and suppression on recall, whereas according to the current version of the working memory model, the passive phonological store is responsible for only the phonemic similarity effect of these three. We can see that this would amount to quite an important change within the model. It certainly seems plausible to suppose that something like the privileged loop would be present from a relatively early stage in development, since the ability to repeat heard inputs readily and without cognitive effort would seem to be useful for the acquisition of speech and language (see Halliday & Hitch, 1988). It is clearly unwise, however, to speculate further in the absence of more definitive evidence about the degree to which young children engage in active rehearsal. My main point in going through this discussion in some detail has been to demonstrate as forcefully as possible another way in which results obtained from children can assist in the fractionation of the adult system, in this case with the generation of fresh questions and insights about the nature of the articulatory loop.

9.7. Visual working memory in young children Our discussion so far has been confined to just one aspect of the dissociation between memory for spoken and visually presented materials in young children, namely, the mechanisms whereby spoken information is stored. We now consider how these children store information about visually presented items. This is especially interesting because here there is a sharp contrast between younger and older children. If one uses the working memory model to generate hypotheses, and if one assumes that the articulatory loop is not involved, then storage must involve either the visuospatial scratch pad or the central executive, or perhaps both.-We opted to begin our investigations by testing for the use of visually based storage. The experiments are more fully reported elsewhere (Hitch et al, 1988), so only some of the more important findings will be summarized here. One problem in tackling this question is that the visuospatial scratch pad is less well understood than the articulatory loop and is consequently less well specified in the working memory model. In addition, we found that manipulations thought to affect the scratch pad in adults were not practicable with young children. We therefore adopted an exploratory approach, and attempted to test the more general hypothesis that young children's memory for visually presented materials involves some form of specifically visual storage. The more precise question of how any such storage might be related to

236

Hitch CONTROL

(pig)

(cake) VISUALLY SIMILAR

(pen)

N^>

(fork)

LONG NAMES

(umbrella)

(kangaroo)

Figure 9.6. Examples of materials used in experiment on effects of visual similarity and word length on recall (from Hitch et al., 1988, with permission of the Psychonomic Society, Inc.).

the concept of the visuospatial scratch pad in adult working memory was left in abeyance, pending an answer to the first. In our first experiment we looked at the effect of visual similarity of pictures presented to 5- and 10-year-old children for immediate spoken recall. We argued, by analogy with the phonemic similarity effect, that traces of visually similar items should be harder to discriminate and thus less well recalled. We used three sets of materials, as shown in Figure 9.6. The visually similar drawings were of elongated objects depicted in the same same oblique orientation and had one-syllable names. The control set also had one-syllable names, but differed in shape and appearance. In the third set, the drawings were also visually dissimilar, but had three-syllable names. On each trial, the child was shown a series of drawings at a rate of one item every 2 sec. Young children were shown series of three items, older children five. Recall was spoken, in the order of presentation, with the child saying "Blank" or "Don't know" if a particular item could not be recalled. The dependent variable was the number of items recalled in the correct serial position, and the results are shown in Table 9.4. As expected, the younger children's recall was impaired by visual similarity, and, somewhat to our surprise, there was also a small but significant effect of word length.


of working memory

237

Table 9.4. Effects of visual similarity and word length on 5- and 10-year-old children's memory for simple pictures Materials

5-year-olds (max = 3) (n = 18) 10-year-olds (max = 5) (n = 18)

Control

Visually similar

Long names

2.07

1.40

1.76

3.68

3.61

2.81

Source: Hitch et al. (1988).

The older children were unaffected by visual similarity and showed a highly reliable effect of word length. Other studies have also found that young children's immediate recall is sensitive to visual similarity (Brown, 1977; Hayes & Schulze, 1977). Hence our main conclusion is that younger children use some form of visual storage for visually presented materials. We regard the presence of a small word length effect in our younger group as suggesting that the children in this sample were just beginning to acquire the tendency to recode visual materials into spoken form. Our data also suggest that older children may come to rely exclusively on the articulatory loop, since there was no sign that they were using visual storage. However, subsequent experiments using more sensitive techniques have shown that older children do continue to make some use of visual storage, in parallel with their very much greater reliance on the articulatory loop (Hitch, Woodin, & Baker, 1989). In a second type of experiment we have examined the effects of a visual interference task on recall. In one such study (Hitch et al., 1988, Experiment 4), 5-year-old children were presented with a series of either three drawings or three spoken words. We used a backwards spoken recall procedure whereby the child recalls the last item first, then the second from last, and so on. This method of recall gives rise to a large recency effect when there is no interfering activity prior to the start of recall. In the visual interference task that we used, children were shown three additional drawings, two of which were identical and had to be matched by being placed on top of one another. This task took 4 sec to complete. The control condition consisted of a 4-sec unfilled delay prior to the start of recall. The design was therefore a 2 x 2 factorial in which the factors were modality of presentation and the presence or absence of visual interference. The results are shown in Figure 9.7. Visual and auditory presentation gave rise to strong, roughly equal recency effects under control conditions, but visual interference had a selective effect, reducing recency only for the visually presented sequences. We interpret this as further evidence that 5-year-olds use a visual store, which is subject to overwriting by subsequent inputs in the same modality.

238

Hitch 100 drawing- control

90

#

d r a w i n g - v i s u a l Rl

A s p o k e n - control

80

•

s p o k e n - v i s u a l Rl

rQp1

A

T-GP2/'/

J / /o

70

-

60

o O 50

40 0> Q.

30

20

10k

Serial

Position

Figure 9.7. Effect of a visual task interpolated before recall of visual and auditorily presented sequences in 5-year-olds (data from Hitch et al., 1988).

In another experiment involving the same general procedure, we compared the effects of visual and auditory—verbal interference tasks in 5- and 11-year-old children (see Hitch et al., 1988, Experiment 3). In these tasks, children classified items as either animate or inanimate, with, in one case, auditory presentation of the name of the item and a spoken response, and, in the other, presentation of a picture of the item and a manual response. The results were in accordance with our expectations. Thus, with visually presented memory materials, younger children were more sensitive to visual than auditory-verbal interference, but older children showed the opposite effect. With


239

spoken presentation of the memory items, on the other hand, both age groups were more sensitive to auditory-verbal interference. These results supply further confirmation of young children's reliance on a visual store for visual materials, and suggest that older children are more reliant on auditory-verbal storage. For auditory materials, however, young and older children alike appear to use the auditory-verbal store. To summarize, we have consistent and fairly convincing evidence in support of the hypothesis that young children's immediate memory for visual materials involves a specifically visual storage system. The key findings are the disruption of recall by visual similarity of the materials and by a visual interference task. Unfortunately, we cannot yet say how this system relates to the concept of the visuospatial scratch pad as described in the adult working memory model. It would of course be interesting to test the idea that our tasks tap the same subsystem, albeit in an immature form, and to try to trace its development. For our present purposes, however, the main point is to confirm that young children use some sort of visual short-term store for visual materials, and to contrast this tendency with the way they remember spoken materials and with the way older children go about remembering the same visual materials.

9.8. Developmental fractionation of working memory Our results are consistent with a very simple theoretical interpretation that assumes (a) that the basic structure of the working memory system remains the same from as early as age 5 upwards (results for 10- and 11-year-olds being effectively the same as in adults) and (b) that young children have a selective deficiency in the recoding process, whereby nameable visual materials are normally recoded and stored verbally in the articulatory loop. We therefore regard our data as providing a developmental fractionation of working memory, broadly analogous to the more familiar neuropsychological fractionations that have been reported. As has been seen, a number of important questions have yet to be fully explored. Although these concern detailed aspects of interpretation, they could have quite a strong bearing on how the working memory system is modelled. The first of them concerns the nature of the articulatory loop and in particular the question of whether the auditory word length effect is critically dependent on the control process of active rehearsal. Our data on young children are consistent with this aspect of the current working memory model, but we remain sceptical as to whether the word length effect they display is a consequence of the type of voluntary, strategic rehearsal that the model implies. The second question concerns the nature of the visuospatial scratch pad. Our data confirm the involvement of a visual store in young children, but so far we have not been able to link its properties with those in the working memory model. Thus, although our results make a very clear case for marked and important developmental differences in children's short-term memory performance, the usefulness of our present

240

Hitch

interpretation as a straightforward fractionation of the adult model must await further empirical investigation.

9.9. Some links with neuropsychological fractionation Developmental and neuropsychological attempts to fractionate working memory can from one point of view be regarded as parallel lines of attack in the task of understanding the normal adult system. Eventually, evidence from these two sources should converge on a common theoretical interpretation. It should be understood, however, that the two types of fractionation differ substantially. Neuropsychological fractionation typically involves investigating patients with progressively more specific disorders of function on the assumption that this will reveal an increasingly accurate view of the underlying modular structure. This is a more empirical procedure than the theoretically driven developmental fractionation approach I have described here, where the starting point was a model of intact adult function and the goal was to see whether developmental data could be explained in terms of this model minus some components. One might, as a consequence, observe children whose state bears a qualitative resemblance to some specific neuropsychological state. However, this would arise fortuitously through a shared subtractive logic rather than because of a more general identity between the methods themselves. Even where a developmental state does appear to mirror a neuropsychological state fairly closely, it would be inappropriate to expect to draw simple inferences from direct, point-by-point comparisons between child and patient. There is clearly no basis for supposing that a deficit to a specific subsystem in a previously normal adult would lead to the same pattern of performance as in a normal child in whom this subsystem has yet to develop. For example, even if the functional architecture of working memory in a child was identical to that of a particular type of patient, we would expect to see differences due to learning (which would influence the availability of normal strategies in the child and compensatory strategies in the patient). There would in addition be differences in parameters of the system, such as rehearsal rate (Hitch & Halliday, 1983), speed of processing (Dempster, 1981), or "operational efficiency" (Case, Kurland, & Goldberg, 1982). For these reasons, I believe that relating neuropsychological and developmental fractionations of function requires considerable caution. At the theoretical level, the best way to do this seems to be through the links between each type of fractionation and common models of normal adult function. In the case of working memory, the way the two types of evidence can show theoretical convergence can be illustrated with reference to the performance of patients with conduction aphasia who show a repetition deficit. Such patients have been interpreted as suffering from an impairment to auditory-verbal short-term memory (see Warrington, Logue, & Pratt, 1971), which


241

within the working memory model would correspond to a deficit associated with some aspect of the articulatory loop. Patients of this sort typically show a reduced auditory digit span, but a rather higher span when the materials are visually presented/This result can be readily interpreted within the working memory model through its assumption of separate subsystems for visual and auditory-verbal short-term storage. If so, such patients should show evidence of greater reliance on the visual subsystem than do normal adults. Unfortunately, this prediction does not seem to have been thoroughly explored, despite its importance for understanding the patients' residual abilities. However, one patient who was investigated in this way could be shown to have a tendency to make visually based confusions in immediate recall (Warrington & Shallice, 1972). Since the evidence from young children suggests that visual working memory is indeed sensitive to visual similarity, it seems reasonable to suppose that patients and young children are accessing the same subsystem. One way of exploring this further would be to see whether patients respond in a similar way to the experimental manipulations that have been used to investigate visual working memory in children. A second illustration of the convergence of theoretical interest is provided by the investigation of PV, a patient with defective auditory-verbal short-term memory reported by Vallar and Baddeley (1984). PV showed the usual pattern of a markedly reduced auditory memory span with relative sparing of visual span (Basso, Spinnler, Vallar, & Zanobio, 1982). Her visual span was unaffected by either phonemic similarity or articulatory suppression, which we note is reminiscent of how 5-year-old normal children behave. Indeed, Vallar and Baddeley offer a similar theoretical interpretation, namely, that PV was not making use of the articulatory loop. It would seem, however, that the deficit responsible for her failure must be distinguished from the cause in young children. Thus, although her auditory span was sensitive to phonemic similarity, it was unaffected by word length. As we have seen, 5-year-olds are sensitive to both phonemic similarity and word length when presentation is auditory. Perhaps the simplest interpretation of this difference is that PV has a deficit within the loop that affects its use regardless of whether presentation is visual or auditory, whereas normal children have an intact articulatory loop that is accessed by auditory inputs but that they have yet to learn to employ in remembering visual materials. Indeed, Vallar and Baddeley suggest that PV has a deficit in the phonological store and that consequently the control process of subvocal rehearsal is no longer an effective mnemonic strategy. An alternative interpretation (suggested by Meudell, 1986) is that PV's deficit is not in the phonological store itself but rather in the pathway from this store to the process of articulation, that is, in the privileged loop of McLeod and Posner (1984). This would give a more economical account of the sensitivity of her auditory span to phonemic similarity but not to word length, by making one assumption rather than two. It will not have escaped attention that exactly the same theoretical point has arisen in considering how to account for the origin of the word length effect in our experiments on young

242

Hitch

children's recall of spoken sequences, where I contrasted active, strategic rehearsal with the automatic feedthrough of spoken inputs to output. I should, however, caution that it remains to be seen whether Meudell's interpretation can account for the full range of PV's deficits (see Shallice and Vallar, this volume, chapter 1). We have now identified two instances in which common issues of theoretical interpretation arise in both the neurospsychological and the developmental fractionation of working memory. Such a close convergence of interest in two very separate domains of research is, I think, encouraging and clearly merits further exploration. Having seen how the two approaches can focus on the same theoretical issues, it is useful to consider briefly some of their methodological advantages and disadvantages. For example, a key problem in the neuropsychological fractionation of function is how to rule out the possibility that what appears to be a selective cognitive deficit is actually a consequence of lowered general-purpose processing resources (see, e.g., Shallice, 1979b). Thus, a patient might perform poorly on a group of tasks because they all place high demands on a lowered general resource and not because they each involve a specialized subsystem that is selectively impaired. It is commonly assumed amongst neuropsychologists that this problem of interpretation can be overcome by demonstrating "double dissociations" of cognitive function. According to one definition, a double dissociation occurs when one type of patient can be shown to be impaired on one group of tasks relative to another, while a complementary type of patient displays the opposite pattern of performance (see, e.g., Colheart, 1985). It is often concluded that a double dissociation of this sort demonstrates damage to separate subsystems in the two cases. Shallice (1988) has shown, however, that such a finding could result from lowered general-purpose resources, since the (generally unknown) functions whereby performance is related to the availability of such resources might differ for the two groups of tasks. Shallice (1988) argues for a criterion of double dissociation that is not subject to this criticism. He shows that resource artefacts can be avoided if a double dissociation is defined in terms of a situation whereby one type of patient is clearly superior to another on one group of tasks, but inferior on a second group. Shallice demonstrates, however, that even with this more satisfactory definition, neuropsychological double dissociations must be interpreted with considerable caution. If we turn now to consider the developmental fractionation of function, we immediately become aware that the kind of fractionation I have discussed here is, in neuropsychological terms, a single rather than a double dissociation. The question therefore arises whether the problem of resource artefacts assumes a similar importance in developmental fractionation. The answer is probably not, despite the possibility that lowered general resources may, for example, contribute to young children's inability to use the articulatory loop to store the names of visual materials. It could well be the general information-processing load of combining covert picture naming with active


243

rehearsal that prevents execution of the rehearsal strategy (see Hitch & Halliday, 1983). It would be exceedingly difficult, however, to use a general resources argument to account for the full pattern of results we have obtained, for example, our evidence that young children's memory for pictures is sensitive to visual similarity and to visual modality-specific interference (Hitch et al., 1988). In more general terms, therefore, we can appreciate how resource artefacts are probably less of a danger in the theoretically driven developmental fractionation procedure than in the more empirical neuropsychological fractionation of function. Developmental fractionation is also less vulnerable to artefacts associated with subject sampling. These issues are of central concern in cognitive neuropsychology based on the single case study approach (see Shallice, 1979b, 1988). It may, for example, be difficult to know when two patients should be classified in the same or a different category, especially when symptoms that are not of immediate concern to the investigator do not match. In developmental fractionation, however, subject sampling, and replicating and generalizing results present no serious problems, since the theories concerned apply to all normal children. These are obviously matters of considerable scientific importance. These methodological differences between developmental and neuropsychological fractionation suggest that double dissociations of function will not play a similar role in the two sorts of enterprise. This is perhaps fortunate, since the existence of constraints on the sequence of normal development means that we would not ordinarily expect to observe double dissociations when comparing children with adults. Of course, if there were circumstances in which developmental double dissociations could be demonstrated, these would obviously be of considerable theoretical interest. However, we must accept that we cannot hope to mirror, in our studies of normal development, the wide range of highly selective pathological deficits that have proved so informative to our understanding of normal function. This leads me to reemphasize the point that neuropsychological and developmental fractionation should be regarded as complementary research strategies, with quite different strengths.

9.10. Applicability of developmental fractionation It is important to appreciate that how generally applicable developmental fractionation will prove to be remains to be seen. We do not yet know, for example, whether this methodology will enable the working memory system to be dissected along other dimensions besides those discussed here. We may perhaps be guardedly optimistic about this, since the functional architecture of the system in young children does seem to bear a close similarity to the adult model. It would be dangerous, however, to assume that the same is necessarily true of other aspects of cognitive function. In the case of reading, for example, it seems intuitively far less plausible to suppose that it would be

244

Hitch

profitable to analyse young beginning readers' performance in terms of models of skilled adult performance. How, then, can we decide whether developmental fractionation is likely to be applicable to any given aspect of cognitive function? Earlier in this chapter I discussed some of the relevant theoretical issues, although without going into great detail. The trouble with trying to extend that kind of analysis is that it rapidly becomes too hypothetical given how little we know about the nature of developmental change. Therefore, I would like to propose instead a more pragmatic approach and identify three criteria on the basis of the work described here. These are: (a) that we have a model of the adult system that specifies its subsystems and their interconnections; (b) that we have experimental methods that relate to the operation of individual subsystems that are applicable to children; (c) that there are similarities between several aspects of adult and child performance in the area of interest. These conditions could of course be relaxed somewhat if one wanted to adopt a more empirically based method of developmental fractionation than that described here. A final question is whether developmental fractionation can tell us anything about disorders in the development of cognitive function. This is clearly too vast a topic to discuss properly here. Nevertheless, we can say that in those areas where developmental fractionation is applicable, it will obviously help us to characterize the changes that take place in normal development. This should be of great value in detecting and assessing cases of abnormal development. To do this it might be fruitful to combine the methodologies of developmental and neuropsychological fractionation. I should reemphasize, however, that developmental fractionation is not a theory of developmental change. Thus, while it may help us to assess the consequences of abnormal development, it can tell us nothing about its causes.

9.11. Conclusion I noted early on that both developmental fractionation and neuropsychological fractionation of cognitive function involve very similar general assumptions about the nature of the human information-processing system as a collection of functionally independent subsystems. Subsequently we examined some important differences between these two approaches that give them different strengths. In the investigation of working memory, where a model of normal adult function is available, we showed that developmental fractionation can be a useful method for analysing age differences and for pointing to directions in which the model itself must be further developed and elaborated. We showed further how concepts and methods used in the analysis of working memory in children are applicable to the analysis of memory disorders in specific types of adult patients. These links clearly merit further exploration. My tentative conclusions, therefore, are that developmental fractionation is certainly


245

feasible and that it has the capacity for generating fresh insights into the nature of both normal and abnormal function.

References Allik, J. P., & Siegel, A. W. (1976). The use of the cumulative rehearsal strategy; A developmental study. Journal of Experimental Child Psychology, 21, 316—327. Atkinson, R. C, & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89-195). New York: Academic Press. Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47-90). New York: Academic Press. Baddeley, A. D v & Lewis, V. (1984). When does rapid presentation enhance digit span? Bulletin of the Psychonomic Society, 22, 403—405. Baddeley, A. D., Lewis, V. J., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly Journal of Experimental Psychology, 36A, 233-252. Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575-589. Baldwin, J. M. (1894). Mental development in the child and the race. New York: Macmillan. Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. W. Ellis (Ed.), Progress in the Psychology of Language (Vol. 2, pp. 197-258). Hillsdale, NJ: Erlbaum. Basso, A., Spinnler, H., Vallar, G., & Zanobio, E. (1982). Left hemisphere damage and selective impairment of auditory-verbal short-term memory. Neuropsychologia, 20, 263-274. Broadbent, D. E. (1984). The Maltese Cross: A new simplistic model for memory. Behavioral and Brain Sciences, 7, 55-94. Brown, R. M. (1977). An examination of visual and verbal coding processes in preschool children. Child Development, 48, 38-45. Case, R. M., Kurland, M. D., & Goldberg, J. (1982). Operational efficiency and the growth of short-term memory span. Journal of Experimental Child Psychology, 72, 371-404. Cole, M., Frankel, F., & Sharp, D. (1971). Development of free recall in children. Developmental Psychology, 4, 109-123. Coltheart, M. (1985). Cognitive psychology and the study of reading. In M. Posner & O. S. M. Marin (Eds.), Attention and performance XI (pp. 3-37). Hillsdale, NJ: Erlbaum. Conrad, R. (1971). The chronology of the development of covert speech in children Developmental Psychology, 5, 398-405. Dempster, F. N. (1981). Memory span: Sources of individual and developmental differences. Psychological Bulletin, 89, 63-100. Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press. Hagen, J. W., & Stanovich, K. G. (1977). Memory: Strategies of acquisition. In R. V. Kail & J. W. Hagen (Eds.), Perspectives on the development of memory and cognition (pp. 89-111). Hillsdale, NJ: Erlbaum. Halliday, M. S., & Hitch, G. J. (1988). Developmental applications of working memory. In G. Claxton (Ed.), Growth points in cognition, (pp. 193-222). London: Routledge & Kegan Paul. Hayes, D. S., & Schulze, S. A. (1977). Visual encoding in preschoolers' serial retention. Child Development, 48, 1066-1070. Hitch, G. J., & Halliday, M. S. (1983). Working memory in children. Philosophical Transactions of the Royal Society London Series B, 302, 325-340. Hitch, G.J., Halliday, M. S., & Littler, J. E. (1989). Auditory-verbal memory span in children: The role of articulation rate and item identification time. Manuscript submitted for publication.

246

Hitch

Hitch, G. J., Halliday, M. S., Schaafstal, A. M., & Schraagen, J. M. C. (1988). Visual working memory in young children. Memory and Cognition, 16, 120-132. Hitch, G. J., Woodin, M. E., & Baker, S. L. (1989). Visual and phonological components of working memory in children. Memory and. Cognition, 17, 175-185. Hulme, C. (1987). The effects of acoustic similarity on memory in children: A comparison between visual and auditory presentation. Applied Cognitive Psychology, 1, 45—52. Hulme, G, Thomson, N., Muir, C, & Lawrence, A. L. (1984). Speech rate and the development of short-term memory. Journal of Experimental Child Psychology, 38, 241-253. Kail, R. V. (1984). The development of memory in children (2nd ed.). New York: Freeman. McLeod, P., & Posner, M. I. (1984). Privileged loops from percept to act. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 55-66). Hillsdale, NJ: Erlbaum. Meudell, P. (1986). Auditory-verbal short-term memory. Department of Psychology, University of Manchester. Unpublished manuscript. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view - a tutorial review. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 327-350). Hillsdale, NJ: Erlbaum. Morton, J., & Patterson, K. E. (1980). A new attempt at an interpretation, or, an attempt at a new interpretation. In M. Coltheart, K. Patterson, & J. C. Marshall (Eds.), Deep dyslexia. London: Routledge & Kegan Paul. Nicolson, R. (1981). The relation between memory span and processing speed. In M. P. Friedman, J. P. Das, & N. O'Connor (Eds.), Intelligence and learning (pp. 179-183). New York: Plenum. Schiano, D. J., & Watkins, M. J. (1981). Speech-like coding of pictures in short-term memory. Memory and Cognition, 9, 110—114..

Shallice, T. (1979a). Neuropsychological research and the fractionation of memory systems. In L. G. Nilsson (Ed.), Perspectives on memory research. Hillsdale, NJ: Erlbaum. Shallice, T. (1979b). Case study approach in neuropsychological research. Journal of Clinical Neuropsychology, 1, 183-211. Shallice, T. (1984). More functionally isolable subsystems but fewer "modules"? Cognition, 17, 243-252. Shallice, T. (1988). From neuropsychology to mental structures. Cambridge: Cambridge University Press. Thurm, A. T., & Glanzer, M. (1971). Free recall in children: Long-term store vs. short-term store. Psychonomic Science, 23, 175—176.

Vallar, G., & Baddeley, A. D. (1984). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior,

23, 151-161. Warrington, E. K., Logue, V., & Pratt, R. T. C (1971). The anatomical localization of selective impairment of auditory-verbal short-term memory. Neuropsychologia, 9, 377-387. Warrington, E. K., & Shallice, T. (1982). Neuropsychological evidence of visual storage in short-term memory tasks. Quarterly Journal of Experimental Psychology, 24, 30—40.

10. Adult age differences in working memory FERGUS I. M. CRAIK, ROBIN G. MORRIS, AND MARY L. GICK

10.1. Introduction The notion that some degree of short-term memory impairment is typically found in healthy older people has been current for the last 30 years or so. Welford (1958) surveyed the results of several dual-task experiments and proposed that many of the deficits associated with the normal aging process - in memory, learning, reasoning, and perceptual-motor tasks - may have their basis in the reduced efficiency of short-term memory; in particular, it seemed that both the capacity of this memory store, and its ability to resist the interfering effects of other activities, declined in the course of normal aging. However, by the time that Craik (1977) reviewed the literature on age differences in memory, more techniques to measure short-term (or primary) memory were available, and it appeared that Welford's suggestion was either faulty or too general. Craik pointed out that age differences were minimal in such measures as digit span, the recency effect in free recall, and the slope of the Brown-Peterson function. A possible resolution of the apparent conflict is that age differences do not appear (or are slight) in areas where the task calls for relatively passive storage of some small amount of material and then for its retrieval in much the same form, whereas age differences are substantial when the subject must manipulate the material held, or actively rehearse one set of material while simultaneously perceiving or responding to further material (Craik, 1977). This latter characterization is very similar to the concept of a "general working memory" as described by Baddeley (1986; this volume, chapter 2). It might therefore be suggested that the normal aging process has little effect on primary memory (Waugh & Norman, 1965) but does have a substantially detrimental effect on working memory. This was the view expressed by Craik and Rabinowitz (1984), although they also caution against thinking of "primary memory" and "working memory" as two distinct The work reported in this chapter was supported by a grant from the Natural Sciences and Engineering Research Council of Canada to the first author. We are grateful to Elizabeth Kelly, Maureen Kerr, Linda Lindsay, Lorna Morris, Lily Moysiuk, and Caroline Panabaker for their help in running and analysing the experiments.

247

248

Craik, Morris, and Gick

mechanisms; rather, the two sets of tasks should be thought of as lying on a continuum of processing complexity, such that primary memory tasks require little translation or manipulation of the material between input and ouput, whereas working memory tasks demand considerable reorganization of the material held. This modified view thus agrees well with Welford's original suggestion, although with the qualification that large age differences will emerge only with the working memory type of short-term memory task. In his later writings, Welford (1980) also appears to endorse this view. The present approach also suggests that as a primary memory task is transformed into a working memory task by requiring manipulations of the material held, age differences should be amplified. This prediction was confirmed in an experiment by Gick and Craik (reported by Craik, 1986), in which short lists of words or digits were given to young and elderly subjects for immediate serial recall. In one version of the task, the digits were recalled in their original order of presentation; it was therefore a straightforward digit span task. No age differences were found in this condition. However, when the same subjects were given lists of words, and asked to recall the words in alphabetical order, a substantial age-related decrement was observed. The present view, in agreement with Welford (1980) and Baddeley (1986), suggests that age differences should typically be found in working memory tasks and that such age differences should be exacerbated by increasing the complexity of the component operations. The exploration of such Age x Complexity interactions in working memory tasks was the principal purpose of the experiments reported in the present chapter. The literature on age differences in memory and information processing gives good support both to the suggestion that older people do less well on working memory tasks, and also to the notion that age and complexity interact in perceptual-motor performance, On the first point, age differences in working memory tasks have been reported by Wright (1981) using the Baddeley and Hitch (1974) paradigm, and by Light and Anderson (1985) using the paradigm introduced by Daneman and Carpenter (1980); both paradigms are further explored in the present series of experiments. In addition, several workers have explicitly linked age-related decrements in comprehension and memory for discourse to a reduction in the efficiency of working memory processes (Cohen, 1981; Light, Zelinski, & Moore, 1982; Spilich, 1983; Zelinski, Light, & Gilewski, 1984). On the second point, Cerella, Poon, and Williams (1980) summarize the evidence in favour of the view that the performance of older people suffers disproportionately as tasks increase in complexity. Their review strongly confirms this "complexity hypothesis," and they further suggest that age decrements are particularly severe when the task involves central cognitive processes, as compared to tasks involving relatively peripheral or automatic processes. Salthouse (1982) also concludes that increases in task complexity are especially disruptive to the performance of older people.


249

The present four experiments were undertaken to provide further information on which aspects of working memory tasks pose problems for older people; in particular, whether processing complexity is especially disruptive to performance in the elderly. Such information should help us to understand age-related deficits in various cognitive functions. In the context of the present volume, it also seems possible that studies of normal aging will illuminate other, more severe types of neuropsychological impairment of short-term memory. The general orientation taken was that "processing resources" decline with age (Craik & Byrd, 1982), in which case it might be expected that any resource-limited difficulty would be exaggerated in older people. A further theoretical point of the studies was to gain more information about possible interactions between age and division of attention, Craik (1977) concluded that older people were especially vulnerable to the effects of divided attention, although this conclusion has been qualified by the results of further work (Somberg & Salthouse, 1982; Salthouse, Rogan, & Prill, 1984; McDowd & Craik, 1988). Finally, consideration was given to the compatibility of the results with current models of working memory (e.g., Baddeley, 1986). Two paradigms were used in the present studies. Experiments 1, 3, and 4 (described more fully by Morris, Gick, & Craik, 1988, and Morris, Craik, & Gick, in press) used the working memory paradigm developed by Baddeley and Hitch (1974) and by Hitch and Baddeley (1976). In this paradigm subjects are first given a variable number of digits or words to hold in mind and are then given a reasoning or decision-making task to perform; finally the subject recalls the original memory items. In our case we asked subjects to remember between zero and eight words, and then presented a sentence for verification. Experiment 2 (described more fully by Gick, Craik, & Morris, 1988) utilized the Daneman and Carpenter (1980) reading span task in which subjects are required to read a series of sentences. In our version of the task, the subject had to decide whether the factual statement presented in each sentence was true or false, and had to respond manually using two-choice response keys. In addition to this ongoing decision task, subjects were asked to remember the final word from each sentence, and to report back the series of final words in their original order after all sentences had been presented and responded to. The task thus constrained the subject to simultaneously process each sentence and hold the set of final words in mind. In both paradigms, the difficulty of the task was manipulated both by varying the number of words to be recalled and also by varying the grammatical complexity of the sentence or sentences to be verified.

10.2. Experiment 1 In this first experiment, younger and older subjects were given a single sentence to verify as rapidly as possible while simultaneously rehearsing zero, two, or four unrelated words. The words (in the memory load conditions) were presented first, and

250


subjects were asked to rehearse them continuously and aloud while verifying the sentence; finally subjects recalled the words in their original serial order. Since subjects rehearsed the memory list throughout the trial, and since the maximum list length was only four words, we expected few recall errors; sentence verification latencies and errors therefore formed the main dependent variables. The word lists consisted of two or four high-frequency bisyllabic words (occurring at least 10 times in the corpus of the Kucera & Francis, 1967, word norms). Eight types of sentences were presented, determined by whether the sentence was true or false, active or passive, and positive or negative, following the procedure used by Baddeley and Hitch (1974). The sentences were constructed so that their meaning would be readily accessible, and contained material that was assumed to be widely known (e.g., A sparrow can build a nest or A cat does not hunt mice). Whereas sentence complexity was varied both by presenting active or passive and positive or negative sentences, the data were collapsed over active-passive, and analyses are presented for positive versus negative sentences only. In line with previous work (Fodor, Bever, & Garrett, 1974), the positive—negative contrast gave rise to larger differences in latency and errors than did the active—passive manipulation. Also, we have no theoretical interest in the nature of "sentence complexity" in the present context, wishing only to vary the difficulty of the on-line processing task; collapsing the data over the active-passive manipulation simply clarifies the exposition. Complexity thus refers to the positive—negative difference in the present study. In all four experiments described in the present chapter the younger subjects were university students in their late teens and early 20s. The older subjects were drawn from a pool of volunteers living independently in the local community. They ranged in age from 60 to 80 years, with a mean age of approximately 70. The two groups were typically matched on number of years of formal education, although mean scores on the Mill Hill Vocabulary Test (a test in which synonyms of words must be recognized) were significantly higher for the older group in all cases. The present experiment was designed with age as a between-subject factor and with two within-subject factors - concurrent memory load (zero, two, or four words) and sentence complexity (positive or negative sentences). The material was presented visually on the monitor of a Commodore PET 8200 microcomputer. A 2-sec warning period signalled by the letter r was first presented in the centre of the screen. In the conditions with the concurrent memory load, this was followed by the memory words, presented to the subject at a rate of one item per 2 sec. The subject was required to start repeating the items aloud, cyclically, and at a steady rate as soon as all of the words had been presented. Immediately after the last word was presented, the sentence appeared. The subject had a maximum of 8 sec to respond by pressing one of two keys using his or her right or left index fingers, according to whether the sentence was true or false. Following the verification response, the sentence was replaced by a line of asterisk

Adult age differences in working

memory

251

YOUNG Q

FOUR WORDS TWO WORDS NO MEMORY LOAD

1

6 -

O Ao-

OLD

a A o

CO

LU

D"

1 £ 3 UJ

A "~~~

o

OLD

YOUNG

2

POSITIVE

NEGATIVE

POSITIVE

NEGATIVE

SENTENCE TYPE Figure 10.1. Mean verification latencies as a function of age and experimental condition (Experiment 1).

characters. The subject was required to keep rehearsing the memory words aloud in serial order for 4 sec after the sentence verification response, at which point the asterisk characters were replaced by the word stop. This procedure was adopted to ensure that the subjects continued rehearsing the memory words well past the sentence verification stage, and to enable the experimenter to ascertain more clearly what the subjects were articulating. In the control condition, without the concurrent load, the sentence immediately followed the warning period and the trial was terminated by the subject's response. As expected, the number of recall errors for rehearsed words was very small (4.8% in the case of the four-word condition, much less in the two-word condition), and the incidence of these memory errors did not differ across age groups or experimental conditions. Verification latencies and errors on the sentence task thus constituted the variables of interest. With respect to sentence verification errors, there were no age differences, but error rates increased significantly both with memory load and with sentence complexity. In addition, memory load and sentence complexity interacted significantly: The effects of complexity were greater as memory load increased. These results fit the expectation that sentence verification becomes more difficult both as the complexity of the sentence increases and as more resources must be devoted to rehearsing the concurrent memory load. Figure 10.1 shows sentence verification latencies for the various conditions. Clearly, both memory load and sentence complexity (positive vs. negative) affect latencies, and

252


decision times also appear to be longer for the older subjects. In fact, all three main effects were statistically reliable. In addition, the Age x Complexity interaction was significant, but the Age x Memory Load interaction was not. With respect to the effects of aging on working memory performance, the main results of interest were thus found in the verification latency data. As predicted, increased sentence complexity had a greater detrimental effect on the older subjects, but the absence of an interaction between age and memory load was quite unexpected. It might be argued that ceiling effects may have obscured a possible interaction between age and memory load in the recall data. This point is addressed in the subsequent experiments. The pattern of results from Experiment 1 makes two interesting points: First, age interacted with one form of complexity (grammatical complexity of the sentences to be verified), but did not interact with another (memory load). Perhaps not all forms of difficulty or complexity are equivalent in working memory situations. Second, the lack of interaction between age and memory load also means that there was no interaction between age and divided attention (zero load vs. two or four items) in the present study. The implications of both points for an understanding of age differences in working memory will be discussed following a description of other experiments in the series.

10.3. Experiment 2 The second experiment used a version of the reading span paradigm devised by Daneman and Carpenter (1980). In this task subjects are required to read a series of sentences. After the entire set of sentences has been read, the subjects are required to report the final word of each sentence in the original order. The task thus constrains the subjects to process each sentence simultaneously, and hold the set of final words in mind. In the present study we modified the Daneman and Carpenter task so that subjects were obliged to process the stimulus material actively. Instead of simply reading, the subject had to decide whether the factual statement presented in each sentence was true or false, and had to respond manually using two-choice response keys; after the series of sentences was presented, the subject attempted to recall the series of final words. By requiring the participant to make a decision, the task is more analogous to the Baddeley and Hitch (1974) working memory task and is likely to be more sensitive to age-related deficits in working memory. On each trial, one, two, four, or five sentences were presented successively; the subject's task was to judge whether each statement was true or false, and then to recall the final words from all sentences. That is, the subject recalled a maximum of one, two, four, or five words, in the original order, following presentation and verification of all sentences. Task complexity was manipulated in three ways. First, sentence complexity was varied by presenting either positive sentences (e.g., Cats usually like to hunt mice, or


253

A canary may often be bigger than a horse) or negative sentences (Bookcases are not usually found by the sea or Children never like to play at the beach) (Chase & Clark, 1972). Complexity was also manipulated by varying the necessity to divide attention. That is, whereas most trials required the subject to divide attention between holding words in mind and processing the next sentence, some trials were given in which no sentences were presented for verification - only the set of words to be recalled in order. Finally, task complexity was manipulated by varying the number of sentences presented on each trial, thereby varying the memory load. In addition, a series of "sentence verification" trials was given in which subjects were required to verify the sentences but not to retain the final words. This series thus acted as a control condition for the effects of memory load on verification latency. The sentences and words were presented sequentially on the monitor of a Commodore PET 8200 microcomputer. In the sentence span trials, subjects read the item silently and pressed one of two keys for the verification response. In the wordalone trials they were required to press one of the two response keys to signal their readiness for the next word. In the word-alone and sentence span conditions, the subject's response was followed immediately by the next item. After responding to the last item, the subjects were required to report the words in their original serial order. The procedure was the same for the sentence-alone conditions, with the exception that for each trial there was only one sentence, with no memory requirements. The subjects were encouraged to be as accurate as possible on the sentence verification task and instructed that accuracy was more important than speed. Eighteen young and eighteen elderly subjects were tested in the experiment. The young subjects were college students who were paid for their participation; their average age was 22 years and their average score on the Mill Hill Vocabulary Test was 14.6. The elderly subjects were drawn from our pool of volunteers; their average age was 68 years and their average Mill Hill score was 16.2 - significantly higher than that of their younger counterparts. The young subjects had an average of 14.8 years of formal education, whereas the older group had received 12.5 years on average. This age difference was also reliable. Thus the young group had received more formal education, but the old group scored reliably higher on the synonym vocabulary test; this pattern is typical in our experience. The dependent measures of interest in the study were sentence verification performance — both latency and errors — and the proportions of final words recalled. The sentence verification results showed that the older subjects made significantly more errors than did the young group. Increased sentence complexity and increased memory load also increased verification errors reliably. Error rates ranged from 4% (young subjects with two simple sentences) to 24% (old subjects with five complex sentences). Two interactions are of major interest in the error data; the interaction of age and sentence complexity was statistically significant, but there was no reliable interaction

254

Craik, Morris, and Gick YOUNG WORD ALONE ASIMPLE SENTENCE o——o COMPLEX SENTENCE O — ~ Q

OLD

100

5 -90 rr g .80 o 2

.70 .60

o CL .50 .40 .30 .20

o' SET SIZE Figure 10.2. Mean proportions of words recalled as a function of age and experimental condition (Experiment 2).

between age and memory load. In the verification latency data, age, sentence complexity, and memory load were again significant effects, but in this case neither the Age x Complexity nor Age x Memory load interactions were significant. Overall then, older people were slower and made more errors in the sentence verification task. As in Experiment 1, age interacted with one type of complexity (the grammatical complexity of the sentences to be verified) but not with another (memory load) in the error data. Figure 10.2 shows the proportions of words recalled in the word-alone condition and in the conditions with concurrent sentence verification. Data are shown for trials in which all sentences were correctly verified (i.e., omitting trials with errors), and for set sizes 2, 4, and 5 only, since recall was essentially perfect in the condition where set size = 1. Two major analyses were carried out on these data; in the first analysis, recall from word-alone trials was compared to recall from the simple sentence conditions (i.e. trials involving positive sentences). This analysis allows an assessment of whether the necessity to divide attention between processing and storage in the simple sentence condition is more disruptive to the performance of older people, relative to performance


255

in the word-alone conditions. Figure 10.2 shows little evidence of an interaction between age and words alone/simple sentences, and the analysis confirmed this lack of interaction. Surprisingly, therefore (but replicating the similar finding from Experiment 1), division of attention was not more detrimental to the performance of older people in this situation. A second major analysis was carried out on the recall data from the simple and complex sentences (the bottom four curves in Figure 10.2). The analysis yielded significant effects of age, sentence complexity, and memory load. The Age x Sentence Complexity interaction was reliable (i.e., recall performance of the older group was reduced more than was performance of the younger group by the use of complex sentences), but the Age x Memory Load (or set size) interaction did not approach significance. As in the verification error data, therefore, age interacted with one source of difficulty but not with another. Taken together, the first two experiments yield consistent, if surprising, results. First, and in contrast to many previous studies of age differences in perceptual—motor performance, the necessity to divide attention between processing and storage had no greater an effect on older people than on their younger counterparts. Second, older people's performance was differentially more affected by the increased grammatical complexity of the sentences to be processed, but there was no interaction between age and the number of words held in short-term storage. It appears that not all forms of complexity are equivalent; it also seems that division of attention cannot simply be redescribed as "increased complexity," since divided attention has similar effects to some but not other types of complexity. Two further studies using the Baddeley and Hitch paradigm were undertaken to provide further information on the locus of agerelated difficulties.

10.4. Experiment 3 In this study, young and older adults were required to hold two, three, four, or five unrelated words in mind while judging whether a sentence was true or false. The words to be remembered were given first; they were presented serially on a computer screen at a 1.5-sec rate. The sentence was then presented; subjects were asked to verify the sentence as rapidly as possible and then to recall the unrelated words in their original serial order. The sentences were either positive ("simple") or negative ("complex") in construction. Finally, in some conditions, no sentence was presented; in this case subjects simply maintained the memory load over an unfilled interval and then recalled the words in serial order. The experiment was thus similar to Experiment 1, although in this case the memory load words were not rehearsed aloud. Sixteen younger and sixteen older adults participated in the study. The young subjects had an average age of 22 years, 13.9 years

256


Q UJ

YOUNG NO SENTENCE o -o POSITIVE SENTENCE • •• NEGATIVE SENTENCE *• *

OLD

< O LU

cr Q

cr o

cr

LU 2 GO

2

3

4

5

LIST LENGTH Figure 10.3. Mean numbers of words recalled as a function of age and experimental condition (Experiment 3). of education on average, and a mean Mill Hill Vocabulary score of 13.8. The older subjects had a mean age of 71 years, 13.3 years of formal education on average, and a mean Mill Hill score of 16.7. The two groups did not differ reliably on years of education, but the older group had reliably higher vocabulary scores. The materials and procedure were similar to those used in Experiment 1. No age difference was found in the number of errors made on the sentence verification task, but the older group had significantly longer latencies. In the error data, the effect of sentence complexity was reliable, as was the interaction between age and sentence complexity. The Age x Memory Load interaction was not reliable however. Error rates were in the 5-14% range for positive sentences, and in the 12-22% range for negative sentences. In the latency data, sentence complexity was again a significant factor, but in this case neither the Age x Complexity nor the Age x Memory Load interactions were significant. Thus, although there were some differences in performance on the sentence verification task in Experiments 1 and 3 - notably that an Age x Complexity interaction was found in the latency data of the first experiment but in the error data of the present study - the overall pattern of results is similar in the two cases. That is, in both experiments age interacted with one form of difficulty (grammatical complexity) but not with another (memory load).


257

Figure 10.3 shows the average numbers of words recalled in the various conditions; there are clear effects of age and memory load. It is also apparent that the necessity to verify a single sentence reduces recall levels for both young and older subjects. This last effect was assessed by collapsing over the sentence complexity variable to yield a 2 (young-old) x 4 (two, three, four, five words) x 2 (sentence present or absent) design. An analysis of variance revealed significant effects of age, memory load, and sentence-no sentence; however, none of the interactions were significant. In particular, the interaction between age and sentence-no sentence was not statistically reliable; it therefore again seems that division of attention (sentence present vs. sentence absent) has as great an effect on younger as on older subjects in working memory situations. A second major analysis was carried out to compare the age groups in the conditions with the sentence verification task present. This analysis yielded main effects of age, memory load, and sentence complexity, with a significant interaction between age and memory load. The interaction is partly attributable to ceiling effects for the young group at list length 2, but there are also some signs that recall performance of the old group is levelling off at around two items, whereas the recall levels of the young group continue to rise. The interaction between age and sentence complexity was not reliable, but Figure 10.3 shows that the effect of complexity was actually larger for the young group. Experiment 3 thus confirmed some of the results from the two previous studies, and also suggested some reasons for the obtained patterns. In the sentence verification aspect of the task, age again interacted with complexity but not with memory load; in the recall data, there was again no interaction between age and division of attention - at least when ceiling effects were avoided. With respect to underlying processes, one possibility is that older people rely exclusively on a relatively automatic articulatory form of rehearsal for holding the memory preload. Such a strategy would result in low levels of recall (as Figure 10.3 shows is the case), and it would also mean that recall would be affected only slightly by changes in the complexity of the concurrent sentence verification task. That is, if the older people are simply holding two words on average in a "rote" or "maintenance" rehearsal fashion, such relatively automatic and peripheral processing is unlikely to be affected by the difficulty of a more cognitively demanding concurrent task. We are assuming here that such maintenance rehearsal can occur simultaneously with the decoding and verification of the visually presented sentence. On the other hand, if younger subjects are augmenting their articulatory recall by recall from a central processor (Baddeley, 1976) and perhaps from secondary memory to some degree, this would lead to increased recall as the memory preload increased in size, but would also imply a somewhat greater vulnerability to the effects of concurrent processing. In an attempt to amplify these effects a final study was carried out. This experiment essentially replicated Experiment 3, but used larger preloads (four, six, and eight

258


words), and also involved the free recall of the preloaded words rather than serial recall. The use of longer words strings should reveal more clearly whether recall in this working memory situation relied on other sources beyond articulatory rehearsal, and whether there were age differences in this respect.

10.5. Experiment 4 Sixteen younger and sixteen older people participated in this final study. The younger subjects were again university students, with an average age of 21 years; the group's mean Mill Hill Vocabulary score was 14.7, and the subjects had received 14.9 years of formal education on average. The older people were home-dwelling volunteers from the local community; their average age was 71 years, their mean Mill Hill score was 17.3, and they had received 15.7 years of education. The groups did not differ reliably with respect to education, but the older group scored significantly higher on the vocabulary test. The basic procedure on each test trial was that the participant first read the memory list of four, six, or eight common words from the computer monitor. The words were presented serially at a 1.5-sec rate. An 8-sec interval followed that was either unfilled (no-sentence condition), or in which a sentence was presented for verification. After the 8-sec interval the word recall appeared, and subjects had 30 sec to recall the memory list in any order. The sentences were again of eight types (positive—negative x active—passive x true—false, but following the procedure of previous experiments, the data were collapsed over the second two variables and so "complexity" was again manipulated in terms of positive versus negative constructions. The sentence verification task yielded few results of interest in the present case. The error rates ranged from 2% to 9% in the case of positive sentences, and from 7% to 14% for negative sentences. There was no main effect of age in the error data, and also no interactions involving age. For latencies, the main effect of age was reliable (the older subjects were significantly slower), but again there were no reliable interactions involving age. The main results of interest were thus the recall data shown in Figure 10.4. The figure shows that with no sentence verification task, the recall scores of both age groups increased with increasing list length, but that the young group's scores increased at a faster rate. An analysis of variance on the no-sentence conditions yielded main effects of age and of list length and also a significant interaction between the two variables. These findings confirm the results of earlier studies (e.g., Craik, 1968). In the conditions with sentence verification, Figure 10.4 shows that recall scores increased as a function of list length for the younger but not for the older group. There is also a small detrimental effect of sentence complexity on the recall scores of the younger group, but little evidence of an effect on the older group's scores. These

Adult age differences in working

memory

YOUNG o — -o

NO SENTENCE

POSITIVE SENTENCE D NEGATIVE SENTENCE A-

Q

259 OLD

D -A

7%(75%)

50% (25%) —

PS

Visual PD

100% 100% 82%{65%) 71% (35%)

100% 100% 93% (85%) 71% (40%)

dissimilar lists, omissions were the most frequent error type (68%), followed by order (18%) and letter substitution (14%) errors.

Word length effect Strings of two- {mese, libro, festa, arma, quadro, cura, voce, sposa, treno, fuoco), three- {secolo, lavagna, vacanza, fucile, cinema, medico, numero, marito, motore, tobacco), and foursettimana, professore, domenica, generale, fotografia, ospedale, telefono, matrimonio, autocarro,

sigaretta) syllable words were used. The procedure adopted was similar to the previous experiments. As shown in Table 17.3, ER showed no effects of word length with auditory and visual input. For three-item auditory lists, omissions (71%) were the most frequent error type, followed by order errors (29%).

Word span lexical—semantic and grammatical category effects

ER's immediate auditory memory span for lists of two-syllable (five-letter) words differing in frequency and imagery values and for high-frequency functors was assessed by the previously described procedure. High-and low-frequency words had use values

455

Phonological processing and sentence comprehension

Table 17.3. Serial recall of two-, three-, and four-syllable auditorily and visually presented words: percentage of items correct in the correct serial position; sequences correct in parentheses Word length (syllable)

A Auditory input

String length 1 2 3

95%

90%

77%(75%) 72%(50%)

92% (90%) 68%(60%)

90% 90% (90%) 72% (50%)

100% 100% 68% (50%)

100% 100% 68%{50%)

B. Visual input

String length 1 2 3

100% 100% 58>%(30%)

Table 17.4. Serial recall of high-frequency/high-imagery imagery (HF/LI), low-frequency/high-imagery

{HF/HI),

high-frequency/low-

{LF/HI), low-frequency/low-imagery

(LF/LI),

and HF function words (F): percentage of items correctly recalled, independent of serial position; sequences correct in parentheses

List length 1 2 3

HF/HI

HF/LI

LF/HI

LF/LI

100% 92%(85%) 45% (25%)

45% 45%(20%) *

75% 90%(85%) 60% (35%)

55% 60%(45%)

45% 52%(25%)

* Refused after three totally unsuccessful trials. greater than 20 and lower than 2.15 per 500,000, respectively (Bortolini et al, 1972). High- and low-imagery words had imagery values greater than 5.6 and lower than 4.9, respectively (Cornoldi, 1974). For each of the four frequency-imagery conditions, the items were randomly drawn from sets of 20 words, with the constraint that an item was never repeated within a string. Since this experiment aimed at assessing the contribution of nonphonological factors to immediate serial recall, an item correct (independent of serial position) score was used. As shown in Table 17.4, ER's repetition performance was influenced by imagery value of the memory items (for one- and twoitem lists, %2[1] = 42.68, p < .001), frequency being comparatively much less relevant (X2U1 = 0.08, n.s.). In three-item auditory lists, which included high-imagery words, error analysis revealed omissions {66%), order errors (30%), and a few phonemic paraphasias (4%). Finally, functors were recalled less than high-imagery words

456

Vallar, Basso, and Bottini i

J3d

3CEN T CORRECT

1-0 i

10

J§

i

i

i

.5. •HI/HF\

" • HI/LF r

N®

•

^

'

•

O LI/HF D LI/LF

1 2 SERIAL POSITION

i

|

|

1

2 POSITION

3

SERIAL

Figure 17.2. A: Two-item lists; B: three-item lists. Items correctly recalled by serial position at input and frequency-imagery value. HI/LI: high-low imagery; HF/LF: high—low frequency. (#2[1] = 35.78, p < .001), performance level being broadly comparable to low-imagery items (x2[l] = 0.04, n.s.). Wide normal and neuropsychological evidence indicates that both in free and serial immediate recall of auditorily presented supraspan lists, performance in the terminal positions represents the output of the PSTS (see Vallar & Papagno, 1986; Shallice & Vallar, this volume, chapter 1). Accordingly, we analysed ER's recall of two- and threeitem lists by serial position. As shown in Figure 17.2A, in the case of two-item lowimagery strings the second item was recalled worse than the first (#2[1] = 5.01, p < .05). For three-item lists (Figure 17.2B) no recency (i.e., a superior recall performance of the terminal position) was found (#2[1] = 0.85, n.s.).

Nonword span The stimuli were pronounceable one- (CVC), two-, three-, and four-syllable nonwords. The procedure was identical to that used in the word span task, with the exception that 10 lists were given for each string length in each syllable length condition. ER was dramatically impaired in this task. In the case of individual nonwords, she was able to repeat correctly only 40% of one-syllable and 10% of three-syllable items, and her performance was totally defective for two- and four-syllable nonwords. Two-item strings of monosyllabic nonwords were also given: ER was able to report 10% of items, but no sequence was recalled entirely correctly. Taken together, these results indicate a deficit of phonological processing and shortterm retention of auditorily presented verbal material. ER is impaired in a phonological


457

discrimination task posing a minimal memory load; her repetition of low-imagery individual words and nonwords is grossly defective (see also data from JB in Allport, 1984); and she has a grossly reduced performance level in a number of auditory span tasks. The presence of the phonological similarity effect suggests a phonological encoding of auditory material, while the absence of the effect of word length indicates that auditory items are not rehearsed. Finally, the lack of both effects with visual presentation suggests that visual material is not conveyed to the PSTS through the rehearsal process (see Vallar & Baddeley, 1984a).

17.2.4. Phonological recoding Before entering the rehearsal process, visual verbal items have to be converted into a phonological form by phonological recoding (see Vallar & Cappa, 1987). When this component is defective, visual items do not have access to the PSTS, even if rehearsal is per se unimpaired. The following set of experiments investigated the function of the phonological recoding process. ER's ability to phonologically recode visually presented material was assessed by reading and matching tasks (Sartori, 1984) and by a rhyme judgment test (Vallar & Cappa, 1987). ER had an errorless performance in reading aloud both words and nonwords presented in a random order (score 7&/7S). She was also able to read aloud with appropriate stress 60 regular (e.g., vicino) and irregular (e.g., minimo) words. In a matching task, in which she had to decide whether two nonwords, one upper- and one lowercase, were the same or different (e.g., NEADE/geade, NEARE/neare), the patient made only one error, scoring 31/32. In the picture-picture, word-picture, and nonword—picture subtests of a four-choice rhyming task she scored 75%, 90%, and 100% of correct answers, respectively (four matched controls: 55%, SD = 7; 83%, SD — 13; 89%, SD = 10). To summarize, ER's ability to phonologically recode visually presented material, as assessed by a range of tasks with a minimal memory component, appears to be unimpaired. The absence of the phonological similarity and word length effects in immediate memory span for visual items is then likely to reflect the disordered operation of the rehearsal process, rather than an access deficit due to an associated impairment of phonological recoding.

17.2.5. Sentence comprehension Sentence—picture matching

ER was given the two-choice test of Parisi and Pizzamiglio (1970), which comprises a wide range of auditorily presented material, including semantically reversible sentences and other items in which word order is important, such as relative sentences. The

458

Vallar, Basso, and Bottini

average sentence length was 5.15 words (SD = 1.96, range = 1-9). The patient scored 76/80, well within the normal range (30 normal controls of Parisi and Pizzamiglio had an average score of 75.2). As assessed by this task, ER's syntactic comprehension abilities appear to be unimpaired. She is able to comphrened the type of sentences that pose problems to other patients with a defective memory span (e.g., case MC, Caramazza et al., 1981). In Parisi and Pizzamiglio's task, word order is relevant for comprehension in 24 out of 80 items; ER's errors were confined to this group of sentences, which consisted of two subject-object, one active and one passive, reversible sentences, and two items with a prepositional spatial contrast (on vs. under). This was explored in more detail in the following experiment. The limited number of alternatives available in this test (e.g., The boy is being pushed by the girl vs. The girl is being pushed by the boy) might, however, be a relevant factor. Assuming that processing of sentences where the linear arrangement of words conveys crucial information involves the PSTS (see Vallar & Baddeley, 1984b), the evaluation of the match-mismatch of each alternative picture with the presented sentence might require the availability of its verbatim record, which then would need to be preserved throughout the process of response selection. Following this line of reasoning, scanning two pictures would require retention of the sentence in the PSTS for a comparatively minor period of time than, say, examining four alternatives. Since, as discussed earlier, ER has a defective immediate memory performance and fails to rehearse information stored in the PSTS, her performance might deteriorate substantially on increasing the number of alternative pictures. In addition, since the two alternative choices include only the target and a thematic role reversal, the patient might have used specific strategies, focusing on this aspect of the task. ER was then given two four-choice sentence-picture matching tests. The first task (A) comprised active and passive direct object nonreversible, active indirect object, and relative sentences (average length, 6.70 words; range, 5-9), where word order was not crucial for adequate comprehension, which could be achieved by processing the major lexical items. For instance, the display for the sentence The boy looks at the tree included the target and three pictures, showing a boy looking at a car, at two boys, at a man and at a dog. The second task (B) comprised items where word order is a relevant factor: active and passive direct object semantically reversible, active indirect object, and relative sentences (average length, 6.62 words; range, 5-9). For instance, the display for the sentence The cat is being chased by the dog included the target and three pictures showing the reversed sentence, The dog is being chased by the cat; a cat and a dog involved in different actions, The cat is behind the dog and The cat and the dog look at each other. As shown in Table 17.5, ER had an errorless performance in Task A, but was clearly impaired in Task B. Error analysis revealed that ER chose the syntactic distractor in 9 out of 11 trials.

459


Table 17.5. Correct answer in four-choice sentence—picture matching (M) and repetition (R) tasks, for A and B sentences Direct object

Indirect object

Relative

Total

Active

Passive

Matching A B

4/4 4/6

4/4 lib

4/4 51b

4/4 lib

16/16 (100%) 13/24 (54%)

Repetition A B

3/4 lib

314 lib

114. 31b

1/4 lib

9/16 (5b%) 8/24 (33%)

Sentence verification In this experiment ER was required to judge whether sentences presented by the examiner were true or false. All items were statements about the world, relying on general knowledge to determine truth or falsity. The material comprised four conditions: (A) Short sentences (e.g., Plants grow in gardens; average length, 5 words; range, 4—9), where the false items were created by mismatching a subject and its predicate (e.g., Plants are people). (B) Sentences based on A items, made longer by the addition of verbiage, such as an introductory or a relative clause (e.g., There is no doubt that champagne is something that can certainly be bought in shops vs. Lettuce is the kind of person that one rarely meets in a schoolroom-, average length, 17.43 words; range, 13-28). (C) Short sentences (e.g., Rivers are crossed by bridges-, average length, 6.03 words; range, 4-9), where the false items were created by modifying the linear arrangement of the words, reversing two relevant items (e.g., The world divides the equator into two hemispheres). (D) Sentences based on C items made longer, as B items, by the addition of verbiage (e.g., Many people know that often books contain pictures of various kinds, which are sometimes printed in colours vs. One could reasonably claim that sailors are often lived on by ships of various kinds; average length, 18.25 words; range, 13-28). Each of the four conditions comprised 31 sentences. The material was presented both auditorily and visually (see Vallar & Baddeley, 1984b, for a detailed description of the presentation method). Table 17.6 shows ER's performance in the four conditions. In the two sets (A and B) in which items were made false by a semantic mismatch, the patient's comprehension was preserved and not detectably affected by sentence length, with both auditory and visual input. Conversely, in the two sets (C and D) in which items were made false by a word reversal, performance deteriorated when long sentences were given, dropping from over 80% correct to chance level (#2[1] = 14.18, p < .001). ER's selective impairment in the case of D sentences cannot be accounted for in terms of an aspecih'c

460

Vallar, Basso, and Bottini Table 17.6. Sentence verification task (percentage of correct answers) Auditory

Visual

Total

100 94 81 59

100 97 84 50

100 95.5 82.5 54.5

Condition

A B C D

Note: Chance = 50%. Number of sentences per condition = 64; 32 auditory, 32 visual. effect of increased length and/or syntactic complexity. The sentences of the control B set, where the anomaly is produced by a semantic mismatch, are of comparable length and structural complexity. In Condition D the majority of ER's errors (72%) were false alarms (false sentences were judged to be true), whereas miss errors (true sentences judged to be false) were comparatively less frequent (28%), a significant difference (2 = 2.23, p < .05). This finding may be taken as an indication that ER's judgments are primarily based on a lexical-semantic reading. False D sentences, where a semantic anomaly is produced by a word reversal that she apparently fails to appreciate, tend to be erroneously judged as plausible like true B sentences, since in both cases there is no semantic mismatch among the major lexical items. Conversely, comparatively few errors occur in the case of both B and D true sentences, where there are no semantic anomalies dependent on either lexical items or word order. Finally, B anomalous sentences may be accurately detected, given the presence of a lexical—semantic mismatch. To sum up, the present evidence from ER suggests an inability to take into adequate consideration word order information, coupled with a reliance on lexical-semantic analysis. This makes her prone to fail in the case of implausible D items, where the anomaly is produced by a word reversal. In the case of A and B plausible items, word order is, of course, relevant in that C- and D-type anomalous sentences, on which ER would be expected to fail, may be produced by a word reversal. For instance, the B item There is no doubt that champagne is something that can certainly be bought in shops may become There is no doubt that shops are something that can certainly be bought in champagne. Verification of the plausible A and B

items, which may be provided by lexical-semantic analyses, primarily utilized by ER, remains preserved, however. ER's unimpaired performance in the long and complex B sentences may also be taken as an indication that syntactic processes are largely spared, consistent with the previously reported data from the two-choice sentence—picture matching task. The suggestion has been made by Schwartz et al. (1987) that, were the parser defective, in the complex plausible B sentences some anomalous reading of a lexical-semantic type


461

10-

I

M SERIAL

T

POSITION

Figure 17.3. Sentence repetition. Items correctly recalled in the appropriate serial position. Position at input: / (initial: 1-2); M (middle); T (terminal: n and n — 1).

may occur. For instance, the sentence about champagne could be interpreted as There is no champagne that shops are something that can be bought in doubt.

17.2.6. Sentence repetition Study 1 In this experiment ER was required to repeat verbatim the A and B sentences used in the matching task. As shown in Table 17.5, ER's repetition of B sentences, in which word order is relevant for meaning, was grossly defective, whereas performance for the A items was better preserved. Repetition performance for the two sets of sentences was scored as items correct in the appropriate serial position in the initial (1 and 2), middle, and terminal (n and n — 1) positions. As shown in Figure 17.3, ER's immediate recall performance was better for the A sentences, and no recency effect was present. A two-way analysis of variance revealed significant main effects of sentence type (F[l, 114] = 12.07; p < .01), and of serial position (F[2, 114] = 7.38, p < .01), while the interaction failed to reach significance level (F[2, 114] < 1, n.s.). Error analysis of A sentences revealed omissions (67%), order (22%), and content word substitution (11%; e.g., gatto [cat] vs. cane [dog]) errors. In the B sentence set, errors included omissions (79%), order errors (13%), content (3%; e.g., bambina vs. ragazza [girl]), and function (1% il [the], singular masculine vs. la, singular feminine) word substitutions, active-passive voice shift (1%; tira [pull] vs. tirata [pulled]), inflectional

462


errors (3%; e.g., inseguiva [chased] vs. insegue [chases]). For both sets of sentences there was no overall significant difference in recall accuracy between content and function words. The vast majority of errors, as previously mentioned, were omissions, which occurred most frequently in the terminal serial positions. In a number of items, however, most of the sentence meaning was preserved. For example, the target B sentence // ragazzo che va in bicicletta insegue la macchina (The boy who is on the bicycle follows the car) became // ragazzo inseguiva in bicicletta la macchina (The boy followed on the bicycle the car). In two other sentences the only errors were the substitution of a content word with a synonym and an inflectional error (tense change). These observations seem to indicate that ER's sentence repetition, as in other patients with a defective PSTS (Saffran & Marin, 1975), may be supported by nonphonological (lexical-semantic) components. This is consistent with her pattern of performance in the case of lists of unrelated words of different grammatical category and imagery value. An additional analysis aimed at assessing in greater detail whether semantic gist was preserved in ER's repetition. We employed a method used by Friedrich et al. (1985), which identifies the basic relations within each sentence (subject-verb, verb-object and, in the case of relative sentences, noun-noun). The percentages of semantic links preserved in the A and B sentences were 52% and 29%, respectively, consistent with the previously mentioned recall advantage of the A items.

Study 2 In this study we explored in greater detail the role of nonphonological factors in sentence repetition by using a test devised by Ostrin and Schwartz (1986), which manipulates active—passive voice, semantic reversibility, and plausibility. The test included reversible (e.g., The boy is pulling the girl and The girl is being pulled by the boy), plausible (The boy is pulling the wagon and The wagon is pulled by the boy), and implausible (e.g., The window is washing the man and The man is washed by the window) active and passive sentences. Compared with the material of the previous study, the Italian version of Ostrin and Schwartz's material comprised shorter sentences: The active and passive voice items were five and six words in length, respectively. This could be considered a positive factor for our study, as in the previous experiment overall level of performance in the case of B sentences was rather poor (33% correct) and most errors were omissions, preventing a detailed analysis of the functional components underlying residual performance. The test comprised 48 items (12 reversible and 12 nonreversible — 6 plausible and 6 implausible — sentences for each voice) and 24 fillers. The 72 sentences were subdivided into two sets of 36, balanced with respect to sentence type. For each set the sentences

Phonological processing and sentence comprehension 1-0

T

M

PLAUSIBLE

D

REVERSIBLE

H

IMPLAUSIBLE

463

cc cc

o (J

Q_

ACTIVE

PASSIVE

Figure 17.4. Percentage of correctly repeated plausible, reversible, and implausible sentences by voice (active and passive). Table 17.7. Error analysis for sentence repetition Voice

Active voice

Passive voice

Constituent order Closed class Voice switch Open class Compounds Miscellaneous Total

0 0 1 0 1 0 2

0 3 6 2 2 0 13

(23 %) (46%) (15.5%) (15.5.%)

were auditorily presented in a random order to the patient, who had received instructions to repeat each item verbatim. The average percentage of sentences repeated correctly is shown in Figure 17.4. As compared with her performance in the previous repetition study, ER showed a remarkably better overall performance level, repeating with complete accuracy about 90% and 44% of the active and passive sentences, respectively. This significant advantage of the active voice (%2[1] = 20.18, p < .0001) is also present in normal subjects (e.g. Savin & Perchonock, 1965), though at a higher performance level. Error analysis (see Ostrin & Schwartz, 1986, for details) is shown in Table 17.7. The prevailing error type (46%) was a passive-active voice switch, which reversed the meaning of three reversible sentences, made plausible two implausible sentences, and made implausible one plausible sentence. The single active—passive voice switch involved an implausible sentence, which was made plausible. Hence, all four implausible sentences ER failed to repeat correctly were made plausible, three, as just mentioned, by

464


a voice switch and one by a compound error (the substitution of one open and one closed class word). The only exception to this tendency towards plausibility was the one voice switch error involving a passive plausible nonreversible sentence. Closed class errors included two additions and one ambiguous active-passive. Open class errors comprised one omission and one substitution.

17.3. Discussion 17.3.1. Phonological processing and storage deficits Case ER shows both a defective auditory memory span and clear phonological analysis difficulties. This pattern of impairment is not uncommon among patients with a putative deficit of auditory—verbal short-term memory. Evidence suggesting a phonological processing deficit has been extensively demonstrated in case EA (Friedrich et al., 1984) and suggested by Allport (1984) in case JB (see, however, Vallar & Baddeley, 1984b, and Shallice & Vallar, this volume, chapter 1). Other patients (IL: Saffran & Marin, 1975; MC: Caramazza et al., 1981) do not have an errorless repetition of individual items, a finding that might indicate an input analysis disorder. On the other hand, correct immediate repetition of single items and preserved phonological processing have been found in three patients with a reduced auditory memory span (PV: Vallar & Baddeley, 1984b; EE/EDE: Berndt, 1985, and Berndt & Mitchum, this volume, chapter 5; TB: Baddeley, Vallar, & Wilson, 1987). These findings are consistent with a serial organization of the processing and storage components, where the output of phonological analysis is the input to the PSTS (see, e.g., Vallar & Cappa, 1987). A processing deficit should therefore produce impaired immediate retention, giving rise to a faulty input to the PSTS, but storage deficits that cannot be traced back to processing disorders may also occur, as shown by the dissociations discussed earlier. ER has a pattern of memory impairment comparable to cases, such as PV, where the disorder is confined to storage. With auditory input (see Tables 17.2 and 17.3) ER shows the effect of phonological similarity, but not that of word length, suggesting that the material is encoded phonologically but not rehearsed. The finding that ER's immediate repetition of individual letters in the phonological similarity set is not errorless corroborates the conclusion that her analysis components are defective. In serial recall of auditory strings the recency effect, which represents the output of the PSTS (see Shallice & Vallar, this volume, chapter 1), is absent. ER's defective short-term memory performance cannot, however, be entirely explained in terms of impaired phonological processing. The patient, as shown by the lack of both the phonological similarity and word length effects, also does not make use of the rehearsal process when the material is presented visually, even though her phonological recording skills are


465

preserved. This pattern of results indicates that ER does not utilize the PSTS in the short-term retention of visual verbal material. This could be attributed to a deficit of the PSTS, while articulatory rehearsal might be either impaired or preserved but, in this latter case, not used to feed visually presented information to a damaged PSTS, due to a strategy choice (see Vallar & Baddeley, 1984a). Alternatively, a rehearsal deficit might prevent visual input from gaining access to a per se preserved PSTS. This latter possibility cannot, however, be easily assessed due to the co-occurrence of phonological processing deficits. This analysis holds also for the case of EA, where a similar pattern of impairment has been found (Friedrich et al., 1984; Martin, 1987). The presence of phonemic paraphasias in ER's spontaneous speech may be broadly consistent with the hypothesis of a rehearsal deficit, if one assumes that the rehearsal process utilizes a phonological output buffer, primarily involved in the storing of preplanned sequences of impending speech (see Shallice & Vallar, this volume, chapter 1, for a further discussion). To summarize, on the available data ER appears to suffer from a deficit of both phonological analysis and articulatory rehearsal, the PSTS being either impaired or not adequately used. Since phonological analysis and rehearsal convey, respectively, auditory and visual information to the PSTS, when the former components are defective the latter cannot operate properly. With auditory presentation the store receives a faulty input from phonological processing, whereas visual information, as a result of the rehearsal deficit, does not enter the system, which therefore cannot contribute to immediate retention, On this analysis it remains in principle possible that the PSTS is per se unimpaired, but because of access difficulties by both auditory and visual input, it is nonetheless unable to provide any relevant contribution to immediate retention. Finally, even in the case of an intact but isolated PSTS, the pattern of immediate memory performance of patients such as ER and EA is expected to be very similar to that of patients with pure storage deficits; this, as previously discussed, appears to be the case. Borrowing a distinction originally proposed for amnesic disorders (Baddeley, 1982), selective impairments of immediate retention of auditory-verbal stimuli may be subdivided into two groups: (a) primary deficits of phonological short-term memory, which may be attributed to an abnormally reduced capacity or total disruption of the PSTS, phonological analysis abilities being preserved (see Vallar & Baddeley, 1984b); patients like PV, EE, and TB would appear to belong to this group; (b) secondary deficits of phonological short-term memory, which may be traced back, wholly or in part, to processing disorders; this would be the case for patients like ER and EA. The suggestion has been made that all verbal short-term memory disorders are inseparable from impairments of a central phonological code (Allport, 1984) and therefore secondary to processing deficits. This view, reminiscent of the level-of-processing approach (which considers the memory trace essentially as a by-product of perceptual processing; see

466


Craik & Lockhart, 1972), does not, however, account for the existence of patients with selective storage deficits. It is, finally, worth noting that the present type of analysis predicts the possibility of patients with selective deficits of phonological analysis, the PSTS and rehearsal components being spared. Such cases would show a defective auditory memory span and, with visual presentation, both a better performance level and the preservation of the normal phonological similarity and word length effects. We have argued that phonological processing disorders may produce "secondary" deficits of verbal short-term memory, with a pattern of impairment close to that of "primary" storage deficits. The sentence comprehension impairment associated with this type of phonological memory deficit, which may be expected to show resemblances to the pattern observed in cases with "primary" disorders, is discussed in the following section.

17.3.2. PSTS and sentence comprehension ER shows a defective performance on both the four-choice version of the matching task and on the verification test, and her impairment involves sentences in which the linear arrangement of words conveys information crucial for meaning. In Study 1, ER's repetition deficit parallels her comprehension disorder. Can this deficit be traced back to a specific dysfunction of the syntactic processing systems, independent of the reduced capacity of the PSTS? This possibility appears unlikely, since in the verification task ER's comprehension deficit is dependent on an interaction between length and sentence type: The patient fails only in the case of the long and syntactically complex "D" sentences, where the anomaly is produced by a word reversal, but has a good performance level with short and simple versions of this sentential material, for example The world divides the equator into two hemispheres. Furthermore, she is also unimpaired in the case of both short sentences and their long and structurally complex versions, provided that the processing of the major lexical items may secure adequate understanding. Evidence for a preservation of ER's syntactic abilities is also offered by her unimpaired performance in the two-choice version of the sentence—picture matching task, even though it should be noted that all four errors occurred on items for which word order is relevant. ER's pattern of impairment in the verification task is compatible with the hypothesis that the PSTS is the working space of the syntactic parsing device (Clark & Clark, 1977; Caramazza & Berndt, 1985). According to this view, comprehension performance would be preserved when the material over which syntactic computations are carried out does not exceed the abnormally reduced capacity of the PSTS (see Caramazza & Berndt, 1985, p. 61). However, the deterioration of ER's performance in the sentence-picture matching task when a four-choice paradigm is used does not appear in line with this hypothesis. If the PSTS is the working space of the syntactic parser,


467

comparable degrees of impairment may be expected in both versions of the task, when, as in the present study, similar sentential materials are used. Taken together, the present data clearly indicate that the PSTS contributes to aspects of sentence comprehension and make the hypothesis of a primarily syntactic deficit highly implausible. In addition, they do not appear in line with the view that the building up of a syntactic representation of a sentence requires temporary retention in the PSTS. This latter conclusion is also suggested by recent reports (Vallar & Baddeley, 1987; Saffran & Martin, this volume, chapter 16; Butterworth, Shallice, & Watson, this volume, chapter 8) showing that patients with a defective PSTS may be fairly successful in grammaticality judgments, even in the case of lengthy sentences with a number of words interposed between the two relevant items. It should be noted that this interpretation rests on the crucial assumption that grammaticality judgments require the availability of a complete syntactic representation of the sentence, making use of only a constant, small amount of "working memory," independent of sentence length. Conversely, the interpretation of a given sentence may need a more complete parsing and the allocation of an amount of working memory proportional to the length of the sentence (Pulman, 1987). Following this line of reasoning patients with a defective PSTS might have a residual memory capacity adequate for grammatical judgments but not for a full parsing. A second source of evidence (see Marslen-Wilson, 1984, for review) stems from the observation that, within a sentence, words may be recognized when the subject has heard only the first two phonemes, as quickly as 200 msec after onset, whereas recognition time of words in isolation is substantially longer. This facilitatory effect, which occurs not only with normal but also with "syntactic" prose (i.e., materials grammatically well formed for which, however, no semantic interpretation is possible), takes place very early in the sentence and develops over serial positions. These findings indicate that a high proportion of syntactic and semantic analyses operates "on-line" on the incoming spoken input. These sources of evidence concur to indicate that the PSTS does not provide a major contribution to syntactic analysis of sentential material. Given that lexical-semantic processing is also typically spared in short-term memory patients, is the PSTS of any use in the comprehension process? The sentential materials in which short-term memory patients fail are, as mentioned in the Introduction, very heterogeneous, and processing has been assessed in different ways, such as verification, sentence-picture matching, and repetition. They appear, however, to share, with the possible and problematic exception of patient NHA on record (see McCarthy & Warrington, 1987a, and this volume, chapter 7), the following characteristics: (a) the sentences are grammatically well formed; (b) there are no semantic mismatches between the major lexical items; and (c) word order conveys information relevant for meaning. Accordingly, syntactic and lexical—semantic analyses alone do not fulfil the demands of the tasks, but integration of the results of such levels of processing is needed, since

468


adequate comprehension depends on the order in which the constituent words are presented. A correct interpretation is achieved only when the products of lexical-semantic processing are mapped onto the syntactic representation of the sentence. This, we suggest, would require the availability of the phonological record of the sentence, which preserves the crucial word order information. In line with the normal evidence discussed earlier, we assume that a great deal of lexical-semantic and syntactic analyses performed on an auditorily presented sentence takes place on-line and in a parallel fashion. At the lexical—semantic level, processing would lead to variably constrained interpretations (e.g., reversible vs. nonreversible sentences), which would not be primarily influenced by the linear arrangement of the constituent words in the sentence. At the syntactic level, the output of the parsing process would be a description of the grammatical structure of the sentence, comprising the sequence of word classes and their grouping into higher-level phrases. At this level the linear order of the constituent words needs, of course, to be taken into account, as shown by any grammatical tree. This does not necessarily imply, however, that the parser's output includes any information concerning the lexical items specific to a given sentence, in addition to a description of its grammatical structure. At this level of detail, for instance, sentences with unrelated meanings, such as The angry farmer chased the dog and A missing fuse was the problem would receive the same syntactic analysis (see Pulman, 1987, pp. 165-166). This would also be the case for sentences such as The boy follows the dog versus The dog follows the boy and The boy is eating the jelly versus The jelly is eating the boy. Consider first the case of the two semantically reversible sentences mentioned previously, in the context of a sentence—picture matching task, for instance. If the syntactic representation is confined to a structural description and lexical-semantic processes do not take into account information carried by the linear arrangement of words, normal subjects would show difficulties in discriminating between such sentences, but would be able to reject a lexical distractor, such as The dog follows the cat. Similarly, semantic anomalies produced by a thematic role reversal, such as the previously mentioned sentence The jelly is eating the boy, would not be detected. Conversely, sentences such as The jelly is eating the screwdriver would be recognized as anomalous by the lexical-semantic processing of the major lexical items. These difficulties may be overcome, however, if the assumption is made that the PSTS is involved in the mapping process. This storage component, as repeatedly suggested by independent sources of evidence (e.g., Wickelgren, 1965), preserves item order. The PSTS, providing information concerning the serial position of the lexical items of a given sentence, may therefore allow the mapping of the products of lexical-semantic processes, in which the linear arrangement of the constituent words would not be represented, onto the structural description of the sentence, in which word order is maintained in terms of the sequences of grammatical word classes, without any additional lexical information. Alternatively, some information concerning the lexical


469

items of the sentence and their serial position should be present in the syntactic representation of the sentence. This has been the suggestion of Scholes (1978, p. 173), who in an attempt to deal with these issues, argues that the match between the outputs of syntactic and lexical processes may occur only if they share sufficient syntactic and lexical information. The syntactic representation cannot simply be a grammatical description, but includes "word shapes" or "feature specifications," which are also present in the lexical representation. The precise nature of such "shapes" is not specified in detail, but they should maintain the linear arrangement of words, given their crucial role in allowing the detection of semantic anomalies produced by a word order reversal. The present hypothesis that the PSTS is involved in the mapping of the syntactic description of a given sentence onto its representation based on lexical-semantic processes is consistent with three sources of evidence from patients with a defective PSTS. First, the preservation of syntactic processes, even in the case of long-distance dependencies, may account for their largely unimpaired performance in grammaticality judgment tasks, at least in the few cases investigated at the moment. Second, their basically normal comphrehension when word order does not provide relevant information may be explained in terms of a spared lexical—semantic level of processing. Third, the observation in both ER and PV (Vallar & Baddeley, 1984b, 1987) of an interaction among sentence type, length, and syntactic complexity - so that comprehension is grossly defective only in the case of long complex sentences in which word order is relevant for meaning - is readily consistent with the present interpretation in terms of defective PSTS, which would allow adequate mapping only for material within its abnormally reduced capacity. The previously mentioned alternative possibility that the PSTS has no relevant role in sentence comprehension, since the outputs of syntactic and lexical-semantic processes include the lexical information required for mapping, cannot easily account for both these sorts of information load effects in sentence verification and the preservation of grammaticality judgments. Were this the case, a syntactic deficit may be expected to produce a defective performance in both grammaticality judgments and comprehension of sentences where word order is relevant for meaning. According to the present hypothesis, the deficit of the PSTS component of verbal memory disrupts the mapping of the syntactic representation of a given sentence onto the products of lexical—semantic analyses. However, since both the syntactic and the lexical—semantic levels are spared, patients with the selective disorder of the PSTS might attempt to use alternative strategies to operate the mapping process. Conversely, strategic effects may be less likely to occur in the case of impairments primarily involving the syntactic or lexical-semantic levels. ER's pattern of performance in the two sentence-picture matching tasks might be explained along these lines, even though specific testing of this hypothesis is needed. In the two-choice version of the task, where the two alternative pictures include the target and the thematic role reversal distractor, the patient may focus on this aspect, disregarding, at least in part, the lexical

470


components of the test. This could allow a more direct mapping, minimizing the contribution from the PSTS, even though it is worth repeating that all of ER's four errors involved reversible items. The utilization of specific strategies may be less viable, however, in the case of the four-choice version of the task, where both lexical and syntactic distractors are present. ER's good repetition of subject-verb-object active sentences in Study 1 might reflect the utilization of a strategy, based on the serial order of the constituent words, which assigns agency to the first noun phrase (see Bever, 1970). Over-reliance on such a strategy, however, may be expected to produce a disproportionate impairment in the case of passive sentences, as actually observed in the patient, even though other factors such as sentence length and syntactic complexity need to be taken into consideration. Some additional evidence for the existence of strategic effects in short-term memory patients comes from the finding that PV is unimpaired in a grammaticality judgment task involving long-distance dependencies. Her performance, however, shows a slight but significant deterioration in a verification task including both nongrammatical and semantically anomalous sentence (Vallar & Baddeley, 1987). More direct indications for the use of specific cognitive strategies come from the observation that in immediate free recall of unrelated lists of auditory items PV recalls first the initial items, which reflect the activity of her unimpaired longterm memory processes. Conversely, in the case of normal subjects the terminal items, which represent the output of the temporary PSTS defective in PV's case, are produced first. This represents a strategic choice: If specifically instructed, the patient is able to adopt a recall-from-end strategy (see Vallar & Papagno, 1986). Both the unimpaired performance of patients with a selective PSTS deficit like PV and JB in a wide range of lexical-semantic and long-term memory tasks (see for details and references Shallice & Vallar, this volume, chapter 1) and their normal everyday life are, by and large, favourable conditions for the utilization of effective strategies. These observations suggest that both in comprehension and memory tasks shortterm memory patients may use a number of processing strategies that are probably also available to normal subjects. This does not necessarily imply, however, that all such strategies are usually employed by the normal individual. It might prove, conversely, to be the case that the availability of a temporary storage component may prevent a potentially perilous over-reliance on a specific strategy, as possibly suggested by ER's pattern of repetition performance in Study 2. The use of strategies is a relevant aspect of Caplan and co-workers' hypothesis that the PSTS may be involved in the comprehension process, providing postinterpretative adjudication between alternative readings of material such as semantically reversible sentences and items made anomalous by a word reversal (Caplan & Waters, this volume, chapter 14; Caplan et al., 1986). Caplan and coworkers' view, however, rests on the assumption that normal sentence processing involves, in addition to lexical-semantic and syntactic analyses, the systematic and simultaneous utilization of "perceptual" strategies based on the linear sequence of the


471

constituent words (see Bever, 1970), and therefore produces conflicting interpretations in the case of sentences where word order is relevant. Although this role of the PSTS is an open possibility, which may be tested in short-term memory patients, Caplan and coworkers' assumption about normal processing should also be investigated. The "mapping hypothesis" of Linebarger, Schwartz, and Saffran (1983), which indicates "the translation between descriptions of sentence form and descriptions of sentence meaning" (Schwartz, Linebarger, Saffran, 1985, p. 121) as the functional locus of the sentence comprehension deficit of the so-called agrammatic patients, may appear to bear some resemblance to the present suggestion. A main difference, however, involves the role of phonological storage. On the present view the primary locus of the deficit is the PSTS, the integrity of which is needed to allow integration of the products of syntactic and lexical—semantic processes, per se spared. The pattern of relationships between sentence type and length in ER's performance in the verification task supports this interpretation. Linebarger et al. (1983), if we read them correctly, consider the defective phonological memory of their patients as an associated deficit, which would deprive them of a backup store. If one assumes, as Linebarger et al. do in their Hypothesis 2, that agrammatic patients suffer from a trade-off between syntactic and semantic processing, recovery of the phonological record of the sentence might be useful, allowing repeated processing attempts. ER's data from the sentence verification task indicate a comparable role of the PSTS in sentence comprehension for both auditory and written material. Consistent with these findings, a defective comprehension of sentences where word order is relevant has been observed for both input modalities in a number of other short-term memory patients (PV: Vallar & Baddeley, 1984b; MC: Caramazza et al., 1981; EA: Friedrich et al., 1985). Taken together, these data appar to indicate that, in the case of written presentation, a visual short-term store cannot be successfully used to secure adequate processing of these types of sentences, but phonological recoding and access to the PSTS are needed. Within the hypothesis outlined in this discussion, it would appear that a visual shortterm store component does not typically provide a major contribution to the mapping process, but this may be an issue for further research. Having attempted to specify the role of the PSTS in sentence comprehension, we wish to make clear that we do not confine its contribution to human cognition to this specific aspect of speech processing. The PSTS, instead of being the specific working space of a given device (e.g., the parser, Caramazza and Berndt, 1985; cf. Butterworth et al., 1986; Caplan et al., 1986; and Caplan and Waters, this volume, chapter 14), may serve as a multipurpose storage system for a range of cognitive operations. The PSTS may be useful as a backup store when multiple aspects of a sentence, which cannot be fully analysed on-line, have to be processed in some detail (Vallar & Baddeley, 1987) and to allow repeated processing attempts when a given syntactic or semantic component is rendered less efficient due to brain damage (Linebarger et al., 1983). The PSTS is

472


involved in long-term phonological learning and may contribute to aspects of language acquisition in children (Baddeley, Papagno, & Vallar, 1988). Finally, the suggestion has been made that some sort of short-term storage may be involved in verbal reasoning (Hitch and Baddeley, 1976) and mental arithmetic (Hitch, 1978). Little neuropsychological evidence is at present available, but the two short-term memory patients of McCarthy and Warrington (1987a, b, and this volume, chapter 7) show a pattern of impairment that may also involve aspects of verbal reasoning. On the basis of McCarthy and Warrington's findings we would not, however, rule out a contribution of the PSTS to more strictly linguistic aspects of sentence processing, since these two patients have a clear deficit on the Token Test.

17.3.3. PSTS and repetition: lists and sentences ER's repetition both of lists of unrelated words and sentences is defective. The standard recency effect, assumed here to represent the output of the PSTS (for word lists see a review by Shallice & Vallar, this volume, chapter 1; for sentences, some experimental evidence is given in Butterworth et al., 1986), is absent for both sets of materials. Taken together, these findings suggest an involvement of the PSTS in immediate retention of both sentences and lists of unrelated items, with a major contribution in the final serial positions. As for nonphonological factors, the massive effects of imagery value - and not of frequency - on recall of word lists argue for a main role of the semantic system, lexical components being comparatively less relevant. This does not mean to say, however, that lexical nonsemantic components do not contribute to immediate retention (see Berndt, 1988, for a review of the relevant evidence). In the case of sentences, both ER's better level of performance when the linear arrangement of words is constrained by semantic factors, such as in nonreversible sentences (see Figure 17.3), and the observation that the patient makes plausible all the implausible sentences she fails to repeat verbatim, concur to suggest a relevant contribution of the semantic system. These effects occur throughout all serial positions, including recency (see Figures 17.2 and 17.3). This may represent a not entirely successful attempt to form a compensatory strategy, since in normal subjects the terminal positions of free and serial recall lists are typically unaffected by a range of variables that includes, among others, word frequency (e.g., Watkins & Watkins, 1977).1 As previously noted, the disproportionate impairment in the repetition of passive sentences may also reflect strategic effects. The relative contribution of the PSTS and of the semantic system to immediate memory may vary according to a number of variables such as the nature of the material and its length.2 On the one hand, retention of meaningless nonword lists is likely to rely mainly, if not exclusively, on phonological storage, whereas words may benefit from semantic coding. Similarly, immediate repetition of lists of unrelated functors,


473

which presumably do not have a rich semantic representation, is very poor, with a performance level much lower than high-imagery content words. However, in the context of a meaningful sentence there is no difference in recall accuracy between concrete content words and functors, because of their relevant syntactic and semantic role. On the other hand, in the case of nonreversible sentences semantics may pose strong constraints on the linear arrangement of words, thus minimizing the role of the PSTS. Conversely, sentences in which word order, typically preserved by phonological coding, is crucial for meaning would call for a major involvement of this temporary storage system. ER's pattern of repetition deficit in Study 1, which parallels her comprehension disorder, is in line with these conclusions. ER's performance, as in the case of normal subjects (see references in Butterworth et al, this volume, chapter 8), is better for sentences than for lists of unrelated words. For instance, average fully correct repetition scores are 25% and about 70% for highfrequency/high-imagery three-item lists (see Table 17.4) and sentences five to six words in length (see Figure 17.4), respectively. This sentence superiority effect may reflect both semantic (in the case of meaningful material) and syntactic processing. The relative contribution of these two factors has not been systematically investigated here by repetition of meaningless syntactically well-formed sentences. However, ER recalls implausible sentences five to six words in length (about 65% correct) better than threeitem lists of high-imagery unrelated words (25% correct). This substantiates the case for a contribution of syntactic factors. 17.3A.

Conclusions

The main data and issues of this chapter may be summarized as follows. First, selective deficits of immediate phonological memory may be produced by both "processing" ("secondary"), as in the present case ER, and "storage" ("primary") disorders. These phonological memory processing and storage components are primarily involved in repetition of both lists of unrelated items and sentences, with a differential contribution of nonphonological syntactic and lexical—semantic systems. Second, the pattern of comprehension impairment associated with these phonological short-term memory deficits typically involves sentences in which word order conveys information relevant for meaning. The suggestion is made that this comprehension disorder may be attributed to a defective mapping of the products of lexical-semantic processing onto syntactic representations, which are per se spared, possibly allowing the utilization of more or less effective alternative strategies.

Notes 1. We are not aware of studies concerning possible relations between recency and imagery effects.

474


2. In the case of visual letters, the standard effect of phonological similarity may disappear when long sequences are presented, suggesting that subjects may opt for a more effective alternative (visual, semantic?) strategy (Salame & Baddeley, 1986).

References Allport, D. A. (1984). Auditory-verbal short-term memory and conduction aphasia. In H. Bouma and D. G. Bouwhuis (Eds.), Attention

and performance X. Control of language processes

(pp. 313-325). Hillsdale, NJ: Erlbaum. Baddeley, A. D. (1982). Amnesia: A minimal model and an interpretation. In L. S. Cermak (Ed.), Human memory and amnesia (pp. 305—336). Hillsdale, NJ: Erlbaum.

Baddeley, A. D., Papagno, E., & Vallar, G., (1988). When long-term learning depends on shortterm storage. Journal of Memory and Language, 27,

586—595.

Baddeley, A. D., Vallar, G., & Wilson, B. (1987). Sentence comprehension and phonological memory: Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and performance XII. The psychology of reading (pp. 5 0 9 - 5 2 9 ) . H o w e and London: Erlbaum.

Basso, A., Capitani, E., & Vignolo, L. A. (1979). Influence of rehabilitation on language skills in aphasic patients. Archives of Neurology, 36, 190-196. Basso, A., Spinnler, R , Vallar, G., & Zanobio, M. E. (1982). Left hemisphere damage and selective impairment of auditory—verbal short-term memory: A case study. Neuropsychologia, 20, 263-274. Benton, A. L., & Hamsher, K. (1978). Multilingual aphasia examination. Iowa City: Benton Laboratory of Neuropsychology. Berndt, R. S. (1985). Working memory and sentence comprehension. Paper presented at the Second Venice Conference on Cognitive Neuropsychology, March 25—29. Berndt, R. S. (1988). Repetition in aphasia: Implications for models of language processing. In F. Boiler and J. Grafman (Eds.), Handbook of Neuropsychology (pp. 329-348). Amsterdam: Elsevier. Bever, T. G. (1970). The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language (pp. 279-362). New York: Wiley. Blumstein, S. E., Baker, E., & Goodglass, H. (1977). Phonological factors in auditory comprehension in aphasia. Neuropsychologia, 15, 19-30. Bortolini, U., Tagliavini, C, & Zampolli, A. (1972). Lessico di frequenza della lingua italiana contemporanea. Milano: Garzanti. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A,

705-737.

Caplan, D., Vanier, M , & Baker, C. (1986). A case study of reproduction conduction aphasia: II. Sentence comprehension. Cognitive Neuropsychology, 3, 129-146. Caramazza, A., Basili, A. G., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271. Caramazza, A., & Berndt, R. S. (1985). A multicomponent deficit view of agrammatic Broca's aphasia. In M. L. Kean (Ed.), Agrammatism (pp. 27-63). Orlando: Academic Press. Clark, H. H., & Clark, E. V. (1977). Psychology and language: An introduction to psycholinguistics.

New York: Harcourt Brace Jovanovich. Cornoldi, C. (1974). Imagery values for 310 Italian nouns. Giornale Italiano di Psicologia, 1, 211-225. Craik, F. I. M , & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 6 7 1 - 6 8 4 .

DeRenzi, E. (1977). Le amnesie. In E. Bisiach et al. (Eds.), Neuropsicologia Clinica (pp. 199-246) Milano: Franco Angeli.


475

DeRenzi, E., & Faglioni, P. (1978). Normative data and screening power of a shortened version of the token test. Cortex, 14, 41-49. Goodglass, H., Gleason, J. B., Bernholtz, N., & Hyde, M. R. (1972). Some linguistic structures in the speech of Broca's aphasics. Cortex, 8, 191-192. Friedrich, F. J., Glenn, C. G., & Marin, O. S. M. (1984). Interruption of phonological coding in conduction aphasia. Brain and Language, 22, 266-291. Friedrich, F. J., Martin, R., & Kemper, S. J. (1985). Consequences of phonological coding deficit in sentence processing. Cognitive Neuropsychology, 2, 385-412. Hitch, G. J. (1978). The role of short-term working memory in mental arithmetic. Cognitive Psychology, 10, 302-323. Hitch, G. J., & Baddeley, A. D. (1976). Verbal reasoning and working memory. Quarterly Journal of Experimental Psychology, 28, 603—621. Linebarger, M. C, Schwartz, M. F., & Saffran, E. M. (1983). Sensitivity to grammatical structure in so-called agrammatic aphasics. Cognition, 13, 361-392. McCarthy, R. A., & Warrington, E. K. (1987a). The double dissociation of short-term memory for lists and sentences. Brain, 110, 1545-1563. McCarthy, R. A., & Warrington, E. K. (1987b). Understanding: A function of short-term memory? Brain, 110, 1565-1578. Martin, R. (1987). Articulatory and phonological deficits in short-term memory and their relation to syntactic processing. Brain and Language, 32, 159-192. Marslen-Wilson, W. (1984). Function and process in spoken word recognition. In H. Bouma & D. G. Bouhwuis (Eds.), Attention and performance X. Control of language processes (pp. 125-150). Hillsdale, NJ: Erlbaum. Miceli, G., Gainotti, G., Caltagirone, C, & Masullo, C. (1980). Some aspects of phonological impairment in aphasia. Brain and Language, 11, 159-169. Novelli, G., Papagno, C, Capitani, E., Laiacona, M., Vallar, G., & Cappa, S. F. (1986a). Tre test clinici di ricerca e produzione lessicale. Taratura su soggetti normali. Archivio di Psicologia, Neurologia e Psichiatria, 47, 477-506. Novelli, G., Papagno, C, Capitani, E., Laiacona, M., Cappa, S. F., & Vallar, G. (1986b). Tre test clinici di memoria verbale a lungo termine. Taratura su soggetti normali. Archivio di Psicologia, Neurologia e Psichiatria, 47, 278-296. Ostrin, R. K., & Schwartz, M. F. (1986). Reconstructing from a degraded trace: A study of sentence repetition in agrammatism. Brain and Language, 28, 328-345. Parisi, D., & Pizzamiglio, L. (1970). Syntactic comprehension in aphasia. Cortex, 6, 204—215. Pisoni, D. B. (1973). Auditory and phonetic memory codes in the discrimination of consonants and vowels. Perception and Psychophysics, 13, 253-260. Previdi, P. (1975). Contributo alia elaborazione di un test internazionale dell'afasia. Taratura dei normali. M.D. thesis, University of Modena, Italy, Faculty of Medicine. Pulman, S. G. (1987). Computational models of parsing. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 3, pp. 159-231). London: Erlbaum. Safifran, E. (1985). Short-term memory impairment and language comprehension: Specifying the nature of the interaction. Paper presented at the Second Venice Conference on Cognitive Neuropsychology, March 25-29. Saffran, E., & Marin, O. S. M. (1975) Immediate memory for word lists and sentences in a patient with a deficient auditory short-term memory. Brain and Language, 2, 420-433. Salame, P., & Baddeley, A. D. (1986). Phonological factors in STM: Similarity and the unattended speech effect. Bulletin of the Psychonomic Society, 24, 263-265. Sartori, G. (1984). La lettura. Bologna: II Mulino. Savin, H. B., & Perchonock, E. (1965). Grammatical structure and the immediate recall of English sentences. Journal of Verbal Learning and Verbal Behavior, 4, 348-353. Scholes, R. ]. (1978). Syntactic and lexical components of sentence comprehension. In A.

476


Caramazza & E. B. Zurif (Eds.), Language acquisition and language breakdown. Parallels and divergencies (pp. 163—193). Baltimore: The Johns Hopkins University Press. Schwartz, M. F., Linebarger, M. C, & Saffran, E. M. (1985). The status of the syntactic deficit theory of agrammatism. In M.-L Kean (Ed.), Agrammatism (pp. 83-124). Orlando: Academic Press. Schwartz, M. F., Linebarger, M. C, Saffran, E. M., & Pate, D. S. (1987). Syntactic transparency and sentence interpretation in aphasia. Language and Cognitive Processes, 2, 85-113. Shallice, T. (1979). Neuropsychological research and the fractionation of memory systems. In L. C. Nilsson (Ed.), Perspectives on memory research (pp. 257-277). Hillsdale, NJ: Erlbaum. Vallar, G., & Baddeley, A. D. (1984a). Fractionation of working memory: Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-161. Vallar, G., & Baddeley, A. D. (1984b). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Vallar, G., & Baddeley, A. D. (1987). Phonological short-term store and sentence processing. Cognitive Neuropsychology, 4, 417-438. Vallar, G., & Cappa, S. F. (1987). Articulation and verbal short-term memory: Evidence from anarthria. Cognitive Neuropsychology, 4, 55—77. Vallar, G., & Papagno, C. (1986). Phonological short-term store and the nature of the recency effect. Evidence from neuropsychology. Brain and Cognition, 5, 428-442. Vallar, G., Papagno, C, & Cappa, S. F. (1988). Latent dysphasia after left hemisphere lesions. A lexical-semantic and verbal memory deficit. Aphasiology, 2, 463-478. Warrington, E. K., Logue, V., & Pratt, R. T. C. (1971). The anatomical localisation of selective impairment of auditory verbal short-term memory. Neuropsychologia, 9, 377—387. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory verbal short-term memory. Brain, 92, 885-896. Watkins, O. C., & Watkins, M. J. (1977). Serial recall and the modality effect: Effect of word frequency. Journal of Experimental Psychology, 3, 712-718. Wickelgren, W. A., (1965). Short-term memory for phonemically similar lists. American Journal of Psychology, 78, 567-574.

18. Working memory and comprehension of spoken sentences: investigations of children with reading disorder STEPHEN CRAIN, DONALD SHANKWEILER, PAUL MACARUSO, AND EVA BAR-SHALOM

18.1. Introduction Our goal is to investigate the role of the verbal working memory system in sentence comprehension, by presenting a model of working memory in sufficient detail to allow specific predictions to be made and tested. In testing this account, we draw on experimental methods that have recently been used in research on language development. These methods are designed to control the various sources of potential difficulty in the standard laboratory tasks used to assess children's grammatical knowledge and their use of this knowledge in sentence comprehension. We illustrate how our proposals about working memory, together with the recent innovations in method, allows us to infer that abnormal limitations in phonological processing, and not absence of grammatical knowledge, are at the root of the difficulties in spoken sentence understanding that are apparent in children with reading disability. Since reading problems are most transparent at the beginning stages of learning to read, we focus our attention there, by investigating the linguistic abilities of poor readers in the early school years. By "poor readers" we mean children who show a marked disparity between their measured level of reading skill and the level of performance that might be expected in view of their intelligence and opportunity for instruction. Our research compares performance by these children with age-matched controls — children who are proceeding at the expected rate in the acquisition of reading skills (for discussion of the issues regarding subtypes of reading disability and choice of control groups, see Shankweiler, Crain, Brady & Macaruso, in press). Much of the research on poor readers finds the source of their problems in the Portions of this research were supported by a Program Project Grant to Haskins Laboratories from the National Institute of Child Health and Human Development (HD-01994). We wish to thank the second-grade students and teachers, reading instructors, and administrators at the Coventry, CT, elementary schools. We also thank Suzanne Smith for her help with Experiment 3, Henry Hamburger for extensive discussion of the issues raised in this paper, and Brian Butterworth, Myrna Schwartz, and Tim Shallice for their comments on an earlier draft.

477

478

Crain, Shankweiler, Macaruso, and Bar-Shalom

language domain, not in the area of visual perception or in general analytic ability. We shall take this for granted (for reviews, see Shankweiler and Liberman, 1972; Vellutino, 1979; Perfetti, 1985). Within the language domain, many sources of evidence converge on the conclusion that poor readers' problems reflect deficiencies in phonological processing (see Liberman & Shankweiler, 1985, and Stanovich, 1982, for reviews). However, there is one finding that raises the possibility that their limitations extend beyond phonological processing to syntactic processing as well: the discovery that poor readers characteristically fail to comprehend complex spoken sentences accurately under some circumstances. This finding has led researchers to the hypothesis that these children have not mastered all of the complex syntactic properties of the adult grammatical system (Byrne, 1981; Fletcher, Satz & Scholes, 1981; Stein, Cairns, & Zurif, 1984). We have called this the structural lag hypothesis (SLH). The SLH provides a coherent account of some factors that may make reading hard to learn and that may distinguish good and poor readers. This hypothesis attributes poor readers' difficulties in spoken language comprehension to their level of attainment in the acquisition of syntax. According to the SLH, language is acquired in stages, beginning with simple syntactic structures and culminating only when the most complex structures have been mastered. To explain why language acquisition conforms to a developmental schedule, the SLH endorses the idea that syntactic structures are ordered in inherent complexity. The late emergence of a structure in the course of language development is taken as an indicator of its relative complexity as compared to structures that appear earlier. It is clear that the SLH deserves serious consideration. Reflecting some common assumptions about language acquisition and linguistic complexity, this hypothesis makes the following prediction about the language-related difficulties of poor readers: The linguistic structures that beginning readers and unsuccessful older readers will find most difficult are just those that appear last in the course of language acquisition. Thus, the SLH would point to findings of language acquisition studies that suggest that some syntactic structures emerge later than others in language development, and to studies showing the late mastery of these structures by poor readers.1 Although the SLH gives a plausible account of some of the difficulties encountered by poor readers, it has a major limitation. It gives no way to tie together poor readers' problems at the level of the sentence with their problems at the level of the word. Specifically, the postulated syntactic deficit of poor readers is independent of their deficit in processing phonological information. This means that the SLH abandons the possibility of achieving a unitary explanation of the whole symptom picture of reading disability. In our research we have sought support for an alternative hypothesis, which we call the processing limitation hypothesis (PLH).2 In contrast to the SLH, this hypothesis attempts to tie together all of the symptoms of the poor reader, viewing them as derived from inefficient processing of phonological structures. Several problems can be

Working memory and reading disorder

479

securely tied to a deficiency in phonological processing, including the difficulties that poor readers have in word segmentation, object naming and verbal working memory. Consider first their well-attested problems in bringing phonological segments to consciousness. It has been shown in several language communities that on analytic tests requiring conscious manipulations of the phonemic structure of spoken words, poor readers are less proficient than children who are more successful in learning to read (Lundberg, Oloffson, & Wall, 1980; Bradley & Bryant, 1983; Morais, Cluytens, & Alegria, 1984; Cossu, Shankweiler, Liberman, Tola, & Katz, 1988). Another problem that has claimed a good deal of attention is their impaired performance on tests of object naming (Jansky & de Hirsch, 1972; Denckla & Rudel, 1976; Wolf, 1981). Analysis of the errors reveals that the mistakes are often based on phonological confusions rather than on semantic confusions (Katz, 1985). This suggests that this problem, too, is a manifestation of underlying phonological impairment. This same line of reasoning also applies to verbal working memory. Because the verbal working memory system depends on the ability to gain access to phonological structure and use it to (briefly) maintain linguistic information, we might expect people who have phonological difficulties to show various limitations on tests of ordered recall (Conrad 1964,1972; Liberman, Mattingly, & Turvey, 1972; Baddeley, 1986). For poor readers, as in other language-impaired populations, there is ample evidence in the literature testifying to deficiencies in short-term retention of verbal materials. Differences in recall have been obtained with a variety of verbal materials, including words and spoken sentences, but they are not typically found with materials that cannot be coded linguistically (see Liberman, Shankweiler, Liberman, Fowler, & Fischer, 1977; Wagner & Torgesen, 1987). Moreover, there is direct evidence from memory experiments that poor readers in the beginning grades are less affected by phonetic similarity (rhyme) than age-matched good readers. This is another indication of their failure fully to exploit phonological structure in working memory (Shankweiler, Liberman, Mark, Fowler, & Fischer, 1979; Mann, Liberman, & Shankweiler, 1980; Olson, Davidson, Kliegl, & Davies, 1984). In addition to these symptoms, we noted earlier that poor readers are sometimes unable to comprehend spoken sentences as well as comparable good readers. Our central aim in this chapter is to explain how the difficulties of poor readers in understanding spoken sentences may be derived from deficient phonological processing. On the face of it, these difficulties might seem to require another kind of explanation. But suffice it to say here that the findings of our recent research, including the results of the experiments presented in section 18.4, have persuaded us that the source of their spoken language comprehension failures is also tied to an underlying deficiency in phonological processing, as proposed by the PLH, and is not the result of a lag in syntactic development, as predicted by the SLH. Given these sharply contrasting hypotheses about poor readers' problems in sentence comprehension, we now turn to the kinds of evidence that can decide between

480


them. One source of evidence may be obtained by examining the pattern of errors good and poor readers make in response to sentences of different types. If poor readers suffer from a limitation in processing, it makes sense that the pattern of errors on different structures should be similar for both groups, with the poor readers showing a decrement of roughly the same magnitude on each sentence type. The prediction that the error pattern of poor readers should parallel that of good readers serves as the foundation for one of the experiments reported in section 18.3. Another research strategy that has proved useful in distinguishing between the PLH and SLH is to examine the performance of good and poor readers on laboratory tasks that differ in how severely they tax the resources of working memory. Marked improvement in performance in the face of reduction in memory load is anticipated by the PLH but not by the SLH. In the absence of requisite structures, poor readers should fail in comprehension even when memory load is minimal. On the other hand, if a processing limitation is the source of the problem, even the most unskilled reader should prove competent with highly complex linguistic constructions in spoken language, within the constraints imposed by their limitations in processing capacity. This prediction, too, is tested in the experiments we report here. Before we give details of the experiments, it will be useful to describe our view of the working memory system and its role in language processing.

18.2. Organization of the language apparatus Our conception of the language apparatus shares much common ground with the modularity proposal advanced by Fodor (1983). It grows out of a biological perspective on language that has long guided research on speech at Haskins Laboratories. According to this viewpoint, the language faculty functions autonomously in the sense that it is supported by special brain structures and operates according to principles that are specific to it and not shared by other cognitive systems. One source of evidence for this conception of modularity comes from studies of speech perception (Mattingly & Liberman, 1988). Another source is from the study of aphasia and related disorders where there is evidence that a circumscribed lesion in the left hemisphere may selectively perturb certain aspects of language performance, leaving other linguistic and nonlinguistic abilities relatively intact (Marin, Saffran, & Schwartz, 1976; Linebarger, Schwartz, & Saffran, 1983; Shankweiler Crain, Gorrell, & Tuller, 1989). There is also evidence that ability to process language may be preserved in the face of massive losses to other systems, as in cases of "isolation aphasia" (e.g., Whitaker, 1976). Another source of evidence for modularity comes from the study of language development, where it has been found that complex linguistic principles emerge in young children at a characteristic pace that is independent of the emergence of other cognitive systems or principles (e.g., Hamburger & Crain, 1984). Also important are


481

research findings demonstrating children's early mastery of linguistic principles that go beyond the data provided by the environment (e.g., Crain & McKee, 1985; Crain & Nakayama, 1987; Crain, Thornton, & Murasugi, 1987). Taken together, all of these findings sustain the notion that language is a biologically coherent system, as the modularity proposal maintains. An extension of the modularity proposal supposes that the language faculty itself is composed of several autonomous subcomponents (or submodules). This componential view of sentence production and comprehension postulates several structures and processors. Roughly, each structure is a stored system of rules and principles corresponding to a level of linguistic representation: phonology, syntax, and semantics. In addition to the independent levels of structural representation, the language apparatus contains special processors, including the phonological, syntactic, and semantic parsers. Each parser is a special-purpose device for rule access and ambiguity resolution corresponding to a specific level of representation. Each parser operates on principles and rules in assigning constituent structure to linguistic input. Because the parsers operate on constituent structure, and not on sequences of words themselves, we can understand sentences of great length, but can retain only relatively short lists of unrelated material. Two further architectural features of the language apparatus are essential to our explanation of the difficulties poor readers have in sentence understanding. We assume, first, that the various submodules are arranged in a hierarchical fashion, with a unidirectional and vertical ("bottom up") flow of information such that a lower level passes results to higher levels but not the reverse. It is also critical to our view that transactions between the parsers take place "on-line," with the results of low-level analyses being quickly discarded, to make room for subsequent input (for related discussion, see Carpenter & Just, 1988).

18.2.1. How working memory functions in the language processing system In keeping with the modularity hypothesis, we conceive of verbal working memory as a domain-specific system that subserves the language apparatus.3 The primary function of verbal working memory is to facilitate the extraction of a meaning representation corresponding to the linguistic input. Assuming that the extended modularity hypothesis is correct, this involves the interaction of several structures and processors. As we conceive of it, verbal working memory is an active processing system in which the analysis of verbal material by these structures and processors takes place during language processing. In common with other contemporary approaches, we assume that there are two components to the working memory system (Baddeley & Hitch, 1974; Perfetti & Lesgold, 1977; Daneman & Carpenter, 1980; Baddeley, 1986; Carpenter & Just, 1988). First, there is a storage buffer where rehearsal and initial (phonological) analysis of

482


phonetically coded information take place. This buffer has the properties commonly attributed to short-term memory. It can hold information only briefly, perhaps only for 1—2 sec, in the order of arrival, unless the material is maintained by continuous rehearsal. The limits on capacity of the buffer mean that information must be rapidly encoded in a more durable form if it is to be retained for subsequent higher-level analysis. Our conception of the storage buffer bears obvious similarities to other discussions in the literature. What is new in our conception of the verbal working memory system concerns its other component. We view this as a control mechanism whose primary task is to relay the results of lower-level analyses of linguistic input upward through the language apparatus. Its regulatory duties begin at the lowest level by bringing phonetic (or orthographic) input into contact with phonological rules, for word-level analysis. Phonologically analyzed information must be rapidly transferred out of the storage buffer and shunted to the syntactic processor, at the same time freeing the storage area to accept the next chunk of phonetic material. By synchronizing information flow with input, the control mechanism is able to push results upward through the system rapidly enough to promote on-line extraction of meaning (Marslen-Wilson & Tyler, 1980; Wingfield & Butterworth, 1984; Crain & Steedman, 1985). In processing spoken language, on-line parsing explains how individuals with drastically curtailed working memory capacity — capable of holding only two or three items of unstructured material — are sometimes able to comprehend sentences of considerable length and complexity (Martin, 1985; this volume, chapter 15; Saffran, 1985). Previous research has found it paradoxical that aphasic patients with a severely restricted phonological short-term store are sometimes capable of understanding at a level far exceeding what would be expected on the basis of their span limitations. This result is fully consistent with our model of working memory. In reading, on-line processing of syntactic and semantic representations necessarily depends on prior orthographic and phonological processing. Until the reader is proficient in decoding from print, we would expect that reading is more demanding than speech of working memory resources. Sometimes it is assumed that print confers an advantage because the reader can look back. It is important to appreciate, however, that only the skilled reader can exploit the opportunity to reexamine sentences in text that were not successfully parsed on first reading. In the unskilled reader, the working memory system is usually preoccupied with orthographic decoding.

18.3. Identifying the source of reading disability We are now in a position to show how the architectural arrangement of the language faculty can be exploited to provide an explanation of the sentence comprehension difficulties of poor readers. A modular view of the language apparatus raises the


483

possibility that a single component may be the source of the entire symptom complex that characterizes reading disability. It is clear that failures in sentence comprehension could arise, in principle, from a deficit (or deficits) at any level that ultimately feeds into the semantic component. It is also conceivable, however, that the entire symptom complex of poor readers, including their difficulties in spoken language comprehension, implicates the phonological component. Let us explain how. Recall that the submodules of the language faculty act in strict sequence ("bottom up") to assign a partial structural analysis, which can then be passed on to higher levels. To keep information flowing smoothly, the control mechanism must avoid unncessary computation that would delay the rapid extraction of meaning. This means that, in ordinary circumstances, the working memory buffer need not store many segments of unanalyzed linguistic material. But suppose that the phonological analysis of material in the buffer is impeded for some reason. Given the architectural features of the language apparatus we have proposed, this would also have the effect of curtailing the operation of higher-level analyses of verbal material. In short, the functions of an otherwise intact system would be depressed. This is exactly what happens in cases of reading disability, in our view. Since poor readers are deficient in setting up and organizing phonological structures, sentence comprehension is compromised because inefficient phonological analysis creates a "bottleneck" that constricts information flow to higher levels of language processing. Although the remaining components of the language apparatus may be completely intact, their operation will be hobbled by poor readers' limitations in phonological processing. In effect, a lower-level deficit in phonological processing masquerades as a deficit at higher levels. At this point, however, we cannot rule out the possibility that the comprehension problems of poor readers are caused by a deficiency in some other component of the language apparatus (e.g., in syntactic parsing). But since poor readers' comprehension problems follow automatically from their well-attested limitations in phonological processing, it becomes unnecessary to postulate additional impairments within the language system. Moreover, we will provide evidence of the acquisition of complex syntax for both good and poor readers, as anticipated by the modularity hypothesis (see also Shankweiler & Crain, 1986). It is important to underscore another expectation of our model, namely, that poor readers should display successful comprehension on sentences that are not especially taxing of phonological resources. This distinguishes our view from other proposals about the relation between working memory and sentence comprehension (e.g., Baddeley, Vallar, & Wilson, 1987; Vallar, Basso, & Bottini, this volume, chapter 17). As long as the control mechanism of working memory is intact, even persons with abnormal limitations in phonological short-term storage capacity should be able to understand sentences of considerable complexity, if they do not impose excessive demands on phonological resources. Since the control mechanism of working memory

484


plays such a prominent role in explaining why impaired comprehension should appear on specific sentences and not others, it will be worthwhile to describe it in more detail.

18.3.1. The compiling analogy Pursuing an analogy with the compiling of programming languages, we view the control mechanism of working memory as a control structure whose function is to carry out a series of translations, each being a translation from a relatively high-level language (the source language) to a more detailed language (the target or object language). This concept is familiar in computer science, where high-level languages like Pascal or Lisp are compiled into lower-level languages such as assembly language or machine language. But the notion of compiling is quite general, and has proved useful in modeling human language processing as well. Cognitive compiling occurs in natural language processing in experiments in which a subject is asked to act out the interpretation of a sentence using toys and figures provided in the experimental workspace. Here, the source language (e.g., English) must be translated into a more detailed language that underlies the overt actions the subject makes in response to the input. We will refer to the mental language that serves as the target language for observable physical actions as the language of plans. In our view, several interesting properties of the control component of working memory can be illuminated by considering the translation between input sentences and the plans that they evoke (see Hamburger & Crain, 1984, 1987, for further discussion and for empirical data).4 In the paragraphs that follow, we focus on the difficulties that may arise for the executive component of working memory in the process of translating from language input to plans. We first consider situations that are amenable to simple translation between source and target language (Hamburger & Crain, 1984). Then we will look at particular linguistic forms that deviate from the best-case scenario, thereby exacting a toll from the resources of working memory. In the simplest case, (a) each well-formed fragment of target language code is associated with a single constituent of source language code, (b) the fragments of target language code can be concatenated to form the correct representation of the input, (c) the fragments can be combined in the same order they are accessed, and (d) each fragment is processed immediately after it is formed, permitting the source code to be discarded. These conditions form a straightforward process of sequential look-up-andconcatenation. Rarely, however, are all the conditions met in ordinary language. And when they are not, the computations involved in reaching the target code (e.g., the semantic interpretation or plan associated with a linguistic expression) could stretch the resources of verbal working memory. It will not be possible to spell out each condition in detail, but it may be helpful to make a few remarks about each, focusing on the linguistic constructions that appear in the experiments reported in section 18.4.


485

(a) The first condition is an isomorphism between any two levels of representation. Correspondence of this kind is maintained between syntactic and semantic constituency in Montague Grammar in order to provide a systematic account of the assignment of semantic values to linguistic expressions. The foundation of this account is the principle of compositionality, which states that the meaning of a linguistic expression is determined from the meanings of its constituent expressions (and their mode of combination). Despite the appeal of a straightforward relationship between the syntax and semantics, linguists working in the generative framework have argued that syntax and semantics are largely autonomous. In our terms, this is liable to add to the complexity of translating between syntactic and semantic structural representations.5 (b) Whether or not the first condition is met, it seems reasonable to suppose that the simplest way to combine nodes of the target language is by concatenation. Unfortunately, it is clearly not possible to concatenate meanings even in parsing simple natural language phrases like expensive socks or second bear. Since expensive socks are not expensive, it would be a mistake to evaluate this phrase on a word-by-word basis, for example, by forming a semantic value for expensive (say, the set of expensive things), and then combining this with the semantic value of the following word, socks. Similarly, the second bear is not necessarily in second position in an ordered array. On occasion, concatenation of word meanings is possible, for example with NPs that contain absolute adjectives, like green, fuzzy, Albanian, and so on, where the denotation of the adjective is not dependent on the linguistic context (e.g., naked Albanian wrestler). But since no unique semantic value can be given to relative adjectives, (e.g., expensive) or to ordinals, the human sentence processing mechanism must hold off interpreting these prenominal modifiers until the head noun has been received. Translations that require the parser to splice together dissociated pieces of code at some level also violate the simple process of look-up-and-concatenation (see Hamburger & Crain, 1984, 1987). An example of this source of distress for working memory is second striped ball An analysis of the logical structure of the plan corresponding to this phrase shows it to consist of a nested loop structure in which fragments of plan associated with striped ball are inserted into the piece of code associated with the ordinal second. Breaking apart the code needed to increment a counter is required in order to test objects (for stripedness and ballhood), to ensure that the counter is advanced only as green balls are located. This process is referred to by Hamburger and Crain as compiling discontinuity.6 Empirical support for the claim that phrases like this present difficulties for young children comes from several acquisition studies that find that children often choose object (b) from an array like the following in response to a request such as "point to the second striped ball" (Roeper, 1972; Matthei, 1981). That is, children incorrectly select the object that is second and striped and a ball, instead of the second of the striped balls (d).

486


(a)

(b)

(c)

(d)

(e)

(f)

Hamburger and Crain (1984) suggest that this error may be the spurious result of premature interpretation of the second as applying to the entire set of objects in the array. They found that children were dissuaded from this concatenative response if they were asked to handle the subsets of objects before these were placed in the array. This presumably inhibits premature execution, since it is unclear in this circumstance which (sub)set of objects the ordinal second modifies. (c) There is another locus of difficulty in translating from a source language form to target language code: Condition (c). This condition requires the order of concatenation of plans to mirror the linguistic input. Let us call any violation of this condition a sequencing problem. In addition to compiling discontinuity, a sequencing problem arises in the example of second striped ball As we saw, the locus of the difficulty with this phrase is not in either the source code or the object code, but only in their relationship. This suggests a possible alternative to the merging of code in the formation of a plan. The alternative would be to hold onto the code associated with the ordinal second until after the remaining elements have been combined (establishing the data structure for striped ball). But setting aside a constituent to await the preparation of other elements with which it is to associate is assumed to be costly of memory resources. In terms of the model of working memory we are considering, this would constitute a violation of Condition (c) and, as a consequence, also Condition (d).7

18.3.2. Relative clauses A second example of the difficulties a sequencing problem may pose for comprehension is from a study on the acquisition of restrictive relative clauses. This study (Hamburger & Crain, 1982) discovered that many children who performed the correct actions associated with sentences like (1) often failed, nevertheless, to act out these events in the same way as adults. 1. The cat scratched the dog that jumped through the hoop.

Most 3-year-olds and many 4-year-olds acted out this sentence by making the cat scratch the dog first, and then making the dog jump through the hoop. Older children and normal adults act out these events in the opposite order, the relative clause before the main clause. Intuitively, acting out the second-mentioned clause first seems conceptually more correct, since the dog that jumped through the hoop is what the cat scratched.


487

It is reasonable to suppose that this kind of conflict between the order of mention and conceptual order (and most appropriate order of execution) stresses working memory because both clauses must be available long enough to enable the hearer to formulate the plan that represents the conceptual order. Presumably, this difference between the responses of children and adults reflects the more severe limitations in children's working memory in coping with sentences that pose sequencing problems. The response we have characterized as conceptually correct (with the relative clause acted out first) requires the formation of a two-slot template, and a specification of the particular sequence in which the actions are to be carried out. Since on the simple lookup-and-concatenate scenario, processing occurs on-line (i.e., on a left-to-right word-byword basis), it seems to us that the difficulty presented by the conflict between order of mention and conceptual order occurs because the information in both clauses must be held in memory long enough to put the first-mentioned action into the second slot. If memory is overloaded, a subject may adopt a default procedure of acting out clauses in their order of mention — that is, according to the simple translation routine of look-upand-concatenate. To explain this phenomenon we draw on another analogy to translation among programming languages. Here we appeal to the distinction between compiling, which completes the translation before starting to execute, and interpreting, which interleaves translation and execution. We can use this distinction in explaining children's conceptually incorrect responses to sentences like (1). Since children are unable to hold information long enough in working memory to compile a conceptually correct plan, it makes sense to suppose that they opt instead to interpret in cases like (1). Consistent with this supposition is the observation that children often begin to act while the sentence is still being uttered.

18.3.3. Temporal terms A third example of the sequencing problem arises with sentences containing the temporal terms before and after. These terms explicitly dictate the conceptual order of events, and they too may present a sequencing problem by introducing conflicts between conceptual order and order of mention. This is illustrated by sentence (2). 2. Jabba flew the X-Wing fighter after Hans Solo sped away in the Millennium Falcon.

A sequencing problem arises in (2) because the order in which events are mentioned is opposite to the conceptual order. Again, research in language acquisition has found that young children frequently interpret these sentences in an order-of-mention fashion (Clark, 1970; Johnson, 1975). As with relative clause sentences, it is likely that this response reflects an inability to hold both clauses in memory long enough to formulate a plan for acting them out in the correct conceptual order. Once again, children's failure

488


to segregate translation and execution explains their incorrect, default decision to adopt the simple look-up-and-concatenate translation. In this case, however, an alternative account of children's difficulties has been proposed. It has been argued that a structural explanation, and not a processing explanation, is called for. Proponents of the structurally based explanation (Amidon & Carey, 1972) point out that the same children who failed to act out sentences like (2) correctly emitted a high rate of correct responses to sentences similar in meaning, but with simpler syntactic structure, as in (3). 3. Push the motorcycle last; push the helicopter first. There is direct evidence that processing factors, and not lack of syntactic competence, are responsible for children's errors in comprehending sentences with temporal terms. The evidence is this: Once processing demands are reduced, most 4- and 5-year-old children usually give the correct response to test sentences like (4) and (5). 4. Push the helicopter after you push the motorcycle. 5. Before you push the motorcycle, push the helicopter. To minimize processing load, one must take cognizance of a presupposition on the use of temporal terms. The presupposition associated with sentences (4) and (5) is that the hearer intends to push a motorcycle. To satisfy this presupposition, one simply has to ask the child in advance to select one of the toys to play with before each trial. These sentences are felicitous only if the subject has first indicated an intent to play with a motorcycle prior to receiving the test sentence. When young children were given this contextual support, they displayed unprecedented success in comprehending sentences with the temporal terms before and after (Crain, 1982; Gorrell, Crain, & Fodor, 1986).8 The same finding was also obtained in a recent study of mentally retarded adults (Crain, 1986). We should also mention a superficial linguistic property that forestalls premature execution, and thereby eases the burden on working memory, in sentences like (5). This is the presence of a temporal term in the initial clause, which indicates that a two-slot template is required. Notice that in the corresponding sentence with after, the temporal conjunction appears in the second clause. The account of memory difficulties we have proposed would therefore lead us to expect this type of sentence to be harder, especially if it contains after. This prediction is confirmed in the experiments reported in section 18.4. In our discussion of the control component of working memory, we have assigned to it as few combinatorial duties as possible. This makes it essentially a simpleminded traffic controller for symbolic representations that are being composed within the submodules of the language apparatus. It is also apparent that the structure-building operations that take place within these modules are frequently at odds with the efficient


489

9

management of information flow. Much is gained, however, by having them incorporated into the language faculty, since they supply the generative capacity for producing and understanding (an uncountable number of) novel sentences.

18.3.4. Garden path effects Corresponding to each intermediate level of representation is a processing mechanism, or parser. The task of each parser is to assign structure to the incoming code as it is being transmitted from the next-lower parser. This analysis phase of the compiling process was aptly referred to by Miller (1956) as chunking. The syntactic parsing mechanism is probably the best understood of the parsers. This mechanism consists of a number of routines for accessing syntactic rules and principles and resolving ambiguities that arise when more than one analysis is compatible with the current input. We assume that access to rules during on-line processing uses hard-wired portions of the language apparatus - almost reflexlike in character - that are sparing of processing capacity in most cases. However, natural languages permit massive local ambiguity, and, despite the flexibility that this allows, this surely incurs some cost to memory resources. In fact, there is considerable evidence that local ambiguities are quickly resolved, perhaps within one or two words after they arise. One parsing tendency that seems to have evolved to meet the twin exigencies of ambiguity and working memory limitations is called Right Association (see Kimball, 1973; Frazier, 1978). Right Association explains why listeners or readers connect an incoming phrase as low as possible in the phrase marker that has been assigned to the preceding material. This "strategy" reflects the functional architecture of the language apparatus, which has many computations to perform and little space for their compilation and execution. As a result, strategies like Right Association dictate that incoming material is integrated into the most readily available (i.e., local) node in the phrase marker under construction. So, for example, Right Association dictates that the adverb yesterday will be attached to the lower of the two VPs in the ambiguous sentence (6) and will therefore be interpreted as related to the last-mentioned event. 6. Bush said he apologized to the UAW, yesterday.

In keeping with Right Association, there is a strong tendency for people to interpret (6) to mean that Bush apologized yesterday, and not that he uttered a sentence to that effect yesterday. It is reasonable to suppose that memory limitations promote rapid on-line integration of material into a structural representation. Although parsing strategies may enable the parser to circumvent the limitations of working memory, they sometimes introduce problems of their own, because the decision dictated by a strategy may turn

490


out to be incorrect in the light of subsequent input. In this case, the perceiver is led down a garden path by the parser. The existence of "garden path" effects (illustrated in [7]) shows that for some sentences even full knowledge of the grammar is not powerful enough to overcome the liability of a tightly constrained working memory. 7. Bush said that he will apologize to the UAW, yesterday. Recovery from garden paths is possible only within the limits of working memory, because this determines whether the grammatically correct attachment site is still available. Since sentences that tax working memory heavily have been found to present problems for poor readers, they should be less able than good readers to recover from incorrect analyses prompted by parsing strategies like Right Association. Therefore, they should be even more susceptible than good readers to garden path effects. Experiment 3 (reported later) tests this prediction by asking good and poor readers to respond to several types of garden path sentences. An examination of how the test sentences were constructed may help to clarify the logic of this experiment. Suppose you are looking at a picture in which a girl (Mary) is using a crayon to draw a picture of a monkey who is drinking milk through a straw. The corresponding sentence is given in (8). What is the unspecified NP in this situation? Both a crayon and a straw are grammatically well formed, but the analysis favored by Right Association has with NP modifying drinking milk rather than modifying drawing a picture, so the general preference is to cash out the NP as a straw. 8. Mary is drawing a picture of a monkey that is drinking milk with NP. This parsing preference is still present if the NP in (8) is extracted by Wh- movement, as in (9). 9. What is Mary drawing a picture of a monkey that is drinking milk withl The preposition with again coheres strongly with the relative clause, rather than with the main clause. The result is that one is tempted to make an ungrammatical analysis of (9) in which what has been extracted from the relative clause, violating a putative universal constraint on extraction called Subjacency. Research in language acquisition, using a picture verification task, found that many children succumb to this temptation, in an "apparent" violation of Subjacency, responding to (9) by saying "a straw," rather than giving the grammatically correct response, "a crayon" (Otsu, 1981; Crain & Fodor, 1985). This incorrect response clearly bears on the choice of the two hypotheses we are considering about the source(s) of reading disability. Since Subjacency is part of Universal Grammar, the PLH would maintain that it should be adhered to by good and poor readers alike from the earliest stages of language development. On the other hand, the processing limitations of poor readers would lead us to expect them to make more


491

apparent violations of the Subjacency constraint. It is then incumbent on the PLH to show that the relatively poor performance of poor readers on sentences like (9) is due to parsing pressures (viz., the effect of Right Association) rather than to ignorance of universal constraints on syntax. There are two critical ingredients in determining whether responses that violate the Subjacency constraint reflect a processing limitation or, instead, arise from a structural deficit. As noted earlier, if poor readers suffer from a processing limitation, this should be revealed in the pattern of errors across sentence types for both reader groups: Poor readers should show a decrement in performance across sentence types, but there should be no Group x Sentence Type interaction. This pattern emerges from comparison of the responses of the reader groups in Experiment 3. The final ingredient is a demonstration of the grammatical competence of poor readers with the construction under investigation. This is the objective of Experiment 4.

18.4. Applying the working memory model to identify the causes of sentence comprehension failures in poor readers In this section we elaborate on the specific problems that should be incurred by poor readers, given our model of language processing. Four experiments are reported here. These experiments were designed to test between the two competing hypotheses (sketched in section 18.1) about the source of impaired comprehension of spoken sentences by poor readers. Specifically, we ask whether the sentence processing difficulties are due to a syntactic deficit, as claimed by the SLH, or alternatively, whether they reflect a limitation in processing involving working memory, as claimed by the PLH. To explore both possibilities, we selected good and poor readers in the second grade. Reader groups were established on the basis of combined word and nonword scores on the Decoding Skills Test (DST) of Richardson and DiBenedetto (1986). To ensure that the difficulties experienced by the poor reader group could not be attributed to a general deficiency in cognitive function, the reader groups were equated on intelligence as well as on chronological age. (For discussion of the general efficacy of this experimental design, see Shankweiler et al., in press).

18.4.1. Temporal terms (Experiments 1 and 2) In the first two experiments, we were interested to discover how variations in processing load affect the performance of poor readers relative to good readers. In the preceding section, we saw that sentences that contain temporal terms are of particular interest in deciding between the competing hypotheses because (a) temporal terms have been found to emerge late in the course of normal language development, and (b) the source of late mastery has been attributed to syntactic complexity, as the SLH

492


would suggest, as well as to their demands on memory resources, as the PLH would have it. In order to test between these hypotheses, Experiments 1 and 2 used a figure manipulation paradigm, with input sentences containing adverbial clauses with the temporal terms before and after. This task engages children in a game in which they are asked to move toys as dictated by orally presented sentences. The set of objects available in the experimental workspace was the same in both experiments; it comprised nine objects (cars, trucks, horses) of different colors and sizes. The purpose of the first experiment was to establish a baseline of linguistic competence by good and poor readers with sentences containing temporal terms. In the second experiment, we sought to manipulate processing demands in two ways. First, an additional modifier was added to one of the noun phrases in half of the test sentences. This maneuver increased the possibility that subjects would make errors in selecting the objects to be moved on each trial. A second change involved presenting the test sentences in contexts that satisfied the presupposition associated with the use of the temporal term. We hypothesized that poor readers would show appreciable performance gains when processing demands were minimized through the satisfaction of this presupposition. It should be kept in mind that if the poor reader group displayed a sufficiently high level of correct performance in any condition, this would argue against the hypothesis that the relevant syntactic structures are missing from their grammars. But, in addition, an increase in successful comprehension in felicitous contexts would lend credibility to a processing explanation of their performance failures in less than optimal contexts. Each experiment was carried out with a different set of 14 good and 14 poor readers. The mean combined reading scores (on the DST) for the good and poor readers were 92.9 and 23.7 out of 120, respectively (Experiment 1), and 97.2 and 37.9 (Experiment 2). The IQ of subjects was calculated on the basis of their performance on the Peabody Picture Vocabulary Test - Revised (Dunn & Dunn, 1981). Performance on this test was used to ensure that both groups were in a similar IQ range, and that the differences between good and poor readers could not be attributed to different levels of vocabulary knowledge. The mean Peabody scores for good and poor readers were 110.6 and 105.0, respectively (Experiment 1), and 115.4 and 109.0 (Experiment 2).

18.4.2. Experiment 1 The purpose of Experiment 1 was to assess the level of linguistic competence for both reader groups with sentences containing temporal terms. This experiment employed simple NPs and, like many previous studies in the acquisition literature, provided no contextual support.10 In half of the 12 test sentences the order of mention of events corresponded to the conceptual order of events, as in (10). In the other half, the order in which events were mentioned was opposite to the conceptual order, as in (II). 11


493

10. Push the red car before you push the largest horse. 11. Push the smallest horse after you push the blue car. First of all, we found that poor readers made significantly more errors than good readers: F(l, 27) = 4.92, p < .04. However, the overall performance of both groups was high, with the poor reader group performing well above chance {87.5% correct). This indicates that the poor readers were not lacking the necessary competence to successfully interpret temporal term sentences even when they contain inessential prenominal modifiers. The near-ceiling performance of the good reader group (96% correct) meant that subsequent analyses of their error patterns would not be revealing, so the remainder of our analysis focuses on the pattern of errors by the poor readers. In particular, we were interested in determining whether the sentences we expected to be most demanding of memory resources do indeed cause special problems for poor readers. These sentences are the ones that present a conflict between the conceptual order and the order of mention and contain the temporal term after, as in (11). Poor readers' 21.4% errors on these sentences reflects their highest error percentage for any sentence type. In fact, it is a significantly higher error rate than for before sentences (7.1% errors) of the same type: F(l, 13) = 4.50, p = .05. This confirms our expectation that sentences like (11) would be the most difficult for poor readers, given their inherent memory limitations. In Experiment 2, we asked whether a high proportion of correct responses is still characteristic of poor readers in contexts that are even more demanding of working memory resources. If not, the combined data would lend support to the hypothesis that poor readers suffer from a limitation in processing. This difference across tasks would defy explanation on the hypothesis that they suffer from a developmental lag in the acquisition of complex syntax. 18.4.3. Experiment 2 The purpose of this experiment was to test the effects of varying memory demands on good and poor readers.12 According to the account of the working memory system presented earlier, poor readers should be highly sensitive to alterations in processing load that give rise to problems in cognitive compiling. We sought first to exacerbate the processing load beyond the level imposed in Experiment 1 by including an additional prenominal modifier in half of the test sentences. As exemplified in (12) and (13), these sentences contained the ordinal term second, which introduces discontinuity in related statements in the plan that one must compile in order to respond accurately to the noun phrase in which the ordinal appears. We will refer to sentences with NPs of this sort as complex NPs. 12. Push the second smallest horse before you push the blue car. 13. Pick up the second largest truck after you pick up the blue horse.

494


A second change in design was introduced in order to increase the ease of processing. We took advantage of a pragmatic property that is often associated with sentences containing subordinate clauses, namely, their presuppositional content, and we exploited this property to reduce the burden imposed on memory by satisfying the presupposition associated with test sentences of this type, as discussed earlier. In the revised procedure, children are asked, before each test sentence is presented, to identify one object they want to play with in the next part of the game. The experimenter subsequently incorporates this information into the subordinate clause introduced by the temporal term. For instance, sentence (12) would have been presented only after a subject had selected the blue car. This will be referred to as the felicity condition. In the no felicity condition, the presupposition inherent in the use of temporal terms was not satisfied; sentences were presented in the "null context," as in Experiment 1. In the null context, unmet presuppositions must be "accommodated" into the listener's mental model of the discourse setting (Lewis, 1979). In order to compensate for unmet presuppositions, the subject must revise his or her current mental model by averring that the presupposition was met. Updating one's knowledge representation in this way is known to be costly of processing resources (see Crain & Steedman, 1985, and references therein). In light of these considerations, the PLH anticipates a high rate of successful comprehension by both reader groups in the felicity condition, but it predicts that poor readers' performance will suffer in contexts that are more taxing of working memory, as in the no felicity condition. The stimuli in Experiment 2 consisted of 16 sentences with temporal terms before and after. In contrast to Experiment 1, only four sentences were presented in which the order of mention of events was the same as their conceptual order, as in (12). In the remaining 12 sentences, order of mention was opposite to the conceptual order, as in (13). All children encountered the test sentences in both contexts, that is, in the felicity and no felicity conditions. This required two testing sessions for each child, with half of the children receiving contextual support in the first session, and half in the second session. Overall analyses of the results reveal main effects of reader group (F[l, 26] = 14.16, p < .001), felicity (F[l, 26] = 6.50, p < .02), and NP complexity (F[l, 26] = 6.13, p < .02). In addition, there is a marginally significant NP Complexity x Reader Group interaction (F[l,26] = 3.92, p = .06) and a trend toward a Felicity x Reader Group interaction (F[l, 26] = 2.89, p = .10). The main effect of reader group tells us that poor readers performed less well than good readers. However, the main effect of felicity indicates that satisfying the felicity conditions (i.e., reducing the processing demands created by conflicts in sequencing) produced a significant reduction in errors for both groups. The marginally significant Felicity x Reader Group interaction suggests that the satisfaction of presuppositions increased performance for poor readers to a greater extent than for good readers. As displayed in Figure 18.1, there is a greater disparity between their performance for no felicity than for felicity. This lends credibility to the

Working memory and reading disorder 50 -i

495 Good Readers Poor Readers

40 -

30 -

20 0) Q.

10 -

No

Felicity

Felicity

Figure 18.1. Percentage of incorrect responses to temporal term sentences (Experiment 2).

hypothesis that, without contextual support, poor readers' limitations in working memory are exacerbated. The fact that poor readers perform at a success rate of 82.4% when the felicity conditions were satisfied, even when half of the test sentences contained complex NPs, calls into question the claim of the SLH that poor readers lag in their mastery of complex syntactic structures. Averaged over the felicity and no felicity conditions, the main effect of NP complexity tells us that complex NPs evoke significantly more errors than simple NPs. However, the marginally significant NP Complexity x Reader Group interaction (see Figure 18.2) indicates that poor readers were more adversely affected by changes in NP complexity than good readers. The special difficulties that the poor readers displayed with the sentences containing complex NPs presumably reflect the fact that these sentences are more taxing on working memory resources. This is explained by the model of working memory presented earlier as the outcome of the cumbersome problem of compiling a plan that violates another of the conditions for simple translation from input sentence to target plan, the problem of discontinuity compiling. As in Experiment 1, we took a closer look at the sentences that we hypothesized would be the most difficult for poor readers, namely, after sentences that pose a conflict between order of mention and conceptual order, as in (13). Restricting our analyses to the no felicity condition only, we find that poor readers made the most errors on just these sentences. In fact, they produced significantly more errors on them than they produced on before sentences of the same type: F(l, 26) = 6.86, p < .02. This difference is not reflected in the good readers' errors for these same sentences under the same conditions. The significant After vs. Before x Reader Group interaction (F[l, 26] = 5.56, /?, 40, 42, 69, 134, Routh, D. A., 277 Rowe, E. J., 20, 21 403, 452, 456, 470, 472 Rubens, A. B., 39, 47 Parisi, D., 365, 457 Rudel, R. G., 479 Pate, D. S., 433, 434, 437, 446n, 449, 460 Rugel, R. P., 47 Patterson, J. V., 95, 108 Rumelhart, D. E., 4, 44, 75, 84 Patterson, K. E., 1, 2, 12, 43, 74, 81, 82, 90, Rundus, D., 37 222 Ryan, J., 20 Penney, G G., 404 Ryder, L. A., 397 Perani, D., 9, 38>, 42 Perchonock, E., 188, 190, 428 Sachs, J. S., 55, 188, 391, 397 Pereira, F. C N., 189, 209, 394 Saffran, E. M., 2, 16n, 24, 29, 30, 37, 39, 68, Perfetti, G A., 332, 377, 400, 478, 481 83, 116, 140, 146, 157, 161, 168, 176, Peterson, L. R., 28, 48, 107, 109, 191, 319, 198, 199, 200f, 208, 209, 272, 339, 348, 337 357, 358, 359, 367, 369, 390, 403, 405, Peterson, M. J., 48, 107, 109, 191, 319 419, 423, 429, 431, 433, 434, 437, 441, Peynircioglu, Z. F., 35 442, 446n, 448-50, 460, 462, 464, 471, Philipchalk, R., 20 480, 482 Phillips, L. W., 132, 326 Salame, P., 19, 46, 59, 60, 61, 76, 87, 89, Picton, T. W., 94 337, 474 Pisoni, D., 74, 78> Salasoo, A., 138 Pizzamiglio, L, 365, 457 Salter, D., 145 Plaut, D. C, 11, 44, 69

Name index Salthouse, T. A., 249, 263 Samar, V., 271 Sartori, G., 45, 457 Satz, P., 478, 503n Savin, R, 188, 190, 428 Schaafstal, A. M, 231, 233, 235, 237, 238, 243 Schankweiler, D., 477-80, 483, 491, 500-3 Schiano, D. J., 228 Schneider, W., 85, 344 Scholes, R. J., 168, 403, 469, 478, 503n Schraagen, J. M. G, 231, 233, 235, 237, 238, 243 Schreuder, R., 190 Schulze, S. A., 237 Schwartz, M. F., 43, 176, 198, 199, 367, 403, 405, 422, 431, 433, 434, 437, 441, 442, 446n, 449, 460, 462, 463, 480 Schwent, V. L, 94 Scott, D., 46 Segarra, J. M, 41 Seidenberg, S., 343 Selfridge, J., 187, 399 Shallice, T., 2, 7, 12, 13, 16n, 18, 20-3, 25, 26f, 27, 29, 30, 35-6, 39n, 41, 43, 56, 57, 61, 74-6, 82, 83, 108, 116, 133, 161, 167, 168, 187, 188, 190-2, 205, 208, 210n, 211, 221, 222, 241-3, 263, 331, 349, 353, 358, 359, 397, 401, 403, 405, 448, 450 Sharp, D., 223 Sheldon, A., 503n Sheremata, W. A., 41 Shiffrin, R. M, 2, 3, 7, 11, 17, 56, 217, 222, 344 Shoben, E. J., 162 Shulman, H. G., 188 Siegel, A. W., 227 Simon, H. A., 11, 37, 44, 311 Simpson, M., 94 Slobin, D., 347 Smith, E. E., 190 Smith, S. T., 500-3 Snodgrass, J. G., 170 Somberg, T. A., 249 Speelman, R. G., 17, 22, 29, 32 Sperling, G., 8, 17, 19, 22, 28, 29, 32 Spicuzza, R. J., 107 Spilich, G. A., 248, 264, 265 Spinnler, R, 16n, 18, 35, 37, 38, 39n, 61, 241, 258, 263, 364, 403, 428, 448 Spoehr, K., 79

515

Spreen, O., 123 Springer, G., 145 Squire, L. R., 323 Stanovich, K. G., 230 Starr, A., 95, 101, 107, 108, 346 Steedman, M., 391, 482, 494 Stein, C. L, 478, 500, 503n Stenning, K., 208 Steinberg, S., 94, 107 Stevenson, R., 188 Stowe, L, 346 Strauss, S., 177, 184, 371 Strub, R. L, 23, 38, 39, 358 Summerfield, Q., 269, 270 Swanson, N. G., 36 Swinney, D. A., 343 Tager-Flusberg, H. B. T., 503n Tagliavini, C, 451, 455 Talland, G., 327 Tanenhaus, M. K., 343, 346 Tash, }., 405 Tavakilian, S. L, 500, 503n Tejirian, E., 391, 397, 399 Thibadeau, R., 391 Thomson, N., 28, 29, 30, 31, 46, 59, 226, 228, 289, 337, 403 Thornton, R., 481 Thurm, A. Tv 222 Tinzmann, M. B., 64 Tola, G., 479 Torgeson, J. K., 69, 479 Touretzky, D., 44 Treisman, A. M, 63, 163, 397, 399 Trollope, J., 297 Tuller, B., 480 Tulving, E., 22, 36, 319, 320, 326-7 Turvey, M. T., 337, 479 Tyler, L K., 392, 482 Tzeng, O. J. L, 57 Tzortzis, C, 16n, 20, 23, 358 Vallar, G., 2, 3, 9, 16n, 18, 21, 24, 27-30, 32-40, 42, 46, 60-2, 67, 69, 74, 76, 83, 116, 123, 125, 126, 134, 176, 208, 209, 210n, 230, 241, 326, 334, 337, 338, 340, 342, 345, 348, 352, 358, 364, 365, 367, 368, 370, 371, 375, 381, 384n, 390, 400, 403, 405, 412, 419, 421, 428, 429, 446n, 448-50, 452, 456-9, 464-7, 469-71, 483 Vanderwart, M., 170

516

Name index

Van Dijk, T. A., 390, 397 Vanier, R, 208, 338, 340, 374, 446, 449, 450, 470, 471 Vanner, E., 209 Varney, N. R., 123 Vellutino, F. R., 478 Vignolo, L. A., 40, 41, 119, 407, 420, 450 Vines, R., 19, 37 Von Eckhardt, B., 428 Wagner, R. K., 479 Wall, S., 499 Wanner, E., 209, 395, 409, 428 Warrington, E. K., 2, 7, 12, 13, 16n, 18, 20-3, 25, 26t 29, 30, 35, 36, 38, 39n, 41-3, 45n, 47, 56, 57, 61, 82, 108, 116, 133, 151, 168, 169, 171, 172, 174-6, 179, 183, 187, 189-91, 205, 208, 209, 218, 240, 241, 324, 327, 338, 345, 349, 353, 358, 370, 382, 390, 403, 405, 418, 428, 448, 449, 467, 472 Waters, G. S., 88, 290, 343, 358, 374, 385n Watkins, M. ]., 19, 35, 37, 46, 57, 81, 140, 151, 161, 207, 228, 401, 472 Watkins, O. C, 19, 35, 37, 46, 82, 140, 151, 161, 472 Watkins, S. H., 157 Watson, R. T., 403 Waugh, N. C, 2, 3, 7, 11, 13, 17, 21, 44, 190, 247, 260 Webber, M. L, 183 Weinberg, A., 344, 391, 392, 394, 418, 444

Welford, A. T., 247, 248, 262, 265 Welsh, A., 59 Whitaker, K, 480 White, W., 289, 290 Whitten, W. B., 22, 35 Wickelgren, W. A., 17, 18, 19, 27, 44, 75, 76, 337, 468 Wight, E., 59 Willette, M, 326 Williams, D., 248 Wilson, B., 16n, 21, 24, 32, 39, 62, 67, 307, 334, 348, 358, 367, 368, 384n, 404, 464, 483 Wilson, K. P., 64 Wingfield, A., 151, 187, 482 Wolf, M, 479 Wood, F., 320, 327 Woodin, M. E., 237 Wright, G, 401 Wright, R., 249, 262 Yeni-Komshian, G. R, 272, 339, 347, 353 Yuille, ]., 151, 164n Zampolli, A., 451, 455 Zangwill, O. L, 56 Zanobio, E., 61, 241, 358, 364, 403, 428, 448 Zelinski, E. M, 248 Zhang, G., 37, 311 Zurif, E. B., 168, 347, 367, 478, 500, 503n Zwicky, A. M, 348, 433

Subject index

acoustic coding, 55; see also auditory codes; phonological coding age effects: in Brown-Peterson function, 247; and complexity, effect of, 248ff; and complexity, grammatical, 255—6; and complexity in sentences, 250-2; and dividing attention, 255; and electrophysiological measures, 98-100; list length and recall, 255-60; in recency effect, 247; in working memory, 247-67 agrammatism, 176; and comprehension deficits, 406—7; and mapping hypothesis, 47.1; and structural vs. lexical contrast tasks, 433 alphabet, reciting, 62 amnesia (see also ecphory; episodic recollection; long-term memory; remote memory): STM and LTM in amnesics, 328; time estimation in, 218, 319-28 aphasia, 146 {see also conduction aphasia; transcortical motor aphasia; Wernicke's aphasia); acoustic-phonetic processing, 340; and articulatory loop, 63; fluent, and semantic anomalies, 434-6; and inner speech rates, 405; "isolation aphasia," 480; and lipreading, 268-9; and modularity of language, 480-1; nonfluent and articulatory difficulty tasks, 405; nonfluent and span tasks, 406-7; and sentence comprehension, 332-3; STM and comprehension deficits in 482 apraxia, ideomotor, 430 arithmetic, STM demands of, 472 articulation rate: and span, 400; and word length effect, 226-7 articulatory codes, 75-7 (see also acoustic coding; auditory codes; phonological 517

coding; semantic coding); input-output phonology conversion, 288 articulatory loop, 224-6 (see also rehearsal; repetition; working memory); and aphasia, 63; and articulatory code, 76; and articulatory suppression, 6 0 - 1 ; automatic and controlled processing in, 377; in children, 230-5, 236-7, 241; and dysarthria, 62—6, 405; and dyslexia (developmental), 68-70; and dyspraxia, 63; and feedback, 63; and language comprehension, 66-8; and language processing, 339-41; patients with deficits in, 372-7; and phonological representation strength, 337; and phonological similarity, 59; and rehearsal in children, 230-5; and sentence processing, role in, 404—11; and STM, 61-2; word length effect, 59 articulatory supression: and articulatory loop, 6 0 - 1 , 225-6; and homophone substitution, 297; and immediate recall, 6 0 - 1 , 90; and list memory, effect upon, 289; and patient MK, 312-15; patterns in normals, 312—15; and phonological similarity, 76; phonological to auditory conversion, 312; and reading, 65-6; rhyme vs. homophone judgments, 76, 290 attention, age effects in division of, 260 auditory codes, 75—7; see also acoustic coding; phonological coding; semantic coding auditory-verbal STM, 11-53 (see also phonological buffer; phonological processing; phonological storage; precategorical acoustic store; short-term

518

Subject index

auditory-verbal STM (cont.) memory); basic patterns of performance in patients, 13; cortical localization of, 331; defined, 167; definition of deficit, 338-9; electrophysiological measures, 100—8; function of auditory-verbal span, 182-4; functional architecture, 7-93; interaction with visual store, 109; localization of, 169-70; neural correlates, 7-53; 94-110; in patients, 101-2 (see also patients); and phonological analysis, 464; and phonological store, 7; relation with span tasks, 168; and speech comprehension, 208 (see also speech comprehension); and temporal orientation, 218-19 auditory word identification, 14-16 automatic and controlled processing, see models, Shiffrin and Schneider Benton Phoneme Discrimination Task, 123-4 Boston Diagnostic Aphasia Examination (BDAE), 118-19, 363, 430 Brown-Peterson task, 14-16; age differences in, 247; and amnesia, 56; and concreteness effects, 162; and word lists, 189 center-embedded relative structures, 395-6, 409 central executive, 58, 216, 225 (see also articulatory loop; working memory); age effects upon, 264-5; and supervisory attentional system, 263 children, 215; cognitive abilities in, 223; extrapolation from adult data, 223-4; phonological memory in, 424; visual memory in, 235-9; and working memory, 221—45; working memory and sentence comprehension, 477-503 chunking, 489 coding, see acoustic coding; articulatory codes; auditory codes; phonological coding; semantic coding complexity effects, and syntactic comprehension, 370 comprehension (see also comprehension of sentences; speech comprehension; Token Test); and articulatory rehearsal, role of, 407; deficits and preserved phonological processing, 452-3; deficits and strategies, 445; of language and STM, 66-8, 168,

337-84; in reading, 398-9; response selection in, 398; in right hemisphere patients, 438-9; of sentences, see comprehension of sentences; single word and phonological processing, 357; of speech, see speech comprehension; speech perception and lexical access, 393; and STM deficits in patients, 390 (see also neuropsychological evidence); theories of, 391; and Token Test, see Token Test; vs. repetition in STM patient, 437-41; of words and STM, 2 - 3 comprehension of sentences, 197-8 (see also comprehension; speech comprehension; Token Test); in aphasics, 332; centerembedded relative structures, 395-6; and complexity, 405-6; components analyzed, 390-3; the "garden path effects," see garden path sentences; and gist recall, 206; meaningless sentences and STM patients, 198-9; memory requirements of, 419-24; methodological difficulties in proving deficits, 347-8; and naming, 175; and parsing, 346-8; and phonological decoding impairments, 357; and phonological memory, 331-3; and phonological processing, 448—73; and phonological store, 380; and phonological store, hypothesized role of, 469ff; in poor readers, 477-80; and relative clauses, 486-7; and repetition, 113; requiring rehearsal, 417; role of STM, 183-4; sentence-picture matching in ER, 457-58; subject-object relation factors, 177-8; and syntactic parsing, 393-7 (see also parsing); word order effects, 199-200; word order processing, independent of, 179-180; and working memory in children, 477-503 computerized tomography (CT) scan, 451 conduction aphasia: lipread vs. auditory span, 283; pattern of impairments in, 451-2; and phoneme discrimination, 116; and phonological coding, 80; span tasks in, 453; vowel and consonant discrimination, 80 connectionism, 4, 11-12; and interactive models, 84-7 Corsi Block Tapping Test, 422 Decoding Skills Test, 491 Denver Auditory Phoneme Sequencing

Subject index Task, 124 digit span, see span tasks discourse abnormalities, detecting, 366-7 dissociations: associations vs. dissociations, 331; interpretation difficulties, 242; list and sentence retention, 181-2; span and sentence processing, 180—1; span vs. sentence repetition, 112 distraction, effect upon recency, 55 dysarthria, and articulatory loop, 6 2 - 3 , 405 dyslexia: and articulatory loop, 64-6, 68—70; in children and span performance, 64; developmental, 64, 68-70; effects of phonological similarity, 64; phonological, 68; span performance in developmental dyslexics, 332 echoic memory, 117; see also phonological buffer; phonological storage; precategorical acoustic store; recall, immediate; short-term memory ecphory, 327; see also episodic recollection; long-term memory electrophysiological measures, see eventrelated potentials episodic recollection, 319, see also amnesia; ecphory; long-term memory; remote memory event-related potentials, 9, 94-110; 346; age effects on P450, 98-100; localization of auditory-verbal STM, 108; memory vs. acoustic store distinction, 106-7; modality effects, 99; and P300, 105; peak amplitude and latency, 97; reaction time and P450, 9 7 - 9 evoked potentials, see event-related potentials Famous Faces Test, 119 fractionation methodology, 215; see also methodology frequency effects, 151-54; in serial recall, 161 fusion and blend illusions, 270-2 garden path sentences, 200-2, 210, 414-15, 489-91 (see also comprehension of sentences); good vs. poor readers' performance, 497-502; normals' performance on, 505; and working memory, 489-90 grammatical complexity, 250-2 (see also

519 syntactic structure); age effects, 255-6, 259 grammaticality (see also grammatical complexity); effect upon sentence recall, 203-4; judgments of, 197-8; and narrow window hypothesis, 441—3, 446 Hebb paradigm, 159-60 homographs, 392 homophone judgments, 289-90, 307-8 imageability, word, 134, 151-4; and comprehension, 303-4; explanation of in patients, 162; and lexical influence, 158; in normals and patients, 162 immediate recall, see recall, immediate interactive activation, 161; and immediate memory, 84-91; models of, 190-1; segmental representations for speech, 117; of speech perception, 84; of visual word perception, 84 interference tasks: and event-related potentials, 109; in reading for meaning, 88; visual interference and recall, 237; visual vs. verbal in children, 238-9 interference theory, 54-5 internal speech, see speech, internal intrusion errors, 153-4, 180; and nonsegmental features, 129; in serial recall, features of, 116 Korsakoff's psychosis: normal digit span in, 56; time estimation in, 321-5 language processing: a compiling analogy, 484-6; a framework for understanding, 342-4; model of, 481-3 levels of processing model, see models, Craik and Lockhart lexical decision tasks, 136-9: abstract vs. concrete words, 130; reaction times, 139 lexical discrimination, and rehearsal loop, 284 lexical organization, and word repetition, 293 lipreading, 217, 268-84; and aphasia, 272; cortical localization of, 272-3; effect of rehearsal on recall, 278; fusion and blend illusions, 270; normals performance on, 133; and phonological processing deficits, 273-5; and precategorical acoustic store, 278-80; and prosopagnosia, 271; and

520

Subject index

lipreading (cont.) pure word deafness, 272-3; and recency effects, 281; and recency in spoken lists, 276-7; right hemisphere contribution, 271-3; sensory effects, 277; sites of functional impairment, 273-5; and speech perception, 262-71; suffix effect, 267-7; visual evoked response studies, 271 list length, age effects, 255-60 list matching, 299-300 list recall, see recall long-term memory (LTM): and lexical influences in STM, 160; relation with STM, 167-8 mapping hypothesis (of phonological processing), 471 methodology (see also dissociations); associations vs. dissociations, 331; fractionation, 215; normals vs. patients, 163-4; and parsing deficits, 346-8 Mill Hill Vocabulary Test, 250, 253 modal model, see models, Atkinson and Shiffrin models {see also connectionism, interactive activation); Atkinson and Shiffrin, 2, 7, 56-7; Baddeley and Hitch (see working memory); Berwick and Weinberg parser, 392-3, 396, 418; Craik and Lockhart, 8, 57-8, 465-6; Hebb, 54; interactive models, see interactive activation; levels of processing, see models, Craik and Lockhart; modal model, see models, Atkinson and Shiffrin; multiple store, 4; Norman and Shallice, 263; Shiffrin and Schneider, 85, 344-6, 377; single multicomponent language, 168; READER, 391-2; TRACE, 84, 217, 270, 276, 280; Waugh and Norman, 2, 7; working memory, see working memory modularity: development of the system, 222-3; hypothesis of, 222; and language organization, 480-1; and working memory, 226-7 multiple store models (of STM), see models, multiple store

evidence, 244-5; developmental fractionation, 221-4, 243-4; and fractionation, 240-3; and lipreading, 283-4; LTM vs. STM, 320-1; for multiple phonological representations, 75—82; phonological processing and sentence comprehension, 448-73; sentence comprehension, 332; STM and language comprehension, 337-84; 403-4; for STM and sentence comprehension links, 448-50; for STM in sentence processing, 390-424; summary of neuropsychological STM findings, 382-4; supraspan and phonological store, 456; working memory, 224-7 normals, studies of: articulatory suppression, 312-15; concreteness effects, 162; and garden path sentences, 505; and lipreading, 133; and neuropsychology, 54-70; and nonword repetition, 157; performance on homophone substitution, 297; probe recognition of nonverbal sounds, 127-8; speaking aloud and memory, 311; studies using span tasks, 167; suppression, pseudohomophones and nonwords, 306; supraspan and phonological store, 456; supraspan serial recall performance, 154 optic neuritis, evoked potentials in, 104

P450, see event-related potentials parallel distributed processing (see also connectionism; models) memory, 69-70; of phonemic features, 270 paraphasias: phonemic, 192; phonemic in word repetition, 340; verbal, 192 parsing, 3, 209-10; Berwick and Weinberg model, 392-3, 396; and compiling analogy, 485; and comprehension, 429; defined, 481; immediate memory component, 331; and lexical-semantic errors, 460-1; and phonological STM, 333, 401-2, 418; pre-parsing buffer, role of, 396, 402; relation with phonological representation, 371; and semantic narrow window hypothesis, 429, 441-3, interpretation, 397-8; and sentence 446 comprehension, 346-8, 443; syntactic, and comprehension, 393-7; syntactic, neologisms, 182 defined, 489; syntactic, and phonological neuropsychological evidence, 90—1 (see also store, 466-7 patients); amnesia and span performance, 56; contrast with developmental patients: AB, 416-17, 420; AK, 406; AL,

Subject index 349ff; articulatory loop deficits, patients with, 372-7; BO, 342, 372ff, 418; central phonological deficits, patients with, 349-62, 377-82; children, see children; CN, 147—63; complex sentence recall in STM deficits, 204; contrast with normals, 215; contrasted with normals on supraspan, 161-2; DB, 349ff; DRB, 273-4, 283-4; EA, 349ff, 412-17; EDE, 118-42, 377ff, 464; EE, 464; and electrophysiological measures, 101-7; ER, 358^, 450-73; GI, 358ff, 383, 412, 422; HM, 56; IL, 358ff, 464; JB, 191-5, 197, 202-10, 349ff, 464, 470; JO, 328; JS, 349ff; JT, 328; KC, 349ff;. KF, 358^; MC, 358^, 429, 449, 464; MK, 217-18, 273, 283-4, 291-313, 342, 372ff; Ms D, 271; Ms T, 272-3; NHA, 169-83, 358^, 418, 449; NHB, 169-83; performance on probe recognition tasks, 209; phonological memory deficits in, 334, 358-7A; PV, 116, 134, 241-2, 358ff, 449, 464, 470; RAN, 169-74, 358ff, 449; RE, 190, 209, 377^, 420-3; RL, 372

Neuropsychological Impairments of Short-Term Memory

Neuropsychological Neurology: The Neurocognitive Impairments of Neurological Disorders

Memory: Neuropsychological, Imaging and Psychopharmacological Perspectives

Essentials of Neuropsychological Assessment

Neuropsychological Assessment of Neuropsychiatric Disorders

Neuropsychological Evaluation of the Child

Pediatric Neuropsychological Intervention

Handbook of Normative Data for Neuropsychological Assessment

Neuropsychological Management of Mild Traumatic Brain Injury

Attention: a neuropsychological approach

Neuropsychological Evaluation of the Child (Baron, Neuropsychological Evaluation of the Child)

Behavioural Approaches to Neuropsychological Rehabilitation: Optimising Rehabilitation Procedures (Neuropsychological Rehabilitation)

Neuropsychological Research: A Review

Clock Drawing: A Neuropsychological Analysis

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Memory

MEMORY

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Neuropsychological Impairments of Short-Term Memory

Neuropsychological Neurology: The Neurocognitive Impairments of Neurological Disorders

Memory: Neuropsychological, Imaging and Psychopharmacological Perspectives

Essentials of Neuropsychological Assessment

Neuropsychological Assessment of Neuropsychiatric Disorders

Neuropsychological Evaluation of the Child

Pediatric Neuropsychological Intervention

Handbook of Normative Data for Neuropsychological Assessment

Neuropsychological Management of Mild Traumatic Brain Injury

Attention: a neuropsychological approach

Neuropsychological Evaluation of the Child (Baron, Neuropsychological Evaluation of the Child)

Behavioural Approaches to Neuropsychological Rehabilitation: Optimising Rehabilitation Procedures (Neuropsychological Rehabilitation)

Neuropsychological Research: A Review

Clock Drawing: A Neuropsychological Analysis

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Memory

MEMORY

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Memory

Recommend Documents