Neuropsychological Impairments of Short-Term Memory

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

Department of Psychology

University of Milan

The University of Utah

Brian Butterworth

Mary L. Gick

Department of Psychology

Department of Psychology

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

General introduction

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

General introduction

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

General introduction

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.

General introduction

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

<2(?)*

—

—

—

CN12

Boston

> 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

Impairment of auditory-verbal short-term storage

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

Impairment of auditory—verbal short-term storage

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

Impairment of auditory—verbal short-term storage

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

Impairment of auditory—verbal short-term storage

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

Impairment of auditory-verbal short-term storage

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,

Impairment of auditory-verbal short-term storage

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

Impairment of auditory-verbal short-term storage

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.

Impairment of auditory—verbal short-term storage

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

Impairment of auditory-verbal short-term storage

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.

Impairment of auditory-verbal short-term storage

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

Impairment of auditory—verbal short-term storage

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.

Impairment of auditory—verbal short-term storage

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).

Impairment of auditory-verbal short-term storage

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.

Impairment of auditory-verbal short-term storage

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.

Impairment of auditory—verbal short-term storage

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.

Impairment of auditory-verbal short-term storage

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

Development of the concept of working memory

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

Development of the concept of working memory

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

Development of the concept of working memory

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.

Development of the concept of working memory

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

Development of the concept of working memory

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

Development of the concept of working memory

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

Development of the concept of working memory

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.

Development of the concept of working memory

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.

Development of the concept of working memory

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.

Multiple phonological representations

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

Multiple phonological representations

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.

Multiple phonological representations

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

Multiple phonological representations

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

Multiple phonological representations

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.

Multiple phonological representations

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

Multiple phonological representations

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.

Multiple phonological representations

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.

Multiple phonological representations

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.

Electrophysiological measures of short-term memory

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

Electrophysiological measures of short-term memory

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

Starr, Barrett, Pratt, Michalewski, and Patterson

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

Electrophysiological measures of short-term memory

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

Starr, Barrett, Pratt, Michalewski, and Patterson

(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

Electrophysiological measures of short-term memory

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

Starr, Barrett, Pratt, Michalewski, and Patterson

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

Information sources in immediate recall

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,

Information sources in immediate recall

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.

Information sources in immediate recall

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

Information sources in immediate recall

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.

Information sources in immediate recall

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

Information sources in immediate recall

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

Information sources in immediate recall

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)

Note: N = 10 in each serial position.

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.

Information sources in immediate recall

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

Note: N = 10 in each serial position.

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

Information sources in immediate recall

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.

Information sources in immediate recall

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

Information sources in immediate recall

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

Information sources in immediate recall

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.

Information sources in immediate recall

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

Saffran and N. Martin

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

Position in input string

O.OC 1

2

3

4

0

Position in input string

••- 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

Position in input string

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

Position in input string

Position in input string

0.6 05

Q4

0 4

0.3

03

0.2

0.2

01

0.1

0.0

00 1

2

3

Position in input string

• - 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

Position in input string

4

Lexical involvement in short-term memory

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

Position in input string

"•" 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),

Lexical involvement in short-term memory

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

Saffran and N. Martin

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

Lexical involvement in short-term memory

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

Saffran and N. Martin

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

Lexical involvement in short-term memory

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

Saffran and N. Martin

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.

Lexical involvement in short-term memory

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

Saffran and N. Martin

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-

Lexical involvement in short-term memory

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

Saffran and N. Martin

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

Lexical involvement in short-term memory

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

Saffran and N. Martin

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.

Lexical involvement in short-term memory

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

A phenomenon in search of a function?

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

A phenomenon in search of a function?

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

McCarthy and Warrington

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].)

A phenomenon in search of a function?

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

A phenomenon in search of a function?

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:

A phenomenon in search of a function?

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

McCarthy and Warrington

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

A phenomenon in search of a function?

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

McCarthy and Warrington

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

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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,

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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

Butterworth, Shallice, and Watson

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

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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

Short-term retention without short-term memory

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.

Short-term retention without short-term memory

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.

Short-term retention without short-term memory

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.

Short-term retention without short-term memory

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

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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:

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

(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,

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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

Short-term retention without short-term memory

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

Butterworth, Shallice, and Watson

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.

Short-term retention without short-term memory

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

Developmental fractionation of working memory

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

Developmental fradionation

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 &

Developmental fractionation of working memory

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

Developmental fractionation of working memory

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.

Developmental fradionation

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

Developmental fractionation of working memory

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

Developmental fradionation of working memory

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

Developmental fradionation of working memory

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

Developmental fradionation of working memory

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.

Adult age differences in working memory

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

Craik, Morris, and Gick

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

Craik, Morris, and Gick

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

Adult age differences in working memory

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

Adult age differences in working memory

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

Craik, Morris, and Gick

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).

Adult age differences in working memory

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

Craik, Morris, and Gick

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

<

or CO Q

a: o D—

UJ 00

i

4

6

8

LIST LENGTH Figure 10.4. Mean numbers of words recalled as a function of age and experimental condition (Experiment 4).

observations were confirmed by an analysis of variance on the conditions involving verification. The analysis yielded significant effects of age, list length, and sentence complexity. There were also significant interactions between age and list length, and between age and sentence complexity; no other interactions were significant. Further analyses of variance within each age group confirmed the impression given by Figure 10.4 that there were significant effects of list length and sentence complexity for the young group but not for the old group. Finally, an overall analysis of variance on all conditions yielded a marginal Age x Sentence-no sentence interaction (F[l, 30] = 3.50, p < .10), but inspection of Figure 10.4 makes it seem likely that this small effect is attributable to list length 4 in which there appears to be a ceiling effect for the young group. For list lengths 6 and 8 there is again no indication that division of attention (sentence absent vs. sentence present) has a greater disruptive effect on the recall performance of the older group. These results confirm and extend the results of Experiment 3. When no sentence verification was required, older subjects' recall scores increase with list length, but at a slower rate than that of the younger group. One interpretation of this finding is that the younger people were more successful at augmenting their primary memory recall by

260

Craik, Morris, and Gick

retrieving further words from secondary memory (Waugh & Norman, 1965; Craik, 1968). In the working memory conditions (with concurrent sentence verification) the old subjects recall just over two words in all conditions. The absence of effects of either list length or sentence complexity adds further weight to the suggestion that their recall is drawn exclusively from a relatively automatic articulatory type of rehearsal. Younger subjects' recall scores do increase with list length and are affected by sentence complexity, suggesting that their articulatory recall is augmented by retrieval from the central processor and from secondary memory. Finally, Experiment 4 again showed no evidence for an interaction between age and division of attention in a working memory situation.

10.6. Discussion The four experiments yielded some expected and some unexpected effects. First, all four studies revealed marked age-related decrements in working memory performance, both in the recall measure (Experiments 2, 3, and 4) and in the sentence verification data (Experiments 1, 2, 3, and 4). There seems little doubt then that older people - even highly intelligent and well-educated people such as those used in the present studies — perform less well than their younger counterparts in working memory situations. This conclusion stands in contrast to the slight or nonexistent age differences that are found in primary memory tasks (Craik, 1977). It seems that the necessity to hold some items while deciding about or manipulating others is especially difficult for older people. The main prediction for the present series was that age decrements would be amplified by any increase in the difficulty or complexity of the working memory task. However, this prediction was only partially confirmed. An increase in the grammatical complexity of the sentences to be verified was associated with increased verification latencies and an increase in verification errors for both age groups in all four experiments. Additionally, increased sentence complexity was associated with decreased recall in all experiments except Experiment 1, in which subjects rehearsed the memory set aloud and thus made very few errors. This pattern of results confirms the general view of working memory theorists that an increase in the processing complexity of a concurrent task results in a decreased ability to hold other items in short-term storage. When the interaction between age and sentence complexity is considered, it was found that older people were differentially handicapped by the increase in complexity in the sentence verification task itself - shown either as an increase in errors or as an increase in latency. This was true for Experiments 1,2, and 3; the interaction was absent in Experiment 4, possibly because the older people simply maintained two items by articulatory rehearsal and concentrated primarily on the verification task. For the recall

Adult age differences in working memory

261

measures, the Age x Complexity interaction was reliable for Experiment 2, which used the Daneman and Carpenter paradigm, but not for Experiments 1 and 3, using the Baddeley and Hitch "preload" paradigm. One major difference between the paradigms is that in the Daneman and Carpenter situation, memory items are added during processing of the concurrent task, whereas in the Baddeley and Hitch paradigm, the memory items are presented before the second task is given. If older people are principally affected by the processing aspects of working memory tasks, rather than by the storage aspects per se, it would be expected that they would show decrements in the processing task itself, but would show recall decrements only when the ongoing processing and memory-holding aspects of the situation could not be kept apart - as in the Daneman and Carpenter paradigm. In Experiment 4, increased sentence complexity was associated with a differential drop in recall performance for the young subjects; it was suggested that the young people augmented their articulatory recall by some storage in the central processing component of the working memory system — thereby making their recall performance vulnerable to the increased processing demands of the concurrent verification task. The manipulation of memory load (or list length) again had somewhat different effects depending on the specific situation. The effect of list length on recall itself followed the standard pattern: As the number of memory items increased, the absolute number of words recalled increased, but the proportion recalled decreased. The theoretical interest in this variable therefore lies in its effects on the processing task, and in the associated interactions with the age variable. With respect to the recall measure, younger subjects recalled differentially more words as list length increased in the Baddeley and Hitch paradigm (Experiments 3 and 4); this result accords with the finding of an interaction between age and list length in free recall reported by Craik (1968). Experiment 1 also used the Baddeley and Hitch paradigm, but in this case only two and four items were presented and words were rehearsed aloud; no age differences were found in this case. In the Daneman and Carpenter task (Experimertt 2), age and memory load did not interact in the recall measure, although recall did show an interaction between age and complexity. This was, therefore, one case in which one source of difficulty (sentence complexity) interacted with age while another (memory load) did not. Performance on the sentence verification task showed the same dissociation between the effects of memory load and sentence complexity. Age interacted with complexity (either for errors or for latencies) in Experiments 1, 2, and 3, whereas age and memory load did not interact reliably for either measure in any of the four experiments. It cannot be concluded that the experiments were simply insensitive to the effects of memory load, since increasing load was associated with reliable increases in both verification latencies and verification errors in Experiments 1 and 2. The main effect of load was not significant in Experiments 3 and 4, possibly because subjects in those studies relied

262

Craik, Morris, and Gick

largely on articulatory recall and devoted more primary attention to the verification task. In any event, the initial expectation that older people would be differentially penalized by any increase in task difficulty was clearly not upheld by the present results. At one level of analysis, the finding that older people were differentially affected by one form of difficulty (sentence complexity) but not by another (memory load) means that not all forms of task complexity can be considered equivalent. To understand the effects it is necessary to understand which underlying processes are involved, as well as how aging and various experimental manipulations affect these processes. In this connection, the working memory analysis (although it is not the only possible way to dissect shortterm memory phenomena) is demonstrably useful. By suggesting that older people rely largely on articulatory recall in Experiments 3 and 4, the low levels of recall, the absence of a sentence complexity effect on recall, and the absence of an effect of memory load on sentence verification can all be understood. When subjects are forced to rehearse aloud (Experiment 1) or must add items to short-term storage during processing (Experiment 2), the pattern of results changes, but in understandable ways. Similarly, it was first expected that if processing resources decline with age as suggested by Craik and Byrd (1982) among others, then all aspects of working memory performance should decline in the older group. The differential patterns discussed earlier disconfirm this expectation. At one level, then, the usefulness of the processing resource notion is seriously questioned. However, the puzzle may again be resolved by proceeding to a finer-grained analysis in terms of component processes. Ongoing decision processes such as sentence verification do require "resources" and are vulnerable to the effects of aging; moreover, these age-related effects are exacerbated by such variables as sentence complexity. However, other aspects of working memory performance - notably articulatory rehearsal - may be less affected by aging, and also (because of their relatively automatic mode of operation) be less affected by other variables. Therefore, rather than attempting to understand the effects of aging on performance in terms of global ideas such as processing resources, it again seems necessary to dissect performance into its underlying component processes. A final puzzle associated with the present results is the complete absence - in all four experiments - of an interaction between age and division of attention. On the one hand this finding confirms the absence of an interaction between age and divided attention in working memory paradigms reported by Baddeley, Logie, Bressi, Delia Sala, and Spinnler (1986), by Light and Anderson (1985), and by Wright (1981), but on the other hand this consistent group of findings is apparently opposed to a second group of welldocumented results showing the presence of an Age x Divided-attention interaction in various perceptual-motor tasks (e.g., Welford, 1958; Salthouse et al, 1984; McDowd & Craik, 1988).

Adult age differences in working memory

263

The difference between the two groups may lie in the fact that the first group's tasks involve a memory component - perhaps the processes of subvocal rehearsal are sufficiently automatic and insulated from other processes to keep interference to a minimum. The tasks used in the second group of studies are typically more "on-line" in nature; they may present a greater problem of managing the allocation of attention between the two subtasks, and this difficulty may be exacerbated in the elderly. Alternatively, tasks in the first group appear to require "microdivision" of attention between rather similar components - verbal information from the same source, for example. This situation contrasts with the "macrodivision" of attention often required by tasks in the second group, where the tasks may involve quite different types of information or be presented in different modalities. Again the older person may experience more difficulty in managing and allocating his or her attentional resources between very disparate tasks. Clearly these speculative suggestions will require more systematic research before the puzzle can be resolved. How do the present results fit with current models of working memory? So far, aspects of the results have been ascribed to components of working memory described in rather general terms. This description can be equated, more or less, with what Baddeley (1986) refers to as the "general concept of working memory" - that is, the notion of a single limited-capacity short-term storage and processing system that operates across a wide range of cognitive tasks. However, Baddeley has also suggested more specific versions of the working memory model. In 1976 he discussed the idea that the short-term retention of linguistic material was mediated by a central processor of limited capacity, in conjunction with an articulatory rehearsal loop. The central processor had some storage capability, as well as the ability to perform reduction coding (e.g., to chunk individual letters into a single word), and to carry out other operations necessary for reasoning, comprehension, and learning; the articulatory rehearsal loop had a capacity of about three items (Baddeley, 1976, pp. 169-187). In later developments of the model, Baddeley (1986) has suggested that the central processor (by now promoted to the "central executive") may be synonymous with the "supervisory attentional system" of Norman and Shallice (1980). The central executive functions are essentially to exert supervisory control on a number of relatively automatic slave systems, such as the articulatory loop, the visuospatial scratch pad, and the phonological store (see also Baddeley, this volume, chapter 2). The present experiments were not designed to evaluate different versions of the working memory concept, but their results fit comfortably into Baddeley's (1976) model. The central executive (or central processor) performs the on-line decisionmaking task of sentence verification, and word storage (like memory span) is mediated by the central executive and the articulatory loop working interactively (Baddeley, 1976, p. 176). In line with previous work showing that aging has little effect on

264

Craik, Morris, and Gick

automatic mental processes (Hasher & Zacks, 1979), it seems reasonable to suggest that the articulatory loop is unimpaired in older people, but that both the capacity and flexibility of the central executive are impaired to some degree. The suggestion that articulatory loop operation is not impaired by aging gains further support from Morris and Baddeley's (1988) review of working memory functioning in patients with Alzheimer-type dementia; the authors concluded that the articulatory loop system functions normally in such patients. The present suggestion is also in line with the proposal by McLeod and Posner (1984) that auditory—vocal transformations (as in shadowing) utilize a "privileged loop" that is separate from the general information-processing network and allows repetition of spoken words to be performed independently of other cognitive activities. For example, shadowing is virtually "interference free" in dual-task situations. The privileged loop used in shadowing may also be used in articulatory rehearsal. The finding from Experiment 1 that there were no age differences in memory for preloaded words when they were rehearsed continuously and aloud also supports the idea of no age differences in articulatory loop function. In Experiments 2, 3 and 4 the memory loads were slightly greater; also, subjects were not asked to rehearse aloud - both of these changes would lead to greater involvement of the central executive, and thus to greater age-related losses in retention. In Experiment 2 in particular (using the Daneman and Carpenter paradigm), the necessity to add words "on-line" to the memory load would necessarily involve the central executive — substantial age-related decrements in reading span were found. The notion that younger subjects are more effective than older subjects at augmenting their articulatory loop recall with recall from the central processor is supported in Experiment 3, both by greater overall recall by the young group in the working memory conditions and by the tendency for sentence complexity to have a greater effect in the young group. This latter result was found again in Experiment 4; it was also suggested that the longer word lists and free recall instructions probably gave younger subjects an additional advantage in that further words could be retrieved from secondary memory (see also Craik, 1968). If sentence verification was carried out by the central executive, age differences would be expected in this component of the overall working memory task. Such results were in fact found, with the older subjects making significantly more verification errors in Experiments 1 and 3, and having significantly longer decision times in all four experiments. The suggestion from Experiments 3 and 4, that older subjects rely very largely on articulatory rehearsal when they have to make a concurrent decision, fits well with the results and conclusions of Spilich (1983). He commented that "the possibility of agerelated changes in working memory capacity suggests that either a decrease in the capacity of working memory or a qualitative change in the ability to guide its use would tend to shift the burden of processing upon the articulatory rehearsal loop for the aged"

Adult age differences in working memory

265

(Spilich, 1983, p. 241). Spilich also points out that if this is so, articulatory suppression should be more disruptive to the aged than to the young. Two final comments on the theoretical underpinnings of the present results are, first, that, by the present view, "primary memory" tasks involve both the articulatory loop and the central processor, but that the processor is not called on to manipulate, reorganize, or recode the presented material; also, of course, primary memory tasks typically involve only one set of operations on one set of material (compared to the present experiments in which subjects must both retain some words while making decisions about further materials). Given that age differences in primary memory tasks are slight (Craik, 1977), it seems that it is not the simple involvement of the central processor that leads to poorer performance in the older person, but rather it is the necessity to use the central processor in a flexible manner to manipulate and reorganize the items held that gives rise to problems. The second theoretical comment is that whereas the present results fit well with Baddeley's (1976) model of working memory, there is nothing in the findings that speaks against the later, more specific, model — the present experiments were simply not designed to assess that model. Baddeley's suggestion that "a decline in the capacity of the central executive is a particularly important feature of ageing" (Baddeley, 1986, p. 244) is certainly supported by the present results, although it is perhaps the flexibility and computational abilities of the central processor that are most vulnerable to the effects of aging, rather than simple "capacity." A decrease with age in the effectiveness of the central executive's higherlevel supervisory attentional control is also an attractive topic for future investigation. In summary, the present studies have confirmed the presence of large age decrements in working memory tasks. Given the crucial role of working memory in learning, comprehension, and problem solving, such decrements may well underlie age-related declines in these other cognitive functions (Welford, 1958). The present experiments have also shown that age interacts with some but not other forms of increased task difficulty in working memory situations. These findings, in turn, prompted an analysis in terms of possible processing components underlying performance. There is at least some evidence for the assertion that aging affects the ongoing active decision-making aspects of processing more than it affects the relatively passive or automatic storage aspects. In any event, it seems clear that further studies of the effects of normal and pathological aging on short-term retention can profit substantially from a dissection of the underlying abilities into their processing components. References Baddeley, A. D. (1976). The psychology of memory. New York: Basic Books. Baddeley, A. D. (1986) Working memory. London: Oxford University Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), 77M? psychology of learning and motivation (Vol. 8, pp. 47-90). New York: Academic Press.

266

Craik, Morris, and Gick

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. Cerella,J., Poon, L., & Williams, D. (1980). Age and the complexity hypothesis. In L. Poon et al. (Eds.), Aging in the 1980's. Washington, DC: American Psychological Association. Chase, W. G., & Clark, H. H. (1972). Mental operations in the comparison of sentences and pictures. In L. W. Gregg (Ed.), Cognition in learning and memory (pp. 205-232). New York: Wiley. Cohen, G. (1981). Inferential reasoning in old age. Cognition, 9, 59-72. Craik, F. I. M. (1968). Two components in free recall. Journal of Verbal Learning and Verbal Behavior, 7, 996-1004. Craik, F. I. M (1977). Age differences in human memory. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychology of aging (pp. 384-420). New York: Van Nostrand Reinhold. Craik, F. I. M. (1986). A functional account of age differences in memory. In F. Klix & H. Hagendorf (Eds.), Human memory and cognitive capabilities. Amsterdam: North-Holland. Craik, F. I. M., & Byrd, M. (1982). Aging and cognitive deficits: The role of attentional resources. In F. I. M. Craik & S. E. Trehub (Eds.), Aging and cognitive processes (pp. 191-211). New York: Plenum Press. Craik, F. I. M , & Rabinowitz, J. C (1984). Age differences in the acquisition and use of verbal information. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 471-499). Hillsdale, NJ: Erlbaum. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450—466. Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language. New York: McGraw-Hill. Gick, M. L, Craik, F. I. M., & Morris, R. G. (1988). Task complexity and age differences in working memory, Memory and Cognition, 16, 353-361. Hasher, L, & Zacks, R. T. (1979). Automatic and effortful processes in memory. Journal of Experimental Psychology: General, 108, 356—388. Hitch, G. J., & Baddeley, A. D. (1976). Verbal reasoning and working memory. Quarterly Journal of Experimental Psychology, 28, 603-621. Kucera, H., & Francis, W. N. (1967). Computational analysis of present day American English. Providence, RI: Brown University Press. Light, L. L, & Anderson, P. A. (1985). Working-memory capacity, age, and memory for discourse. Journal of Gerontology, 40, 737-747. Light, L. L., Zelinski, E. M., & Moore, M. (1982). Adult age differences in reasoning from new information. Journal of Experimental Psychology: Learning, Memory, and Cognition, 8, 435-447. McDowd, J. M., & Craik, F. I. M. (1988). The effects of aging and task difficulty on divided attention performance. Journal of Experimental Psychology: Human Perception and Performance. McLeod, P., & Posner, M. I. (1984). Privileged loops from percept to act. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 55-65). Hillsdale, NJ: Erlbaum. Morris, R. G., & Baddeley, A. D. (1988). Primary and working memory functioning in Alzheimertype dementia. Journal of Clinical and Experimental Neuropsychology, 10, 279—296. Morris, R. G., Craik, F. I. M., & Gick, M. L. (in press). Age differences in working memory tasks: The role of secondary memory. Quarterly Journal of Experimental Psychology. Morris, R. G., Gick, M. L., & Craik, F. I. M. (1988). Processing resources and age differences in working memory. Memory and Cognition, 16, 362-366. Norman, D. A., & Shallice, T. (1980). Attention to action. Willed and automatic control of behaviour. University of California at San Diego CHIP Report 99. Salthouse, T. A. (1982). Adult cognition: An experimental psychology of human aging. New York: Springer-Verlag. Salthouse, T. A., Rogan, J. D., & Prill, K. (1984). Division of attention: Age differences on a visually presented memory task. Memory and Cognition, 12, 613-620.

Adult age differences in working memory

267

Somberg, T. A., & Salthouse, T. A. (1982). Divided attention in young and old adults. Journal of Experimental Psychology: Human Perception and Performance, 8, 651—663.

Spilich, G. A. (1983). Life-span components of text processing: Structural and procedural differences. Journal of Verbal Learning and Verbal Behavior, 22, 231—244.

Waugh, N. G, & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89-104. Welford, A. T. (1958). Ageing and human skill London: Oxford University Press. Welford, A. T. (1980). Memory and age: A perspective view. In L. W. Poon, J. L. Fozard, L. S. Cermak, D. Arenberg, & L. W. Thompson (Eds.), New directions in memory and aging. Hillsdale, NJ: Erlbaum. Wright, R. (1981). Aging, divided attention, and processing capacity. Journal of Gerontology, 36, 605-614. Zelinski, E. M , Light, L. L., & Gilewski, M. J. (1984). Adult age differences in memory for prose: The question of sensitivity to passage structure. Developmental Psychology, 20, 1181-1192.

11. Lipreading, neuropsychology, and immediate memory RUTH CAMPBELL

11.1. Introduction Speech therapists often observe that aphasic patients seem to be very reliant on seeing the speaker. Despite this, and despite the obvious relevance of this observation to remediation, there is little in the neuropsychological literature on aphasia that relates to the ability to lip-read. What sort of help should one expect lipreading to provide for the patient? In particular, is it feasible to ask whether impairments in auditory-verbal processing may dissociate depending on whether the patient can only hear the speaker or can see and hear him? I shall discuss, first, ways of accommodating lipreading to the perception of auditory speech and, second, some aspects of lipreading in immediate memory. These are overlapping areas of concern, rather than separate ones. There are two opposing standpoints from which to examine the relationship of lipreading to speech processes. One is that lipreading might provide a source of speech information that is somewhat detached from normal speech perception but that can be called on, like a spare electric generator, in an emergency. The other viewpoint, by contrast, takes lipreading to be integrated into auditory speech perception. The analogy here may be to the spare can of petrol that the motorist may carry; it is carried in order to keep the car running when other sources of fuel may be reduced, but it utilizes the same mechanism as fuel that is bought from a gas station. So is speech reading a specialpurpose mechanism, tangential to normal speech processing (the spare power generator), or is it rather just another source of "speech-fuel" that utilizes just the same mechanisms as speech that is heard? Most studies of lipreading assume the former to be more true; they point to the enormous individual variability in performance on "standard" lipreading tasks, in contrast to the near-universal ability to perform auditory speech repetition tasks (in persons with unimpaired hearing). And many studies have attempted to determine which cognitive skills best predict lipreading in order to foster David Howard, Donald Broadbent and Jeanette Garwood kindly read and criticized this draft. To them all, many thanks for refining my sketchy notions and to David, particular thanks for making sure that my misinterpretations of M. K. were corrected.

268

Lipreading, neuropsychology, and immediate memory

269

them in people who may need to lipread. Yet the most thorough reviews, such as those by Jeffers and Barley (1971), indicate no particularly reliable or valid indicators of lipreading ability; if lipreading is an arcane cognitive process, available to the chosen few, it is singularly hard to predict whose those few will be and what that "spare power generator" may comprise. I tend to the latter view: I believe lipreading to be, on the whole although not entirely, a process that relies on mechanisms that determine and control the perception and production of heard speech. Individual differences in lipreading ability then reflect individual strategies in making the most of impoverished speech input, rather than alternative skills for abstracting information from the seen face. In turn, however, this means that speech perception must be considered to be a broader process than is sometimes conceived. Not only must the speech processor accept visual (lipread) inputs, but such information, like auditory information relevant to speech, may need to work interactively across phonetic, phonemic, lexical, and possibly syntactic levels. And it must work in such a way that context can affect the processing of speech events concurrently, retroactively and proactively, too. At present, I find the simplest way to account for many of the effects of lipreading is to suppose that it adds a phonetic feature — that of seen mouth opening and closing — to the speech stream; this is sufficient to account for most of reported effects of lipreading. What are these effects?

11.2. Some effects of lipreading in speech perception (a) Lipreading disambiguates speech in noise or through impaired hearing. As many investigators have pointed out (see Summerfield, 1987, for review), place of articulation can often be seen on the mouth when noise or disease impairs the processing of the highly transient, often high-frequency acoustic features that discriminate between many consonantal sounds. For example, ba and ga have similar low-frequency components and differ mainly in the higher-frequency distributions of sound energy. Thus, with high-frequency hearing loss or noise, the ability to discriminate these speech sounds auditorily may become impaired, but they are visually distinctive (although not necessarily identifiable from vision; see section 11.3). Thus it can be seen that speech can sometimes complement those aspects of heard speech lost through noise or impaired hearing. (b) Where context constrains what is spoken, lipreading alone can sometimes support speech perception. Barbara Dodd and I have shown that digit lists that are seen, but not heard to be spoken, can easily be perceived correctly; we have used this as the basis for our studies comparing heard, written, and lipread immediate recall (e.g., Campbell & Dodd, 1984). (c) Yet, without context, no consonantal phonemes can be uniquely identified from sight alone (although some groups of consonants can be distinguished from others),

270

Campbell

whereas vowels can be very variable in their lipread characteristics. And, of course, one can understand even quite noise-distorted speech well without seeing the speaker (telephone, radio, etc.). Lipreading can be a sufficient source of speech perception when there is some acoustic or other back-up source of speech information, but it is not a necessary one. (d) However, even perfectly clear acoustic speech inputs can be aided by seeing the speaker. Reisberg, McLean, and Goldfield (1987) have shown that the comprehension of spoken text passages (Kant's Critique of Pure Reason was one excerpt to which the Canadian undergraduate subjects were exposed) can be improved by up to 15% when the speaker is seen as well as heard. Work in our own laboratory (Garwood, in preparation) confirms and extends this finding over a range of settings, texts, speakers, and comprehension tasks. Moreover, the control conditions Garwood used (a still head that does not speak and a blank screen) show that the gain is not due to "attentional capture" by a face, but is specific to the face actions seen. (e) Lipreading integrates with heard speech. The most convincing demonstration of this comes from the fusion-and-blend illusions. McGurk (McGurk & MacDonald, 1976) is credited with the first demonstration of an illusion based on discrepant phonetic seen and heard inputs. When seeing a speaker say ga, while hearing the speaker say ba, one has the impression that da has been spoken. This illusion is compelling and immediate and can be demonstrated in quite young children, although it appears not to reach full "power" until the age of 7 or 8 years (McGurk, personal communication; Massaro, 1987). Blend illusions, where the perceived speech sound is a blend of the seen and heard sounds, rather than a "fused" or new phoneme, can also occur. So if the speaker is seen saying ba while heard saying ga, what is reported is usually bga, or sometimes bda (see Massaro, 1987). Vowels "blend," too (Summerfield & McGrath, 1984). Together with developmental evidence (see Campbell, 1988a, for fuller treatment) these findings suggest strongly that lipreading plays a part in normal speech perception. But only a flexible speech-processing theory, in which incomplete information can be actively processed at all and at every possible stage, would allow these effects to emerge. In particular, the usefulness of lipreading despite its inadequacy as a source of phoneme identification from sight alone and the contingencies of the blend-and-fusion illusions can be more satisfactorily discribed by such an interactive model (such as McClelland & Elman's TRACE [1986]) than by other means. Thus it is a problem for theories that would have parallel processing of phonemic features from heard and seen speech to explain why it is that seen ga and heard ba are perceived as da, whereas heard ga and seen ba generate bda-like percepts (Summerfield, 1987); they seem to require modality-specific contigencies to be built into them. Yet an interactive activation approach, in which heard or seen phonetic features can spread activation throughout the speech-processing system prior to identification, can be used to model these

Lipreading, neuropsychology, and immediate memory

271

different outcomes very satisfactorily without any special assumptions. The details are to be found in Campbell (1988b). I mentioned, though, that while I tend to the view that lipreading is another "fuel can", there may be some evidence that it could be something of a "spare generator"; that is, there may be some mechanisms involved in lipreading that are not used in auditory speech processing. What are these?

11.3. Some right hemisphere component to the early stages of lipreading speech? Clearly the most important difference between lipreading and hearing is that one uses the eyes and one the ears. Special-purpose mechanisms might usefully be sought in connection with visual processing. Two slight pieces of evidence indicate that lipreading may use some processing mechanisms that are not specific to speech and that are related to visual processing. Samar and Sims (1983, 1984) have found a correlation between lipreading (using the standard Uttley test) and component of the visually evoked response. This is an early component of the electrophysiological response to a seen stimulus and is unlikely to relate to strategic cognitive brainwork. It would appear to be related to the attentional aspects of examining a face to decide what is being said. And I have shown (Campbell, 1985) that there is a right hemisphere superiority for lipreading in a unilateral visual reaction time study. In that experiment subjects heard a speech sound — either a CV syllable or a vowel — followed by a brief unilateral exposure of a face photograph. The task was to make a match: Did the heard speech sound match the configuration of lips seen on the tachistoscopic exposure of the fullface photograph? Reaction times and accuracy of response were measured. Both vowels (ah, ee, oo) and consonant-vowel syllables (shuh, muh, thuh, fuh) were more quickly and accurately matched when the face picture appeared in the left than the right visual field. Thus, in this highly artificial task, where a brief visual display had to be matched to a heard speech sound, a consistent right hemisphere advantage emerged. However, if such right hemisphere mechanisms were essential to effective lipreading, we might expect right hemisphere patients, particularly those with problems in analysing faces for other things, to be impaired at lipreading. One case indicates quite conclusively that this need not be so (Campbell, Landis, & Regard, 1986). Mrs. D is densely prosopagnosic and prosopaffective agnosic, but can read speech from faces perfectly well. Mrs. D has an occipito-temporo-parietal lesion, medially placed, in the right hemisphere. At present, the possibility is open that lipreading deficits may emerge in patients with other right hemisphere lesions; Samar and Sim's finding might suggest that problems in visual attention consequent to right hemisphere damage may also give rise to some problems in understanding speech from seen lip movements. And patients

272

Campbell

with posterior right hemisphere lesions may be informative in locating the route(s) by which visually processed, lipread speech might access left hemisphere "speech analysis." In this context, Mrs. T, a patient with left temporomedial occipital lesion (Campbell et al., 1986) is worth considering. Mrs. T was not aphasic in any way, but was impaired at some aspects of lipreading, and she was not susceptible to fusion or blend illusions. Mrs. T's only other cognitive symptom was pure alexia that had resolved, at the time her lipreading was tested, to letter-by-letter reading. A classical Dejerine-type disconnection explanation for Mrs. T's deficit(s) could be advanced; the site of her lesion blocked the transfer of visually based, language-related information, via the direct occipitotemporal routes, to the anatomical site(s) where the phonological representation of a word is realized. The medial siting of the lesion further impaired the direct access of contralateral inputs to these centres: Even if the posterior parts of the right hemisphere in Mrs. T processed information from lips, such information would have difficulty in reaching the left hemisphere speech sites. That is, as far as lipreading goes, Mrs. T suffered a functionally bilateral lesion that impaired this skill. Under these conditions, visual information that requires access to the language processor takes more circuitous routes, possibly via the anterior commissures. It may be, then, that the relevant visual information can be processed in either hemisphere, and that only in cases such as Mrs. T's, where direct access to both routes is blocked, will an impairment be seen. The right hemisphere advantage observed in normal undergraduates by Campbell (1985) may then have its roots in the well-established superiority of the right hemisphere to perform visual processing on brief and impoverished inputs, and/or in a mild advantage to the right hemisphere for lipreading, which will be evident in reaction time and overall accuracy effects rather than all-or-none impairments.

11.4. Pure word deafness and lipreading: neuropsychological evidence for the localization of subprocesses in lipreading There are sparse reports that patients with auditory pure word deafness can be helped by lipreading: Two published cases (Miceli, 1982; Saffran, Marin, & Yeni-Komshian, 1976) state, without further support, that their patients were "helped by lipreading." In these patients auditory phonetic discrimination was poor. They provide some, very indirect, evidence concerning lateralization and phonetic processing of lipread material. Saffran et al.'s patient had a left temporal lesion. Under dichotic stimulation he could not report right ear inputs. Careful auditory testing using synthesized speech stimuli showed that this patient systematically confused the consonants in isolated syllables. He had special difficulty in discriminating manner of articulation (in particular, presence or absence of the phonetic feature of voice), whereas place of articulation was relatively less impaired. Miceli's patient, by contrast, had bilateral temporal lesions and

Lipreading, neuropsychology, and immediate memory

273

did not show suppression of right ear inputs in dichotic testing. She was relatively more impaired in processing place than manner of articulation. By subtraction, this might suggest that the phonetic feature dimension of place of articulation might sometimes be processed in the right hemisphere. Place of articulation can often be lipread, whereas manner usually cannot. Could it be that a "spare generator" for the analysis of place of articulation may have developed in the right hemisphere as a result of a slight advantage to the right hemisphere in the analysis of speech from seen lips? This would fit with Mrs. T's lipreading problems in suggesting that, for her, potentially useful phonetic information, derived in the right hemisphere fails to reach the critical left hemisphere speech analysis sites via posterior commissural connections (her left hemisphere lesion was medially placed and would disrupt such connections). It is possible, then, that the early information-processing stages of the analysis of seen speech might involve some mechanisms, sited posteriorly in the brain, and possibly localized somewhat more to the right hemisphere. These could be considered to be extrinsic to "normal speech perception." Once phonetic analysis is under way, however, it seems to me unlikely that this should occur in parallel for heard and for lipread material: Rather, it would seem that a single phonetic processor accepts and uses seen and heard inputs in its analysis. Speech processing is not modality bound - rather it is modular and unique. The speech module accepts seen and/or heard input of the appropriate sort and can effectively integrate complementary and supplementary inputs.

11.5. Disorders of phonological processing and lipreading On the view outlined earlier a more central impairment of language processing (namely, phonological problems in spoken language processing), in the absence of very impaired phonetic discrimination skill, should not necessarily produce dissociated impairments between seen-and-heard and heard speech. Two patients currently under investigation by S. Franklin and D. Howard offer an instructive contrast. Both MK (see Howard & Franklin, this volume, chapter 12) and DRB (Campbell et al, 1988) can distinguish minimal phonetic contrasts and show other indications of intact phonetic registration of speech. Both patients can name pictures correctly and can read well; they do not have deficits in specifying correct phonological output nor in accessing language from the written word. However, they differ in other respects. MK appears to have impaired phonological descriptions of spoken words and cannot reliably make auditory lexical decisions. DRB is fast and accurate at auditory lexical decisions; he is poor, however, at comprehending spoken words unless they are highly concrete. That is, DRB's deficits appear to be postlexical (concerned with semantic access); MK's deficits are lexical (impaired lexical representations). Should seeing the speaker have any effect on these patients' auditory speech

274

Campbell

classification skills? The suggestion made here is that lipreading could help an aphasic patient to the extent that it provides a "spare fuel can" - an extra phonetic resource - that will help provide bottom-up information to the phonological speech recognition system. If this system is intact, lipreading might help patients by "kick starting" the speech recognition device and have beneficial effects "onwards and upwards." The extent to which lipreading might help patients like these should then reflect the utility of this resource as tasks involving lexical, syntactic, semantic memory processes are imposed. Remember that Reisberg et al. found that lipreading was useful to undergraduate subjects in following clear auditory text that was cognitively demanding. Lipreading, which adds extra phonetic "fuel," can have beneficial effects on higher processing in normal speech processing. DRB appears to be able to use lipreading in just this way. His ability to make semantic classification judgments when he can see as well as hear the speaker is significantly better than when he can only hear the speaker (concreteness effects are diminished). The interactive activation view to which I subscribe would predict, however, that lipreading may not be the only way to improve his aural classification skills; making the speech signal clearer (louder) or adding phonetic information in some other way should also have beneficial effects. And, conversely, DRB could be predicted to be worse than normal when auditory lexical decision is made harder by adding white noise. These tests are at present under way. The dissociation between DRB's ability to process heard and seen-and-heard speech may be a reflection of a more general debility: All or any phonetic "boosters" will improve his performance because he has intact phonological representations at the most abstract level. He knows what speech sounds should be. His problem is that acoustic information is not making effective contact with phonological representations. MK, however, seems to be less helped by lipreading. Why should this be? Other evidence (see Howard & Franklin, 1988) suggests that MK's poor auditory lexical decision performance reflects the fact that his auditory lexicon does not contain sufficiently well specified phonological forms for use in such discrimination tasks. Under these circumstances an extra phonetic resource is less likely to be useful: If the phonological descriptions of words are awry, then however good the phonetic input, it will still be unable to produce reliable auditory word identification. In MK, then, lipread speech, like heard speech, is blocked from processing beyond the phonetic discrimination level. In this context it is worth noting a classical description: " Her husband said that she understood spoken language better when she did not see the speaker - when she was spoken to from a distance, and when spoken to quietly and not loudly . . . she herself volunteered that they (grown-up people) had too many teeth — " (Bramwell, 1897, 1987, pp. 253-254). If this account is taken, it makes the substantial if negative point that lipreading cannot provide a "third route" to speech comprehension when the phonetic-phonological analysis system is impaired. The only way in which lipreading

Lipreading, neuropsychology, and immediate memory

275

left hemisphere

right hemisphere

Audition

Vision

(heard speech)

(seen face)

i

,... t

iphbnelic analysis! • of seen place , ""'of articulation i

phone fie"/ phonological processor

i

i

attentjonal pragmatic, prosodic ^processes

phonological (non-lexical) repetition

fc

Speech Output Processes Sites S

of Impairment

Mrs. T's disconnection

$ £ DB's (partial) impairments of MK's (dense) impairments

Figure 11.1. Possible information flow (and sites of impairment) for lipread speech. can help patients with varieties of cortical "deafness" is by providing an extra phonetic input when central phonological representations are usable. Figure 11.1 summarizes some of the possible relationships of lipreading to auditory speech processing discussed earlier. Possible sites of deficit are sketched in for DRB, MK, and Mrs. T. The box labelled phonetic/phonological processor in the figure is prized half-open by the dissociation between DRB's and MK's auditory processing skills. The next section of this chapter, which considers repetition and auditory-verbal immediate memory from the point of view of lipreading, opens this box even more and considers its relationship to other structures and processes.

276

Campbell

11.6. Beyond speech registration: immediate verbal memory and lipreading Hearing speech and remembering it for immediate recall are processes that overlap. Interactive models of speech recognition indicate this quite clearly, for it is axiomatic to such models that they use aftercoming context to identify speech sounds, and this demands some maintained representation of what is to be clarified by the aftercoming speech event. So, for instance, the TRACE model of speech recognition (McClelland & Elman, 1986) is quite explicit that the registration and identification of a speech sound leave a primary record; TRACE is not an acronym but rather a description of the persistent state of activation of the speech processor. One phenomenon that is often taken to illustrate the continuity of speech processes and immediate recall in just this way is the modality-specific recency effect and the suffix effects that go with it (Crowder & Morton, 1969; Crowder, 1983). In the immediate recall of auditory lists the last item(s) are relatively well recalled in comparison to the last item(s) of a written list (the auditory recency effect). Auditory recency is abolished by an aftercoming speech sound. This is the suffix effect. These are powerful and robust effects. Because the lexical status of the suffix and strategic task components do not reduce the suffix effect, such effects are said to be precategorical, and a useful way to conceptualize them is to envisage that the auditory speech stream leaves a trace, a sensory residue, which, if not overwritten (as by a suffix) can be accessed at recall; this is precategorical auditory storage (PAS). In the last ten years it has become clear that such effects are not auditory in origin. Lists that are spoken aloud, or that are mouthed while being read, or that are lipread from a silent speaker all show recency in comparison to written lists; this recency, moreover, is reduced or eliminated by an aftercoming speech stimulus — a suffix — as long as this, too, is spoken aloud or mouthed or lipread. There is a good deal of crosscommonality in these suffix effects; so, for instance, lipread recency can be quite specifically reduced by a mouthed suffix and mouthed recency by an auditory suffix — but the modalities are not entirely interchangeable. In general, an auditory suffix wipes out recency for lipread, heard, or mouthed lists most effectively; and some studies suggest that auditory list recency can be somewhat more resistant to mouthed and lipread suffixes than are mouthed or lipread lists, respectively. This suggests to me that there may be two components to such recency and suffix effects in immediate recall: a major phonetic component, common to lipread, mouthed, and heard stimuli, and a minor component that may be modality specific and sometimes has "sensory" characteristics (Greene & Crowder, 1984; Campbell, 1987a, b; Gathercole, 1987). That the major component of auditory recency is phonetic is also suggested by studies using ambiguous acoustic stimuli. In Ayres, Jonides, Reitman, Egan, and Howard's (1979) study, subjects were instructed that an unattended ambiguous sound was a "trumpet note" or "a spoken wah sound." Only in the latter condition did a suffix effect obtain for spoken digit recall.

Lipreading, neuropsychology, and immediate memory

277

Silent-lipread recency and suffix effects can be shown to be more "purely" phonetic than many auditory ones. They are determined less by the sensory similarity between spoken list and suffix than by perceived phonetic commonality. So, for example, a man's voice heard speaking a suffix after a woman has been seen (but not heard) saying the list has just as disruptive an effect on recency as hearing a woman's voice. For heard lists, there is a differential effect of voice; suffix effects are most pronounced when the same voice speaks the suffix and the list (Morton, Crowder, & Prussin, 1971; Campbell & Dodd, 1984). We have shown recently that adding phonetic information to a lipread list, by adding a burst of noise corresponding to vocal cord vibration, produces recency and suffix functions that are predicted only by the extent to which the suffix and list are phonetically specified and not at all by sensory factors. The phenomenal impression of this stimulus configuration seems to be like listening to speech through noise - the "buzz" corresponding to the electronically generated sound signal from vocal cord vibration can integrate with the synchronized lip movements to generate rather effective speech comprehension. In recalling lists of such //buzz-speech," a "buzz-speech" suffix is less effective in reducing recency than a clearly heard (and seen) "natural" speech suffix (Campbell, Garwood, & Rosen, 1988). Any theory that bases an explanation for recency and suffix effects on the physical similarity of list and suffix (see, e.g., Morton, Marcus, & Ottley, 1981) cannot easily encompass this finding. It is the phenomenal, not the sensory, characteristics of the speech signal that seem to be responsible, to a large extent, for recency and suffix effects in auditory immediate list recall. However, this is not the whole story: a lipread or a buzz-speech input does seem to leave a rather more fragile "trace" than a heard one. For we also found (Campbell et al., 1988, experiment 2) that adding an irrelevant pure-tone burst, lasting a quarter of a second, to each buzz speech stimulus eliminated recency and suffix effects for such lists. Natural heard lists still show recency when such a pure tone is added, and suffix effects in this case are determined by whether the suffix also has a pure tone accompaniment (Routh & Lifschutz, 1975). Only a tone-accompanied suffix eliminates recency for a natural, tone-accompanied list. This suggests that - to some extent at least - the "sensory-specific" component of natural auditory recency and suffix effects may often reflect "grouping" or "attentional" phenomena, as Kahneman (see, e.g., Kahneman & Henik, 1979) has long claimed. Apart from the impact of studies of lipreading on the reformulation of auditory memory processes, the immediate memory of silently lipread material has further interesting characteristics. Two are important in the current context. First, memory for lipread material does not seem to use a visuospatial resource, such as the "visuospatial scratch pad" (Logie, 1986) in a particularly demanding way. In one experiment, T. Winterton and I (unpublished) asked subjects to memorize a path through a 3 x 3 cell matrix when the instructions were given by lipreading, in writing, or by ear (the instructions were five single-step instructions of the form "left, right, up,

278

Campbell

down"). There was no systematic effect of mode of presentation of the instructions on the ability to remember a path through the matrix. Nevertheless, subjects in lipreading memory experiments do complain that they find the task tiring, and we wonder whether some extra processing resource, other than that needed to keep one's eyes on the screen (auditory memory tasks do not have this visual alerting component), might be implicated in lipreading. Second, the immediate recall of silently lipread lists is no more and no less affected by concurrent articulation during recollection than is the immediate recall of heard lists. By contrast, written lists are more deleteriously affected than either heard or lipread lists by this manipulation (Campbell & Dodd, 1984). Taken together, these findings strongly suggest that the rehearsal and maintenance characteristics of the immediate recall of silently lipread material are precisely the same as those for heard speech. The recall of silently lipread material by hearing people is no more reliant than heard recall on visuospatial or on articulatory memory systems. We can (tentatively) add a rider to this. Lipreading does not seem to add information to the trace used for immediate ordered list recall. Seeing the speaker does not improve auditory digit span where audition alone can support repetition adequately. J. Garwood and I (unpublished) tested 70 or so Oxford undergraduates' digit span under two conditions - audition alone or audition and vision (seeing as well as hearing the speaker). All subjects were tested at different list lengths - five-, seven-, nine- and eleven-digit list recall. At the "comfortable" auditory levels used, digit span was not increased by seeing the speaker. Nor were there qualitative indications (i.e., serial position differences) of any contribution from the seen speech input. The {short-term memory or list recall) system seems to be relatively insensitive to whether the source of information is lipread, heard, or a natural or artificial ("buzz speech") combination of both.

11.7. PAS and phonological memory: a radical alternative? Since the lipreading and other studies lead me to believe that PAS-like effects are largely phonetically, not sensorily, determined, it seems possible that the traditional distinction between PAS-like storage effects and other phonological effects in short-term memory (phonemic similarity effects; word length effects - see Baddeley, this volume, chapter 2) may be more apparent than real. However, any reformulation of immediate memory processes that blurs the distinction between sensorylike and central-like processes must envisage a metaphor different from that of straightforward box stores. One alternative approach (Barnard, 1985) suggests not that immediate auditory-verbal memory is best conceived as a set of storage systems with somewhat different operating characteristics and capacities but rather that the processing of speech is accomplished by a structured system of reciprocally connected components that

279

Lipreading, neuropsychology, and immediate memory

heard and seen speech .

phonetic units

perceiving speech

phonological units 'abstract' phonemic units

mil

1111 IN

IT

m

P.I. input phoneticphonological system

1 i

A rehearsing and repeating speech

corresponding units ( a t each level) can be reciprocally activated

'abstract' phonemic units

lexical processors

I I I JJ from written

producing speech

(non-lexical) material

phonological units phonetic units

P.O. output phonetic/ phonological system

spoken or mouthed speech Figure 11.2. An "open-box" model of two phonological processors and their interaction. This model is posited to underlie immediate verbal repetition performance.

happen to have the property of maintaining a particular state of activation as a function of input, output, and task demands. Short-term memory "capacities," on this type of account, are by-blows of psycholinguistic processes that need to be buffered in order to produce and perceive speech. It is in the operation of this system that we might seek to understand all short-term memory phenomena. Barnard stops short of suggesting that PAS-like effects could be incorporated into such phonological systems. I do not. Figure 11.2 sketches an "open-box" model that indicates the main features of a system that subserves immediate memory performance without the presumption (for I

280

Campbell

maintain it is unwarranted) of a purely sensory auditory store. This system can readily account for auditory, lipread, and mouthed short-term memory effects. Figure 11.2 bears more than a passing resemblance to the outline of phonological buffer processes shown, for example in Howard and Franklin (this volume, chapter 12). Immediate memory performance is characterized by the reciprocal interaction of a phonetic-phonological speech recognition device (phonological input, PI) and a similar device that specifies speech output (phonological output, PO). Each of these is independently connected to lexical processes and to visual (letter and word) analysis. So far, this arrangement is that of traditional "box models." It is the internal structure of these devices, however, that gives rise to several of the important characteristics of immediate verbal memory, and which leads me to describe this approach as one that "opens the box." Within each of these two systems, levels of representation from the purely phonetic (acoustic and lipread) to the most abstract phonemic (the level that permits some people to "hear" the b in comb) are represented in fully interconnected fashion. A particular speech event is represented as a stable pattern of distributed activity across a subset of units in each system (either PI or PO). Units are organized hierarchically within each level of representation (TRACE-like units work in PI, and comparable speech output units in PO). Within each system (PI or PO) activation of units at a particular level can ramify to levels above and below it. So, for instance, if the word pat is heard, phonetic and phonemic units corresponding to those parts of the word are maintained in mutual activation while units incompatible with it are inhibited. TRACE (McClelland & Elman, 1986) details how and when such activation persists for speech sounds, its activation parameters being set to model auditory speech identification phenomena. A further assumption is that reciprocal activation is between corresponding parts of each system; a fine-phonetic representation in PI will map onto an analogous site in PO, for example. The ability to remember a particular spoken event will depend on the state of activation within each of these systems. The principle of reciprocal activation between corresponding input and output representations helps to capture the "privileged loop" aspect of auditory-verbal repetition, although no claims will be made here concerning the allocation of cognitive resources for repetition tasks. How woulcl such a model cope with immediate memory phenomena? A sketch is provided in the following section.

11.8. Modality-specific recency (PAS-like) effects At first sight the PI-PO model outlined in Figure 11.2 does not appear to generate modality-specific effects in immediate recall - effects that are robust and replicable and that gave rise to the PAS concept. The internal structure of PI and PO, however, can account for modality-specific

Lipreading, neuropsychology, and immediate memory

281

effects. When written lists are read for recall, a phonological representation of those lists is activated in PO. Howard and Franklin (chapter 12) give clear evidence that only PO can be activated directly by written input. Under the usual conditions of such experiments, only a phonological representation is so activated; phonetic details are not needed in order for the visual phonological input to drive the PI system and establish a rehearsal loop. When written material is processed in this way, there is no phonetic activation from sensory sources to PI to "enrich" this phonological record. The maintained, reciprocal phonological activation in PI and PO determines visual memory span. By contrast, a heard (or lipread) list activates phonetic components in PI that in turn activate units at the phonemic level. The process whereby this representation in PI accesses a corresponding one in PO and these are reciprocally maintained constitutes rehearsal. At recall these phonological representations can be accessed, but the phonetic activation of PI will also leave a trace and contribute to recency. Written material that is mouthed at presentation also causes recency in immediate recall. This could be because on reading the list item not only the phonological but also the full phonetic specification of that word form is accessed in PO. The full phonetic form is activated in order to control the sequence of articulatory processes needed to produce the item - whether by speech or mouthing. This activates a corresponding phonetic level in PI and so provides the extra recency resource as well as the rehearsed component of the list. Similar proposals have been made by Crowder (1986) to account for mouthed recency effects, although cast in a rather different form. Why it should be recency rather than, say, overall accuracy that benefits particularly from such phonetic activation is a complex question and answers will not be attempted here.

11.9. The phonological store and rehearsal The most widespread (modal) models of short-term memory suggest that phonological effects in recall can arise from the operation of three rather different stores and processes; one is "precategorical" and was discussed earlier. This can be characterized as transient, nonstrategic, phonetically organized store of limited capacity, with modalitysensitive characteristics (taking modality nonliterally to include speechlike inputs; Crowder & Morton, 1969; Crowder, 1986). The other two could be considered more "central": Thus, in Baddeley's working memory formulation (see this volume, chapter 2) there is a phonological store (PS) that is less transient that the precategorical system and of higher capacity, yet not under strategic control. Then there is the articulatory rehearsal loop (AL) that maintains and refreshes the phonological store and that is under strategic (central executive) control. These latter two phonologically organized components are not particularly modality sensitive, unlike the precategorical,

282

Campbell

immediate registration system. The PS and the AL are thus taken to be the core components of immediate short-term memory and therefore discontinuous with speech perception and speech production mechanisms. On the PI-PO model, however, this is not the case. Characteristic core, or central, short-term memory performance (effects such as phonemic similarity and word length effects in immediate recall) arises when the higher levels of representation - the more "purely" phonological — are activated in PO and PI and can maintain themselves, reciprocally. To the extent that such maintenance (rehearsal) obtains and to the extent that activation persists at the phonological level in PO and PI, so "memory" characteristics will emerge in performance. In the most literal sense (more literal than even in Baddeley's working memory metaphor), immediate verbal memory performance is the simple reflection of the act of speaking to oneself and hearing oneself do it. The systems that allow this to occur can be activated by internal inputs (reading, naming pictures) or by external ones (repeating heard speech). In this maintained interaction, however, it is not necessary for the fully detailed phonetic specification of the items to be upheld or fully activated, although it can be - for example, when remembered material is mouthed at presentation or spoken at recall. Word length and phonemic similarity effects - indicators of the AL and PS components of the working memory system - can be easily accommodated into the PIPO formulation. The word length effect reflects the phonological capacity (number of phonological segments) of reciprocal PI-PO activation under conditions of silent rehearsal. Fewer long than short words can occupy this phonological working space. The phonemic similarity effect reflects the fact that, both in PI and PO, the representation of phonologically similar units is more likely to become corrupted during reciprocal activation than that of dissimilar units. On the PI-PO formulation it is not the case that an input store (the PS in working memory models or PI in the PI-PO model) is necessarily more sensitive to the phonemic similarity effects while an output mechanism (the AL, or PO) is more sensitive to word length effects. Rather, these are qualitatively quite different phenomena reflecting different aspects of the workings of an integrated system whose components may not necessarily show the characteristics of the working memory system components. The subverting of immediate memory by concurrent articulation indicates that such maintained activation of PO and PI is channel limited; when one has to say ba-ba-ba while trying to remember verbal material, recall tends to be impaired. What is left when PO and PI are being used in this way enables us to gauge the structure and capacity of these systems and of their links to lexical and semantic representations. Moreover, this formulation suggests a different way to capture the effects of articulation during presentation and recall; it suggests that the level of phonetic complexity of the activity should have detectable effects. So, for instance, saying Peggy Babcock should affect recency in the repetition of auditory lists more than saying ba-ba-ba.

Lipreading, neuropsychology, and immediate memory 11.10. Lipreading, immediate memory, and neuropsychology Although it is the study of lipreading in the context of immediate memory that has led to this reappraisal of immediate memory mechanisms, nothing has yet been said about the role of lipreading in immediate memory in neuropsychological patients. There is little to say, for no systematic studies have, to my knowledge, been reported. Yet the facts and the perspective offered here make sufficiently clear predictions. First of all, we are unlikely to find a patient with normal hearing and acquired brain damage, who, through lipreading, could access a system for representation and recall that is independent of that which is used when listening to speech. Lipread recall is, in all important respects, indistinguishable from heard recall in normal individuals. Lipreading uses the PI system just as hearing does. Thus, all other things being equal, we would expect the patient with a "classical" impairment of short-term memory (see Shallice & Vallar, this volume, chapter 1) to show just the same advantage for written over lipread list recall as occurs for written over auditory list recall. I have informally tested digit span in two diagnosed "conduction aphasics"; in both cases lipread span and auditory span were comparable, and both were smaller than visual (written letters and digits) repetition span. Additionally, in these patients seeing the speaker did not add anything to their impaired auditory span. I have already mentioned that our laboratory investigation suggests that normal immediate auditory memory performance on span-related tasks is not improved by seeing the speaker either. Would one expect to find dissociations in favour of seen-and-heard over heard speech in any patients? DRB is helped by lipreading in achieving some comprehension of spoken speech. He is much helped by seeing the speaker when he is asked to repeat speech. His auditory digit repetition span is one, and for nonwords it is zero. But he can repeat some nonwords when he sees the speaker. Conversely, their patient MK (see chapter 12), who also has a repetition span of one auditory digit, is not helped by lipreading the speaker. I characterized DRB's impairment and the support he gained from lipreading in terms of somewhat impaired auditory access to the phonological analysis system, the system component described here as PI. This need not necessarily impair auditory lexical decision when lexical connections to the intact PI are good and when some phonetic information gets through (in contrast to "pure word deaf" patients). No further devices are needed to explain DRB's repetition improvement with lipreading. As with his comprehension, the simple addition of a reliable phonetic component, in the form of seen lip movement, can be enough to activate the intact PI to form lasting representations of speech to be used at recall. This analysis of DRB's performance also suggests that some other "centrally deafened" patients might well be aided in repetition when they see the speaker. MK was characterized differently. Lexically specified phonological information that could be used in various discrimination (as opposed to production) tasks is not well

283

284

Campbell

specified in MK. Thus, even if lipreading had access to PI it might do little good interpreting heard words. On its own, this need not necessarily stop MK from repeating phonetically acceptable material quite well. Yet he is unable to do this. In order to explain why MK cannot repeat heard items one must consider his other major deficit. This is described in Howard and Franklin (this volume, chapter 12) as a disconnection between PI and PO. Thus, even though MK can read words aloud (PO is intact and can use written material), he cannot "send" phonetic or phonemic details of this to PI. Nor can he maintain a phonetic record that has registered in PI, for the connection with PO is missing. The absence of PI—PO reciprocal activation (the rehearsal loop) in MK might conflate his difficulties in auditory lexical discrimination. One way to know "what a word sounds like" is to know "in one's head" how the written word is spoken. MK is denied this knowledge. It appears that even mouthing words as he reads them is not sufficient to bridge PO and PI. Neither seeing speech nor speaking silently can help him. 11.11. Conclusion In this review I have tried to show how predictions concerning the role of lipreading in immediate memory in aphasic subjects may be premature, but that some of the groundwork has been done. In particular, I maintain that lipreading can be integrated into a theory of normal auditory speech perception without too much fuss - as long as such a theory is highly interactive. In turn, this generates predictions and analyses of various aphasic associations and dissociations between the processing of speech that is heard and speech that is heard and seen. There may be a small, possibly strategic, component of the early stages of analysing speech information from faces that is not localized to the speech hemisphere, but by far the largest part of the process of identifying speech from seen lip movement is served by auditory speech processes located in their classical home. These processes ramify into short-term memory processes. The view from lipreading suggests that sensory storage in immediate memory may be illusory, and a radical view of short-term memory is offered in which separate stores are discarded in favour of a phonologically (phonetic—phonemic) organized speech input and speech output system where units in each representational component are reciprocally activated in "point for point" correspondence. Such an interactive system demands explicit internal structuring of a highly interconnected, interactive kind. We are now poised to investigate systematically and, within a simple and coherent theoretical framework, the role of lipreading in immediate memory disorders in patients. References Ayres, T. J., Jonides, J., Reitman, J. S., Egan, J. C, & Howard, D. A. (1979). Differing suffix effects for the same physical suffix. Journal of Experimental Psychology: Human Learning and Memory,

5, 315-321.

Lipreading, neuropsychology, and immediate memory

285

Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197—258). London: Erlbaum. 197-258. Bramwell, B. (1897). Illustrative cases of aphasia. Lancet, 1, 1256-1259. Reprinted in Cognitive Neuropsychology (1984), 1, 245-258. Campbell, R. (1985). The lateralisation of lipread sounds: A first look. Brain and Cognition, 5,1-22. Campbell, R. (1987a). Common processes in immediate memory. In D. A. Allport, D. MacKay, W. Prinz, & E. Scheerer, (Eds.), Language perception and production: Common mechanisms in listening, speaking, reading, and writing (pp. 1 3 1 - 1 4 9 ) . N e w York: Academic Press.

Campbell, R. (1987b). Lipreading and immediate memory processes. In B. Dodd & R. Campbell (Eds.), Hearing by eye (pp. 243—256). London: Erlbaum. Campbell, R. (1988a). Lipreading. In H. Ellis & A. Young (Eds.), The handbook of face processing (pp. 187-207). Amsterdam: Elsevier. Campbell, R. (1988b). Tracing lip movement: Towards a theory of lipreading in speech perception. In R. Campbell (Ed.), Visible language: A special issue on lipreading. Providence, RI: Brown University Press. Campbell, R., & Dodd, B., (1984). Aspects of hearing by eye. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and Performance X (pp. 300-311). Hillsdale, NJ: Erlbaum. Campbell, R., Garwood, J., Franklin, S., Howard, D., Landis, T., & Regard, M. (1988). Neuropsychological

studies of the auditory-visual

fusion illusion. Paper presented to the

Experimental Psychology Society, Edinburgh. Campbell, R., Garwood, J., & Rosen, S. (1988). On adding sound to lipread lists. Memory and Cognition, 16, 210-219. Campbell, R., Landis, T., & Regard, M. (1986). Face recognition and lipreading: A neurological dissociation. Brain, 88, 287-294. Crowder, R. G. (1983). The purity of auditory memory. Philosophical Transactions of the Royal Society of London, B, 302, 2 5 1 - 2 6 5 .

Crowder, R. G. (1986). Auditory and temporal factors in the modality effect. Journal of Experimental Psychology Learning, Memory, and Cognition, 12, 268—278.

Crowder, R. G v & Morton, J. (1969). Precategorical acoustic storage (PAS). Perception and Psychophysics, 5, 365-373. Gathercole, S. (1987). Lipreading: Implications for theories of short term memory. In B. Dodd & R. Campbell (Eds.), Hearing by eye: The psychology of lipreading (pp. 2 2 7 - 2 4 2 ) . London:

Erlbaum. Greene, R. L., & Crowder, R. G. (1984). Modality and suffix effects in the absence of auditory stimulation. Journal of Verbal Learning and Verbal Behavior, 23, 3 7 1 - 3 8 2 . Howard, D., & Franklin, S. (1988). Missing the meaning: A cognitive neuropsychological analysis of single word processing in a fluent aphasic. Cambridge, MA: Bradford Books.

Jeffers, J., & Barley, M. (1971). Speechreading. Springfield, IL: Thomas. Kahneman, D., & Henik, A. (1979). Perceptual organisation and attention. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptual organisation (pp. 181-211). Hillsdale, NJ: Erlbaum. Logie, R. H. (1986). Visuo-spatial processing in working memory. Quarterly Journal of Experimental Psychology, 38A, 229-248. McClelland, J. L, & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86.

McGurk, H., & MacDonald, J. (1976). Hearing lips and seeing voices. Nature, 264, 746-748. Massaro, D. W. (1987). Speech perception by ear and eye. In B. Dodd & R. Campbell (Eds.), Hearing by eye: The psychology of lipreading (pp. 53-83).

London: Erlbaum.

Miceli, G. (1982). The processing of speech sounds in a patient with cortical auditory disorder. Neuropsychologia, 20, 5—20.

Morton, J., Crowder, R., & Prussin, H. A. (1971). Experiments with the stimulus suffix effect. Journal of Experimental Psychology, 91, 161—180.

286

Campbell

Morton, J., Marcus, S. M , & Ottley, P. (1981). The acoustic correlate of "speechlike": A use of the suffix effect. Journal of Experimental Psychology General, 110, 56S-593. Reisberg, D., McLean,J., & Goldh'eld, A. (1987). Easy to hear but hard to understand: A lipreading advantage with intact auditory stimuli. In B. Dodd & R. Campbell (Eds.), Hearing by eye: The psychology of lipreading (pp. 97—114). London: Erlbaum. Routh, D. A., & Lifschutz, A. J. (1975). An asymmetrical effect of similarity in the attention of stimulus suffix interference. Journal of Verbal Learning and Verbal Behaiour, 14, 95-104. Saffran, E., Marin, O. S. M., & Yeni-Komshian, G. H. (1976). An analysis of speech perception in word deafness. Brain and Language, 3, 209-228. Samar, V., & Sims, D. G. (1983). Visual evoked-response correlates of speechreading performance in normal-hearing adults: A replication and factor analytic extension. Journal of Speech and Hearing Research, 26, 2-9. Samar, V., & Sims, D.G. (1984). Visual evoked-response components related to speechreading and spatial skills in hearing and hearing-impaired adults. Journal of Speech and Hearing Research, 27, 162-172. Summerfield, Q. (1987). Some preliminaries to a comprehensive account of audio-visual speech perception. In B. Dodd & R. Campbell (Eds.), Hearing by eye: The psychology of lipreading (pp. 3-52). London: Erlbaum. Summerfield, Q., & McGrath, M. (1984). Detection and resolution of audio-visual incompatibility in the perception of vowels. Quarterly Journal of Experimental Psychology, 36A, 51-74.

12. Memory without rehearsal DAVID HOWARD AND SUE FRANKLIN

12.1. Introduction It is widely accepted that rehearsal plays an important role in short-term recall, but the nature of that role is much less clear. Evidence for the functions of rehearsal has come from two different areas; the first involves experiments in normal subjects that investigate the effects of dual tasks such as concurrent articulation of irrelevant material ("articulatory suppression"; see, e.g., Baddeley, Lewis, & Vallar, 1984) or counting (e.g., Peterson & Peterson, 1959), which are assumed to interfere with rehearsal. However, the interpretation of these kinds of studies is problematic; the dual task may interfere with cognitive and mnestic functions other than the process of rehearsal, or rehearsal may continue to be performed despite the dual task. It is impossible to show a priori that the dual task employed interferes only with rehearsal; a circularity is unavoidable. The second general strand of evidence on the importance of rehearsal comes from studies of patients with developmental or acquired disorders of short-term memory or articulation. Here, too, a degree of logical circularity seems hard to avoid. Vallar and Baddeley (1984a, b), for instance, report data from a patient, PV, with restricted shortterm memory span. They show that in recall PV behaves as they expect a patient who cannot rehearse would behave, but note that they have no independent evidence that rehearsal processes are impaired. Indeed, Baddeley (1986) suggests that PV could rehearse if she chose; she does not do so because rehearsal cannot be used to "refresh" a defective phonological memory store. Other investigators have examined memory processes in patients whose articulatory processes are disrupted by congenital or acquired anarthria (e.g., Baddeley & Wilson, 1985; Bishop, 1985; Vallar & Cappa, 1987). Here, despite the failure to articulate, the subjects show the pattern of performance

We thank Dr. R. Zeegan for permission to study this patient; Sally Byng, Veronika Coltheart, and Eirian Jones for allowing us to use their test materials; and Brian Butterworth and John Skoyles for discussing the data and issues with us. Veronika Coltheart, Steve Avons, Derek Besner, Myrna Schwartz, and Giuseppe Vallar provided valuable comments on various drafts. The authors were funded by separate grants from the Medical Research Council.

287

288

Howard and Franklin

thought characteristic of rehearsal; the conclusion drawn is that rehearsal is therefore possible without overt articulation. In this chapter we try to avoid this kind of circular reasoning by reporting data from a patient, MK, for whom we have independent reasons for believing that rehearsal is impossible. We have documented these reasons at length elsewhere (Howard & Franklin, 1987, 1988), but we summarize the relevant evidence in the case history presented here. A widely influential model of short-term memory performance is Baddeley's model of "working memory" (see Baddeley, 1986). With auditory presentation, items enter a phonological short-term store (PSTS) with limited capacity, where they are held in phonologically coded form. The contents of the PSTS can be refreshed by a process of articulatory rehearsal; the operation of this "articulatory loop," which can be prevented by articulatory suppression, is responsible for word length effects. When items are presented visually, they can only access the PSTS via the articulatory loop. The problem in applying this model to data from patients with acquired disorders of memory or language is that it is unclear how the components of the working memory model relate to lexical and sublexical processing capacities. That there must be a close relationship is clear: Short-term recall is, for example, better for real words than nonsense syllables (e.g., Schweikert & Boruff, 1986). Allport (1984) argues for the existence of separate input and output phonological buffers, and Monsell (1987) has proposed that rehearsal involves cycling information between them, using the same processes that are involved in nonmemory tasks with words or nonwords. Monsell argues that rehearsal involves one sublexical process for converting input phonology to output phonology (which is also required for verbal repetition of spoken nonwords). For converting output phonology to input phonology Monsell suggests that two different processes may be available (cf. Crowder, 1983); one involves generation of an articulatory code, whereas the other recirculates phonological information. Recent revisions of the "logogen model" of lexical processing incorporate two processes for converting input phonology to output phonology, which do not involve semantic mediation (cf. Morton & Patterson, 1980). One links the auditory input lexicon and the phonological lexicon directly, and is therefore available only for real, familiar words. The second link is sublexical; it can be used for repeating nonsense words, as well as with real words. A means of converting output phonology into input representations is also required for tasks such as comprehension of pseudohomophones (Howard & Franklin, 1987). We have been able to show that the patient who is our subject here, MK, does not have access to any of these processes for converting input phonology to output phonology or vice versa. On the assumption that these are the processes that mediate rehearsal, MK will be a patient who cannot rehearse. Our investigations will cover two areas in which rehearsal processes have been said to play a crucial role: list memory and judgments of rhyme and homophony. Before presenting

Memory without rehearsal

289

the evidence we will briefly consider the available evidence from studies with normal subjects for the role of rehearsal in these two areas.

12.1.1. List memory The effects of suppression on list memory are now well established. With auditory list presentation without suppression, list recall is worse for phonologically similar lists (e.g.CG TV) than for dissimilar lists (e.g., QL YR), and is better for lists of short words than long words. With suppression during presentation and recall, the word length effect disappears but the phonological similarity effect remains (Baddeley et al, 1984). With visual presentation without suppression both phonological similarity and word length affect recall; suppression abolishes both of these effects (Murray, 1968; Baddeley, Thomson, & Buchanan, 1975; Besner & Davelaar, 1982). On the basis of this pattern Baddeley (1986) has proposed that rehearsal is responsible for the word length effect, and is necessary for visually presented items to access the PSTS, which is responsible for the effects of phonological similarity.

12.1.2. Judgments of rhyme and homophony There are experimental data on the effects of suppression on three different kinds of tasks. The ways in which these have been described have caused considerable confusion in the literature. The terminology we will adopt is as follows: (a) Rhyme judgments. Pairs of words (or nonwords) are presented and the subject has to decide if they rhyme or not. In well-designed experiments orthographic similarity will be no guide to whether words are phonological rhymes, by using nonrhymes that are orthographically similar (i.e., "eye rhymes," e.g., rear bear) and rhymes that are orthographically distinct (i.e., "ear rhymes," e.g., fare pair). (b) Homophone judgments. The subject has to decide whether pairs of words or nonwords sound the same. Again, with suitably designed experimental materials, this decision cannot be made on orthographic similarity. (c) Pseudohomophone detection. The subject has to decide whether a nonword string would sound like a real word when it is pronounced; so the response to krain is yes and to prain is no. (Note that Besner [1987], while drawing a clear distinction between rhyme judgments and the other tasks, calls both homophone judgments and pseudohomophone detection homophone judgments.) Suppression impairs rhyme judgments with written presentation of pairs of real words (Besner, Davies, & Daniels, 1981, Experiment 3; Wilding & White, 1985; Johnston & McDermott, 1986; Brown, 1987, Experiment 1; cf. Kleiman, 1975), with pairs of nonwords (Besner et al, 1981, Experiment 3; Brown, 1987, Experiment 1), and a real word paired with a nonword (Richardson, 1987). We know of no published reports

290

Howard and Franklin

concerning the effects of suppression on auditory rhyme judgments; Wilding (personal communication) in an unpublished experiment found no effect. Baddeley and Lewis (1981) found no effect of suppression on subjects' ability to perform homophone judgments with pairs of nonwords, or in judging whether one real word and one nonword were homophones. Richardson (1987) replicated this latter result. Besner et al. (1981, Experiment 4) and Brown (1987, Experiment 2) with homophone judgments on pairs of words or pairs of nonwords found no effect of suppression on reaction times, but did find a small and reliable effect of suppression in the error rates. There are three experiments that involve pseudohomophone detection. Besner et al. (1981, Experiment 5) found clear effects of suppression on error rates, but when, in Experiment 6, they instructed their subjects to suppress at a slower rate, they found no significant effect. Baddeley and Lewis (1981) found no statistically significant effect of suppression in this task. There are two reasons for treating this result with suspicion: First, in their experiment there was an increase in the mean reaction time of 140 msec and error rates from 14.4% to 18.5% under suppression - a difference that did not reach statistical significance. Effects of this magnitude are typical in experiments on suppression and rhyme judgments; Wilding and White (1985, Experiment 1), for instance, found an average increase in reaction time (RT) of 136 msec under suppression. Second, the mean RTs of their subjects were around 2 sec - which is approximately 900 msec slower than Besner et al.'s subjects in a corresponding task (Experiment 5). With such slow responses, it is hard to exclude the possibility that subjects manage to timeshare between the tasks. Baddeley and Lewis's result is clearly not strong evidence that pseudohomophone detection is not affected by suppression.1 The experiments by Waters, Komoda, and Arbuckle (1985) support the view that suppression interferes with lexical access from pseudohomophones. They found that although suppression had no effect on comprehension of text of real written words, performance declined under suppression when the text was rewritten into pseudohomophones. For two of these tasks the existing literature gives clear answers. Suppression impairs rhyme judgments but not homophone judgments with visually presented words. With the effect of suppression on pseudohomophone detection the evidence is equivocal; real effects of articulatory suppression cannot be excluded. In our paper describing MK's written word comprehension, we argued that pseudohomophone detection must involve generating a phonological code from the written word by means of a sublexical routine, and then using this to access an entry in an auditory input lexican (Howard & Franklin, 1987). This process then involves conversion of an output phonological code into an input code suitable for lexical access. One possibility is that articulatory suppression with normal subjects interferes with this process, which we (inelegantly) called "phonological-to-auditory conversion/' 2 Under this hypothesis, the evidence

Memory without rehearsal

291

from normal subjects suggests that an output phonological representation is sufficient for judging homophony, but that access to an auditory input code is required for rhyme judgments. We should therefore predict that, in a patient who like MK is impaired in phonological-to-auditory conversion, rhyme judgments on written words will be impaired while homophone judgments will be unaffected.

12.2. Case report: patient MK Performance on "standard" tests shows that MK, who suffered a left hemisphere stroke in 1982 at the age of 65, is a "Wernicke's aphasic"; thus his most conspicuous difficulty is in comprehension of spoken language, while his speech is fluent with some grammatical structure. His difficulties in comprehension, production, and reproduction of written and spoken words can be much more precisely defined, and we have documented them at length elsewhere (Howard & Franklin, 1987, 1988). Here we simply summarize: Spoken word comprehension is poor. Comprehension is better for high- than low-imageability words, and better for long words than short ones; word frequency has no effect. Written word comprehension is much less impaired than comprehension of spoken words. MK understands high-imageability words better than low-imageability words, but word length, word frequency, and spelling-to-sound regularity have no effect on comprehension. Oral word reading is better for words that are regular rather than irregular in their spelling-tosound correspondences; this effect is only found with low-imageability words. Nonword reading is good, and reading performance is unaffected by syntactic category, word frequency, or the presence of a morphological suffix. Many of MK's errors in reading aloud with "irregular" words reflect his use of orthographic information; e.g., move —• "/rnsov/," mortgage - • "/motgeicfe/." Oral word repetition has very different characteristics from reading. MK repeats long words better than short ones, high-imageability words better than low-imageability ones, and monomorphemic words better than words with a suffix. Syntactic category has no effect when other variables are properly controlled, nor is there an effect of word frequency. Incorrect attempts at word repetition include semantic errors {girl - • "lady," learn - • "know"), phonologically related real words {missile —• "whistle," trim —• "swim"), and apparently unrelated responses (e.g., push - • "pillow"). When the stimuli have suffixes, MK makes frequent morphological errors involving omission or substitution of the suffix. He fails utterly in repetition of nonwords. Writing to dictation has almost identical characteristics and levels of accuracy to oral word repetition. The only way in which it differs is that word length has no effect on overall accuracy, but this turns out to be because MK has a general tendency to make spelling errors with longer words in all tasks that require written output. Written and spoken naming is only just below the level of normal subjects in accuracy, but semantic errors are common in both tasks. The only difference is that in written naming there is a significant disadvantage for longer words that is not apparent with spoken output. Delayed copying (i.e., copying words onto the other side of a card) is accurate with both real words and nonwords, but there is, as with other tasks involving written output, a small but significant disadvantage for longer words.

292

Howard and Franklin

This pattern of impairment is clearly somewhat complex, and we will not provide any detailed interpretations here. We will, however, summarize our arguments for three different information-processing impairments that will have implications for the predictions we can make about how MK will perform in tests involving short-term memory and/or phonological recoding, and then briefly describe MK's comprehension of spoken and written sentences. 12.2.X. A deficit in converting output phonological representations into input phonology This claim is substantiated in detail in Howard and Franklin (1987). There are a number of lines of evidence. First, in a number of tasks on which phonological recoding of written words might hinder performance, MK shows no evidence for any effects of phonological recoding: 1. In visual lexical decision there is no difference between real words with regular and irregular spelling-to-sound correspondences, or between pseudohomophones and control nonwords in either error rate or latency of response. 2. MK's accuracy of comprehension is as good for words with regular as irregular spelling-tosound correspondence, although he is much less likely to pronounce the irregular words correctly. 3. MK never defines a word with a homophone as that homophone; thus he would never say that bare meant "a large, fierce animal."

Second, in tasks that require phonological recoding MK shows no evidence that he can do so: 1. When asked to choose which of two written nonwords would sound like a real word when pronounced (e.g., stawn - stawk, floo - froo) MK scored 24/39, which was not significantly better than chance. 2. When asked to give the meaning a written pseudohomophone (e.g., bote, tode, Hal...) would have, when pronounced, MK performed extremely poorly (38% correct). He was, however, much better at pronouncing them correctly [83%), and at defining them when the corresponding real words were presented in written form (89%). MK's ability to pronounce these pseudohomophones correctly was related to their length, while his ability to define them was independent of their length, but depended on the orthographic similarity between the pseudohomophone and the corresponding real word. This strongly implies that his ability to understand these pseudohomophones depends not on phonological recoding but instead on "approximate visual access" (cf. Saffran & Marin, 1977). His errors in this task support this view: In definition his errors are definitions of visually similar words (e.g., phite —• "pale" [i.e., white]), whereas his errors in reading seem to reflect a (mildly) defective sublexical reading routine (e.g., trete —> "/tret/").

MK's performance in pseudohomophone definition allows us to identify with some precision the point at which his performance is breaking down. He performs relatively accurately in defining the words when auditorily presented, and he performs relatively accurately in generating the appropriate output phonology from the written pseudohomophone. Thus his difficulty lies in converting output phonological representations into the input phonological form that can be used for semantic access.

Memory without rehearsal

293

12.2.2. A deficit in converting input phonology into output phonology MK cannot repeat spoken nonwords. This is not due to any particular difficulty in perceiving them: In minimal pair judgments with nonwords he scores 94% correct. His difficulty in nonword repetition cannot be attributed to a difficulty in generating nonword phonology; in reading short nonwords (three to four phonemes) aloud he was 88% correct. The problem seems to lie in the process of converting nonlexical auditory input codes into output phonological codes. Many accounts of lexical organization postulate two different lexical routines that might be used for word repetition. One involves "direct" repetition without semantic mediation - entries in an input lexicon activate the corresponding entries in an output lexicon (see, e.g., Morton & Patterson, 1980). The second involves recognizing the spoken word and then retrieving a representation of the word's meaning from a central store; this meaning representation is used to access an entry in the output lexicon. By comparison with his nonword repetition, MK achieves some success with real words. There is strong evidence that on at least some occasions he uses the routine that involves semantic mediation. First, both word repetition and spoken word comprehension are better for high-imageability than low-imageability words. Second, words that are correctly repeated are a subset of those that are correctly comprehended. And third, MK makes semantic errors in oral word repetition. Whether a "direct route" is contributing to word repetition is much harder to establish, simply because little is known about the characteristics of the routine. By analogy with the direct route for reading one might suggest that, in normal people, the direct route is available for all familiar words (cf. Funnell, 1983), and when it is defective low-frequency words will be preferentially affected (cf. Bub, Cancelliere, & Kertesz, 1985). It is clear that, because MK's repetition of real words is defective, a direct route cannot be available for all familiar words. Oral word repetition is unaffected by word frequency; thus MK is unlikely to be using a "direct" repetition route that works only with high-frequency words. MK has no sublexical routine available for repetition, and if a direct route is contributing at all to repetition of familiar words it is very impaired. It is clear that MK does use a semantically mediated routine that also is defective: it operates most effectively with concrete words.

12.2.3. A deficit in the auditory input lexicon In a number of different tasks MK misidentifies auditorily presented words as other phonologically related words. On 40% of trials he will accept a phonologically related real word or nonword as the appropriate name for a picture; thus he might judge either sea or "/gi/" as the correct name of a knee. In defining auditorily presented words he often gives a definition of another similar word; thus he defines cult as "a horse"

294

Howard and Franklin

(i.e., colt). In word repetition he produces another phonologically similar real word (e.g., missile -» "whistle"). He performs badly in lexical decision with auditorily presented words; his rate of misses is greatest with low-imageability words, which suggests that with highimageability words his relatively intact semantic representations may be involved in lexical decision. Both repetition and comprehension are worse for short words (which have many phonologically similar neighbours) than long words (which have very few). These difficulties do not appear to be due to peripheral problems in spoken word identification. MK's hearing is well within the normal range for his age, and in minimal pair judgments with either real words or nonwords his performance is 89—97% correct across different tasks. MK's problem therefore seems to lie in a loss of distinctiveness of representations in the auditory input lexicon. Thus MK appears to lack the processes necessary to convert output phonology to input phonological representations, and the conversion of input phonology to output phonology is possible only for real words and (usually or always) involves semantic mediation. If rehearsal involves the ability to cycle information between input and output phonological representations MK should behave as a patient who cannot rehearse. Further, MK's auditory input lexicon is defective; thus his PSTS may have limited support from lexical representations.

12.2.4. Sentence reproduction Sentence reproduction

MK was presented with a set of simple sentences for oral repetition, and for copying when the written sentence was removed before he could make his written response. Both tasks require within-modality reproduction of sentences in which stimulus and response are successive. There were 20 sentences varying from two to five words in length. In oral repetition MK reproduced 2/20 sentences correctly (/ combed my hair, and What are they doing!) despite his very poor repetition of single function words. Of the individual words 32/81 were repeated correctly. In successive copying his performance was much better: 8/20 sentences correct and 60/81 lexical items.

Sentence comprehension

To assess whether the same advantage for written sentences can be found in sentence comprehension, we used Bishop's (1982) "Test for the Reception of Grammar" (TROG). This is a sentence-picture matching test for the comprehension of a variety of sentence

Memory without rehearsal

295

Table 12.1. Tests of sentence comprehension a. Bishop's (1982) TROG (correct responses; chance = .25) Auditory presentation

43/80

Visual presentation

43/80

b. Byng's (1986) test for comprehension of locative sentences (correct responses; chance = .50) Auditory 19/40 23/40

Reverse role distractor Spatial distractor

Visual 27/40 24/40

c. Jones's (1984) test for comprehension of reversible verbs (n = 30)

Correct response Reverse role distractor Lexical distractor

Auditory

Visual

15 10 5

15 12 3

d. Judgments of visually presented sentences with homophones Sentence type

Example

N

Proportion correct

Correct sentences Homophone error Control real word error Pseudohomophone error Control nonword error

A pear is a fruit. Children pay half fair. She paid the bus fire. He had a red noaze. She blew her neuze.

120 40 40 20 20

0.98 0.40 0.37 1.00 1.00

types of increasing complexity using distractors that should be sensitive to impairments in grammatical processing. The results are shown at the top of Table 12.1. Performance on the TROG is equal for spoken and written sentences, and is extremely poor. A number of reports of patients with short-term memory disorders have found that sentence comprehension was especially poor with sentences where information about the constituent order was critical for correct interpretation (e.g., Saffran & Marin, 1975; Caramazza, Basili, Koller, & Berndt, 1981). We used two tests to assess this: Byng's (1986) test for comprehension of locative sentences examines comprehension of simple locatives (e.g., The hook is above the switch) in two sets of conditions. With the "reversed role distractors" the patient has to choose between a picture of a switch above a hook and a hook above a switch; for correct performance in this condition information about constituent order is essential. In the "spatial distractor" condition the patient has to choose between the correct picture and one portraying a different spatial relationship (e.g., a switch beside a hook). Correct performance in this second condition is possible without any information about constituent order.

296

Howard and Franklin

Jones's (1984) test assesses comprehension of reversible active declarative sentences. The subject has to match a sentence (e.g., The vicar shoots the doctor) to a choice of three pictures; one shows the correct action, the second the reverse action (The doctor shoots the vicar), and the third depicts a different relationship (The vicar weighs the doctor). A selective difficulty with information about constituent order should leave the patient unable to choose between the correct picture and the reverse role distractor, but able to reject the "lexical" distractor. The test has two equivalent forms, each with 30 items; Set A was presented visually and Set B auditorily. The results of these three tests are shown in the centre of Table 12.1. In Byng's test, there is no significant difference between the two types of distractors with either modality, and there are no significant differences between auditory and visual presentation (McNemar z = 1.52, n.s.). The only score that deviates significantly from chance is with visual presentation and reverse role distractors, and even here, MK is only just above the 5% level. Clearly sentence comprehension is very severely impaired, but the difficulty is not confined to comprehension of sentences where order information is required. In Jones's test, MK scores somewhat better than chance and equally well with both modalities of presentation. Combining across modalities, he is significantly more likely to choose a "reverse role" than a lexical distractor (binomial test, p < .01), but he does choose the lexical distractors on a substantial proportion of occasions, which suggests that this difficulty with these sentences is not solely one of knowledge of constituent order. His performance levels in all these tests of sentence comprehension are extraordinarily poor. In tests of comprehension of single written nouns, MK scores well; in general his performance is at a level around the lower limits of the normal population. As we shall show, his memory span is limited, but it should, on any kind of account, be sufficient for comprehension of short sentences. Yet on a number of tests he scarcely performs above a chance level. It is possible that MK's difficulty in these tests reflects a lexical difficulty with the prepositions and verbs whose comprehension is necessary for correct interpretation of the sentences. We therefore presented him with modified versions of Byng's and Jones's tests where only a single word had to be matched to the appropriate picture. For Byng's test we presented him with the pairs of pictures from the spatial distractor condition together with a single preposition; thus, for example, he might have a picture of a switch beside a hook and a picture of a switch above a hook and the single word above. With spoken prepositions he scored 27/40 and in written form, 31/40. His performance is no better than in the sentence-picture matching condition; it is likely, therefore, that a lexical difficulty in comprehension of single prepositions is responsible for his poor performance in Byng's test of phrase comprehension. With Jones's test we presented a single verb, and two pictures, one involving the

Memory without rehearsal

297

correct verb and the other Jones's original lexical distractor. Again, performance with single verbs was very poor, although significantly above chance: 22/30 with spoken and 23/30 with written verbs. Clearly a lexical difficulty contributes substantially to his poor performance in this task.

Comprehension of written sentences with

homophones

In judging whether written sentences are well formed, normal subjects make more errors with sentences that are incorrect because of substitution of a real word homophone (e.g., The king is in his thrown room) than because of substitution of a control real word (The king is in his thorns room), and this effect disappears under conditions of articulatory suppression (Coltheart, Avons, & Trollope, 1988). If MK is not using any phonological recoding in understanding written sentences, then he should be no more likely to accept homophone errors than the control sentences. MK was tested with a set of sentences designed by Veronika Coltheart for use with 8- to 11-year-old children. The results are shown at the bottom of Table 12.1. Overall, his performance was 79% correct - a level comparable with Coltheart, Laxon, Rickard, and Elton's (1988) 8-year-old subjects. He rejected correctly all sentences with nonwords. With real-word sentences his performance was very much better than chance, but still very poor; he accepted 62% of all incorrect sentences composed entirely of words, but showed no tendency to accept more homophone errors than the controls. This result supports the view that phonological recoding plays no role in MK's comprehension of written sentences.

Summary of performance in sentence comprehension tasks

MK's sentence comprehension is very severely limited. It is impaired to a surprising extent when compared to the relative mildness of his problems in single-noun comprehension. We were able to demonstrate that MK's sentence comprehension problem is not more marked where constituent order is necessary for comprehension, which shows that his difficulty cannot be due to a failure to access a short-term store specialized for holding order information. Like normal subjects under articulatory suppression, he has no more difficulty in rejecting sentences made unacceptable by substitution of a homophone than control sentences. This confirms that phonological recoding plays no role in MK's comprehension of written material. The major source of limitation on MK's sentence comprehension appears to be a lexical difficulty with prepositions, verbs, and probably with other words from closed

298

Howard and Franklin

grammatical categories whose interpretation is necessary for the correct analysis of sentence structure. From this perspective, MK's difficulties in sentence comprehension are a consequence of his general impairment in the comprehension of abstract words.

12.3. MK: list memory tasks If MK does not have the processing systems needed for rehearsal, we would predict that he should behave in list recall tasks like a normal subject whose ability to rehearse is experimentally disrupted by articulatory suppression. Thus we predict that: 1. List performance will be affected by phonological similarity with auditory but not visual presentation. 2. Stimulus word length should have no effect with visual or auditory presentation. 3. If the PSTS, rather than the process of articulatory rehearsal, is necessary for maintenance of order information, MK should have no particular tendency to make order errors with auditory list presentation.

We have argued that MK's PSTS would have little support from his defective lexical system. Thus we predict that, with auditory presentation, memory span should be no greater for real words than pronounceable nonwords.

12.3.1. Repetition span With auditory presentation, MK repeats only about 40% of single concrete real words. He is unreliable in repeating digits and never attempts to repeat more than one item from a list of two spoken digits. With visual presentation performance is much better; he manages 3/3 with two- and three-digit lists, 2/3 with four-digit, and 1/3 with both five- and six-digit. Thus auditory repetition span is severely restricted, but because single-word repetition is also severely impaired, this will not be a reliable guide to his short-term memory capacity.

12.3.2. Pointing span With a card with the digits 1-9 on it, MK pointed correctly to 8/9 single digits, but only 1/10 with two-digit lists. His picture-pointing span was assessed with nine pictures of common objects, which he could both point to and name correctly when they were presented as single items. The pictures were removed during list presentation, and their positions were varied from trial to trial, to prevent him from using any spatial coding in the task. With auditory presentation he managed 2/4 with two- and 0/4 with three-word lists. Performance with visual presentation was again better: 3/4, two-word; 2/4, three-word; and 0/2, four-word lists.

Memory without rehearsal

299

Table 12.2. Auditory matching span: the effects of word type and a comparison of the effects of changing item or order in nonmatching lists List lengths 1 2 3 4 5

Nonwords 20/20 20/20 28/30 17/20 -

Concrete nouns

Digits

20/20 19/20 29/30 34/40 39/50

20/20 20/20 2S/30 40/40 40/50

Digits" -

55/60 28/40

"The final column presents digits where nonmatching lists involved exchange of adjacent items. All other nonmatching lists involve item substitution.

MK's pointing span is also clearly impaired. This task, however, requires word comprehension, and his understanding of spoken words is impaired; pointing span may therefore underestimate his memory capacity.

12.3.3. Matching span: effects of word type and a test for selective disruption of order information List matching, a technique used for assessing STM capacity in patients by Allport (1984), requires neither overt list reproduction nor any necessary lexical processing. In these tasks MK was presented with a list at the rate of about two items per second, followed, after a pause of around half a second, by a second list; his task was simply to say yes if he judged the lists identical and no if he did not. These lists were presented over a period of several weeks; lists were blocked by length and stimulus type. To assess for the effects of word type MK was presented with lists of single-syllable, simple nonwords, single-syllable, common concrete nouns, and digits. In all these lists nonmatching pairs were created by substitution of an incorrect item (e.g., ball ship cat - ball ship pen). Half the items in each list matched, and there were equal numbers of substitutions at each position in the list. To assess whether order information is particularly affected we included two lists of digits where "different" items were created by misordering of adjacent items rather than substitution of one of them. The results are shown in Table 12.2. First, as predicted, there is clearly no effect of material type. Matching span is as good with nonwords as with concrete nouns or digits. Thus performance does not benefit from lexical processes. Second, memory capacity as assessed by matching span is considerably better than results from pointing span tasks might indicate; with digits, performance is perfect even with four-item lists. Third, there is no significant difference in level of performance between digit lists where items are permuted and those where items are substituted (length 5, Fisher exact test,

300

Howard and Franklin Auditory 1-

0.875 -

o

r I

0.75-

o Q.

* Dissimilar

0.625-

* Similar

0.5-

Visual

0.875Dissimilar

sc o o

I

0.75-

s

Similar

0.625-

0.5-

T 4 List length

Figure 12.1. The effects of phonological similarity on matching span, with auditory and visual presentation.

Memory without rehearsal

301

z = 0.85, p = 0.20). This indicates that order information is not especially vulnerable in this task.

12.3.4. Matching span: the effects of phonological similarity MK was presented with lists for matching with both auditory and visual presentation. Half the items in each list were made up of a random selection from a set of seven phonologically similar letters (B D C T V G F) and half from dissimilar letters {] Y S H R Z L). The nonmatching pairs that made up half the items of each type in each list were constructed by switching adjacent letters; each pair of letters was switched an equal number of times. The lists, which were blocked by length and modality of presentation, were presented over a period of several weeks. The number of items of each kind was 40 at Length 3; 60, Length 4; 80, Length 5; and 100, Length 6. The results are summarized in Figure 12.1. With auditory presentation there is a clear effect of phonological similarity (#2(1) = 5.81, p < 0.05), and accuracy drops off with increasing length (Jonckheere trend test z = 2.02, p < 0.05). With visual presentation there is no effect of phonological similarity (/2(1) = 0.29, n.s.), but there is a significant effect of list length (z = 2.52, p < 0.01). Considering only lists of lengths 3-5, performance is better with visual than auditory presentation with both phonologically similar and dissimilar (#2(1) = 28.14, p < 0.001, and x2 = 5.43, p < 0.05, respectively). MK's behavior in this task is interesting. With visual presentation he rehearses the words aloud; with auditory presentation he does not attempt to do so (presumably because he cannot). Thus the existence of a phonological similarity effect is entirely unrelated to the ability to use spoken rehearsal.

12.3.5. Matching span: the effects of word length - Experiment 1 The effects of word length were evaluated using lists of exactly the same kind using the same design. Lists were made up of monosyllables (whale, harp, smile, knife, plant, crown, steak) or trisyllabic words (crocodile, orchestra, hospital, telephone, photograph,

tangerine, ambulance). The words were matched for frequency and imageability. The results are shown in Figure 12.2. The word length effect falls far short of significance with both auditory and visual presentation (# 2 (l) = 0.44, n.s., and #2(1) = 0.49, n.s., respectively). Accuracy is a function of list length with both modes of presentation (Jonckheere trend test z = 2.23, p < 0.01, and z = 2.96, p < 0.01, respectively). Considering lengths 3—5, performance with visual presentation is significantly better than with auditory presentation (long words: (%2(1) = 9.75, p < 0.01; short words: #2(1) = 9.00, p < 0.01). In spoken-word comprehension, MK performs better with long words than short ones. Note, therefore, that the result of this experiment - failure to find a word

302

Howard and Franklin Auditory

0.875 -

o

0.75-

oa o 0.625-

0.5

Visual

0.875 -

§

0.75 * Short

oa o 0.625-

0.5 List length

Figure 12.2. The effects of word length on matching span, with auditory and visual presentation.

Memory without rehearsal

303

length effect with auditory presentation - supports our earlier conclusion that MK's performance in list memory tasks.with auditory presentation does not depend on lexical processes. As in the previous task MK rehearses lists aloud only with visual presentation. It is clear, therefore, that the word length effect with visual presentation is not a simple result of articulatory rehearsal.

12.3.6. Matching span: the effects of word length - Experiment 2 From one perspective, the failure to find a length effect with auditory presentation in the previous experiment is rather surprising. If, as we have argued, MK is not using lexical information in matching span, we might expect his short-term capacity to be time limited, and so larger for single-syllable words. And yet we have failed to find a word length effect. The previous experiment involved only seven words of each length; it is possible that MK developed such familiarity with these small sets that length effects disappeared. We therefore ran the same matching span experiment with auditory presentation, where the stimuli were selected at random from sets of 100 one-syllable and 100 three-syllable words; the words were matched for imageability and frequency. MK was given 120 lists of Length 4 for matching; half the lists were made up of trisyllabic words and half of monosyllables. Nonmatching lists were generated by switching adjacent items. The results are straightforward: 38/60 correct judgments with lists of monosyllables, and 41/60 correct with lists of trisyllabic words. Selecting items from a larger pool confirms the result of the previous experiment: MK's auditory matching span does not depend on the syllable length of the words involved, and this result is not an artefact of using small stimulus sets.

12.3.7. Probe memory: the effects of word imageability MK has much better comprehension of high-imageability than low-imageability words. This experiment tests whether there is any semantic influence on his performance on a probe memory task. Lists made up of bisyllabic words of either high imageability (e.g., jacket, marble, salad) or low imageability (e.g., patent, treaty, topic) were presented auditorily or visually, at the rate of two items per second; immediately at the end of the list MK was presented with a probe, and had to decide whether it had been present in the list. In 48 trials the probe was not present; each position in the list was probed on 10 occasions. The stimulus items were drawn randomly from pools of 10 high-imageability and 10 low-imageability words matched for frequency and length. With auditory presentation, MK scored 54/64 (84%) correct with concrete and

304

Howard and Franklin

53/64 (83>%) with abstract word lists. With written presentation he scored 52/64 (81%) with concrete and 49/64 {77%) with abstract words. There is no suggestion of an effect of imageability with either mode of presentation, nor is there a significant effect of modality of presentation (McNemar z = 0.91; n.s.). Thus, despite a strong effect of word imageability on written- and spoken-word comprehension, MK's memory performance is independent of this variable. This supports the claim from earlier experiments that MK does not use lexical or semantic levels of representation to support his performance on list memory tasks.

12.3.8. Visual list presentation: the effects of vocal rehearsal We have shown that, in the matching span paradigm, MK's memory of visually presented lists is not affected by phonological similarity. Because he adopted a strategy of rehearsing the lists aloud, we argued that phonological similarity effects were dissociated from spoken rehearsal. In this experiment we examine this issue experimentally by comparing report of lists of letters MK reads aloud during presentation, with report of lists where he has to remain silent during presentation. Normal subjects show phonological similarity effects whether they say lists aloud or are presented with them silently, but when they say the lists during presentation, performance is improved on the final item (e.g., Greene & Crowder, 1986). This has been explained as an effect of an auditory input store (Crowder, 1983). Since we have claimed that MK cannot use output phonology that he generates to access auditory input processes, we predict that he will show neither a phonological similarity effect nor recency even when he says the list aloud at presentation. MK was given 20 lists of three, four, and five items in length. Half the lists at each length were made up of phonologically similar letters (from the set C G B T P D V) and half of dissimilar letters (from the set H L S Z R Y ]). For half of the lists of each length MK was instructed to read the letters silently to himself from a card, turn it over, and say the letters aloud. For the other half he was told to say the letters aloud, turn the card, and report the letters. As we would expect, MK's performance depended on list length; overall, he was correct on 14/20 lists of Length 3; 10/20, Length 4; and 2/20, Length 5. There was, however, no effect of reading the lists aloud (15/30 lists, 82/120 items) compared to silent presentation (11/30 lists, 85/120 items). Reading the words aloud did not result in recency: With five-item lists read aloud, for example, MK recalled 4/10 of letters in Position 4 and 2/10 final items. There was also no disadvantage for phonologically dissimilar (12/30 lists, 74/120 items) compared to phonologically similar lists (14/30 lists, 93/120 items). In fact, there is a significant advantage for the similar lists (on items correct, Fisher exact z = 2.52; p = 0.012 2-tailed)! This is presumably because

Memory without rehearsal

305

phonological similarity provides a cue for recall by limiting the range of possible responses. The results of this experiment confirm and extend the results we found with the matching span paradigm: Short-term memory of visual lists is not affected by phonological similarity, and the failure to find any effect of saying the items aloud on list presentation confirms that MK cannot access input auditory processes from output phonology that he generates himself.

12.3.9. Recall of pseudohomophones and nonwords: visual list presentation Besner and Davelaar (1982) demonstrated that with normal subjects, lists of visually presented pseudohomophones are recalled more accurately than lists of control nonwords, and this effect is still found under conditions of concurrent articulation. The advantage for pseudohomophones does not therefore appear to depend on articulatory rehearsal. So, if MK is unable to rehearse, we should expect him to show a pseudohomophone advantage in visual list recall. To test this MK was presented lists of three or four items; these could consist of control nonwords (e.g., daik prane powce), pseudohomophones (wotch kace bote), or real words (lake train bruise). The items were matched by list for w-ness (i.e., the number of real words that can be made by changing a single letter in the item). There were 10 lists of each type of each length; items of each type were randomly intermixed, but presentation was blocked by list length. Lists were presented under two conditions. In the first session, MK read all the items on a card aloud, turned it over, and then attempted to recall the list. This condition allows us to check whether a difficulty in generating the correct phonology for the stimuli may be contributing to a difficulty in recall. In the second condition, tested 3 months later, MK read the list silently, turned over the card, and then attempted to recall it. When lists were read aloud on presentation, MK made only two reading errors on the 210 items. He is therefore able to generate the correct phonology from these written stimuli. His recall of these lists is shown in Table 12.3. As in the previous experiment there is no difference between the conditions of silent presentation and reading the lists aloud. There is, however, an advantage for pseudohomophones over control nonwords (by items, combining over list length and conditions, Fisher exact z = 3.22, p < 0.001) and for real words over pseudohomophones (z = 2.80, p = 0.002). Like normal subjects under conditions of articulatory suppression, MK is better at recall of visually presented pseudohomophones than control nonwords. This effect is apparent both under silent presentation and when lists are spoken aloud, and is not attributable to any difficulty in generating the appropriate phonology for the items.

306

Howard and Franklin Table 12.3. Spoken recall of visually presented lists of nonwords, pseudohomophones, and real words when spoken aloud at presentation and when presented silently (10 lists of each type at each length under each condition; proportion lists correct; items correct in brackets)

List length

Nonwords

Pseudohomophones

Real words

Total

.0(.53) .15 (.59)

.6(.9O) .1(.73) 35 (.SO)

.30(.64) .03 (.54) .17(.59)

.2 (.67) .1(.55) .15 (.60)

.4(.83) .0(.63) •2(.71)

.23 (.69) .03 (.49) .13 (.58)

Spoken aloud on presentation

3 4

Total

.0(37) .0(38) .0(37)

3 (.67)

Silent presentation

3 4

Total

•K.57) .O(.3O) .05 (.41)

12.3.10. Summary of the list memory tasks All four of our predictions were confirmed. There is no evidence of any lexical or semantic influence on memory performance: Matching span is as good with nonwords as real words, and probe memory performance is as good with low-imageability as high-imageability words. Like normal subjects under articulatory suppression, MK shows no effects of word length with visual or auditory presentation, and phonological similarity effects are found only with spoken word presentation, but recall of visually presented pseudohomophones is better than control nonwords. There was no evidence that, with auditory presentation, information about item order is more vulnerable than information about item identity. The most intriguing result is perhaps the failure to find a word length effect with visual presentation, even though MK adopted a strategy of articulatory rehearsal; and, even when required to say a list aloud on presentation, there are no phonological similarity effects. These results strongly imply that neither the word length effect nor the phonological similarity effect is a simple consequence of articulatory rehearsal. As with many patients who have short-term memory deficits, MK's performance is consistently better with visual rather than auditory list presentation.

12.4. MK: Judgment of rhyme and homophony Normal subjects' ability to decide if pairs of written words rhyme is impaired by articulatory suppression, while judgment of whether pairs of words or nonwords are homophonous is unaffected. If MK is behaving like a normal subject under suppression we should predict that (a) rhyme judgments should be poor with written words, and (b) homophone judgments should be good with written words and nonwords.

Memory without rehearsal

307

Table 12.4. Homophone judgments: proportions correct (50 judgments of each type) Normal controls Example Regular words Irregular words Nonwords

Lax Bear Max

MK Lacks .94 Bare .90 Nacks .94

Mean

Range

.97 .96 .96

.94-1.0 .92-1.0 .90-1.0

Source: Stimuli are from Coltheart (1980); data on normal controls are from Baddeley and Wilson (1985).

Table 12.5. Rhyme judgments: proportions correct (15 judgments of each type) MK

Eye and ear rhymes Ear rhymes Eye rhymes Nonrhymes

Example

Auditory

Visual

Cream Come Foot Shout

1.0 .S7 1.0 1.0

.53 .53 .33 .80

.97

.55

Total

12A.I.

Team Sum Boot Loot

Homophone judgments

We used the homophone judgments from Coltheart (1980). These are of three kinds: pairs of regular real words (e.g., blew, blue), pairs of real words where at least one is irregular (e.g., bear, bare), and pair of nonwords (e.g., cobe, koab). There are 50 pairs of each type of which half are homophones; orthographic similarity is equal for the homophonic and nonhomophonic pairs. The results are presented in Table 12.4, together with data from 14 normal controls (from Baddeley & Wilson, 1985). MK's performance is within the normal range for regular words and nonwords, and only just outside for irregular words.

12.4.2. Rhyme judgments We used a set of 60 pairs of words. There are 15 items of each of four types: ear and eye rhymes, ear rhymes, eye rhymes, and nonrhymes (see Table 12.5). They were presented to MK in auditory and visual form. Before presentation he was given some practice and told that he could not rely on orthographic similarity for the judgment. With auditory presentation MK performs well; he clearly has no difficulty with the concept of rhyme. With visual presentation, however, his performance is no better than

308

Howard and Franklin

chance. MK's inability to judge if the pairs of written words rhyme cannot be attributed to a difficulty in generating the output phonology for these words: On a separate occasion he was asked to read the 120 words aloud; he made only seven errors.

12.4.3. Auditory rhyme judgments: the effects of delay and interference The results in the previous section suggest that MK has a phonological input store with rather restricted capacity, which he cannot "refresh" by articulatory rehearsal. Baddeley and Lewis (1982) suggest that, with normal subjects, the contents of a phonological store decay in 1-2 sec when rehearsal is suppressed (cf. Schweikert & Boruff, 1986). We therefore investigated MK's performance in rhyme judgments when either interference or a delay is interspersed between presentation of each half of the pair of words. MK performed rhyme judgments on a set of 60 items (a revised version of the previous test) auditorily presented under three conditions: (a) immediate—the two words were presented immediately after each other; (b) silent delay-the two words were presented at an interval of 2 sec; (c) interference - the experimenter counted aloud from 1 to 3 during the delay. The items were presented in the same order in three sessions a week apart; one third of the items in each condition occurred in each session. Each condition was preceded by four practice items. In the immediate condition MK made 57/60 correct judgments, 51/60 with a silent delay, and 54/60 with interference. There was no significant difference in accuracy across the three conditions (Cochran Q[2] = 3.86, n.s.). It is clear that MK can hold a single item for at least 2 sec, and this ability is not subject to interference from irrelevant material.

12.4.4. Summary of rhyme and homophone judgment tasks With written word presentation, MK is normal in judging whether pairs of words and nonwords are homophones, and he performed at chance in judging whether pairs of words rhyme. Since his ability to access output phonology from written words is good, this pattern of performance is strong evidence that access to output phonology is sufficient for homophone judgments but not for rhyme judgments. MK can access input phonological codes from auditory input, and he can perform auditory rhyme judgments. Access to output phonology is therefore not sufficient for performing rhyme judgments on written words.

12.5. Discussion Many claims have been made about the role of rehearsal in tasks of short-term memory and language comprehension. Following Monsell (1987) we suggested in the

309

Memory without rehearsal Visual input

Auditory input

AUDITORY

INPUT

INPUT

PHONOLOGICAL STORE

j_

LEXICON

_ "ALL OTHER LEXICAL AND SEMANTIC PROCESSES

OUTPUT PHONOLOGICAL STORE

Speech output

Figure 12.3. An outline model of the components involved in short-term memory tasks. The input and output phonological stores are linked by the processes of rehearsal. For details see text. Introduction that rehearsal processes could be identified with the lexical processes used to convert input phonology into output phonology, and vice versa. On independent grounds we were able to show that MK cannot use any of these processes; in this sense, he is a patient who cannot rehearse. His performance can then be used to test claims about the role of rehearsal that are derived from the effects of experimental manipulations, such as concurrent articulation, with normal people. Figure 12.3 presents a very simplified representation of the processes involved in phonological short-term memory. The components that are impaired or unavailable to MK are shown in dashed lines. We will frame our discussion in terms of the components shown in this figure.

310

Howard and Franklin

12.5.1. List memory MK is unable to repeat strings of more than one item, and even with single words performance is much less than perfect. Therefore we tested MK's list memory in a variety of different ways that did not require a spoken list as a response. Across the tasks performance was better with visual than with auditory presentation. With auditory presentation performance was affected by phonological similarity but was no better for real words than nonwords, nor were concrete words better than abstract ones; this is consistent with the view that MK is relying on a prelexical store in which items are coded in some form where "phonological" confusions can occur. The information might, therefore, be coded in acoustic, auditory, or some more abstract phonological form. We think it unlikely that MK is relying solely on a simple acoustic store like Crowder and Morton's (1969) PAS (precategorical acoustic storage; see Crowder, 1983; Greene & Crowder, 1986). This is because, in the delayed rhyme judgment task, interpolated auditory material did not affect performance, when a "suffix" of that kind should have abolished PAS. MK's auditory list memory falls off quite sharply with lists of more than three to four items. This suggests that he is relying on a "store" with limited capacity. As performance is unaffected by word length, the capacity cannot be time limited; instead, it seems to be limited by the number of items irrespective of their category or length. MK therefore is using an "isolated" short-term store, which is not supported by either rehearsal or by lexical or semantic storage. It is prelexical and coded in some "auditory" or "phonological" form. Its capacity is not time limited, but it is restricted to three to four items in length. The delayed rhyme judgment task shows that it is able to hold an item for a minimum of 2 sec, and is not subject to interference from irrelevant auditory material. It is difficult to know whether this is a characterization of a "normal" PSTS, shorn of the supporting memory systems that enhance span in normal people, or whether the system is damaged. Unlike many published reports of patients with short-term memory deficits (e.g., Caramazza, Berndt & Basili, 1983; Allport, 1984; Campbell & Butterworth, 1985), MK shows no signs of a global degradation in the whole "phonological" domain; in fact, in a number of tasks he shows remarkably good performance. Thus, he can read simple nonwords accurately (Howard & Franklin, 1987); he can do minimal pair judgments with real words and nonwords (Howard & Franklin, 1988); and he can judge whether pairs of auditorily presented words rhyme. The capacity of MK's PSTS cannot then be said to be limited by a general phonological "problem." We suspect, therefore, that MK's performance gives an accurate picture of the operation of an isolated prelexical phonological store. Although we cannot conclusively demonstrate that its capacity is not limited by damage, our experiments put a clear lower limit on the size of this prelexical store. Craik (1971) reviews several approaches

Memory without rehearsal

311

to estimating the capacity of the presemantic store and concludes that it can hold about three items; this is remarkably similar to our estimate of the size of MK's prelexical store. Schweikert and Boruff (1986) find a similar capacity for normal subjects' nonword span. With visual list presentation, MK has a slightly greater span. That there is no effect of word length or phonological similarity in written list memory confirms that rehearsal is not involved. This is a striking finding because in these tasks MK overtly uses articulatory rehearsal. This strongly suggests that in normal subjects the effects of articulatory suppression on visually presented lists reflect not interference with rehearsal itself but rather the use of phonological recoding to access the PSTS (cf. Baddeley, 1986). Articulatory suppression, then, interferes not with the ability to hold or generate an (output) phonological representation but instead with the process we have called "phonological-to-auditory conversion." In accordance with this, Klapp, Greim, and Marshburn (1981) show that normal subjects are able to assemble an articulatory programme for spoken output (stored, they suggest, in an output buffer) from visual input, under articulatory suppression. MK's memory for visually presented lists was slightly greater than with auditory presentation; it is between three and four items. Since his written word comprehension is affected by word imageability while his memory span is not, this span does not depend on semantic encoding. Evidence from Chinese speakers suggests that, without rehearsal, there is a short-term memory for visually presented items of about this size. Zhang and Simon (1985) estimated the normal span for visually presented Chinese characters as just over six. They suggest that there are two sorts of stimuli that cannot be rehearsed articulatorily; radicals, which have no known name, and characters, which have many homophones. With these stimuli they found that recall span fell to three items; Zhang and Simon suggest that this residual capacity is due to a visual memory store. With visually presented lists of English homophones (right write rite write), Crowder (1978) found normal subjects could recall four items from a seven-word list. We suspect that MK uses the same store for memory of visually presented lists. Unlike normal subjects, MK is no better at recalling visually presented lists when he says the items aloud at presentation than when he reads them silently. In normal subjects, the advantage for lists spoken aloud at presentation is mainly due to good performance on the final item; MK shows no recency in either condition. In normal subjects, the advantage from speaking the items aloud on presentation is eliminated by articulatory suppression, but not by silent mouthing (Gathercole, 1986); this suggests that an auditory memory component is responsible for the effect. MK's performance, which is like that of normal subjects under articulatory suppression, shows no evidence that auditory information can contribute to his recall of visually presented lists. With visually presented lists, MK is better at recalling pseudohomophones than control nonwords; once more this is the pattern found with normal subjects under articulatory suppression (Besner & Davelaar, 1982). This is the only task in which MK

312

Howard and Franklin

shows any difference between pseudohomophones and control nonwords; as we have described, he is unable to detect pseudohomophones, is very poor at defining them, and is unaffected by them in visual lexical decision (Howard and Franklin, 1987). In view of all the evidence we have produced showing that for MK input phonological representations play no part in processing of visually presented items, the advantage for pseudohomophones in recall cannot be due to the use of representations in an auditory input lexicon. The most plausible explanation of this effect is lexical support of phonological representations in an output buffer (cf. Saffran & Martin, this volume, chapter 6; Campbell, chapter 11). That MK behaves in list memory tasks in the way in which we would predict of a subject who cannot rehearse supports our hypothesis that the rehearsal system can be identified with the processes that allow transcoding between (auditory) input representations and (output) phonological representations.

12.5.2. Rhyme and homophone judgments In normal subjects, articulatory suppression interferes with judgments of whether or not two written words rhyme, but not with judgments of homophony. This has been used as evidence for two distinct kinds of "phonological" code (see, e.g., Besner et al., 1981; Besner & Davelaar, 1982). Bishop (1985) suggests that these may correspond to different stages in spoken word production. The evidence from this study forces a different conclusion, namely, that homophone judgments can be made on the basis of (output) phonological representations, but that rhyme judgments require access to an input (auditory) code. This conclusion is driven by the dissociation for MK between the rhyme and homophony tasks. MK, who cannot convert (output) phonology into input codes, is at normal levels in homophone judgments but at chance with rhyme judgments with written words; he is, however, good with auditory rhymes. This demonstrates that output phonological codes are sufficient for homophone judgments but not for rhyme judgments. Access to auditory input codes is necessary for rhyme judgments. This implies that in normal subjects articulatory suppression interferes with phonological-to-auditory conversion, rather than with output phonological representations themselves; this is the same conclusion that we arrived at independently in discussing the list memory tasks. We are not clear why auditory input codes should be necessary for rhyme judgments. Since the critical difference between homophony and rhyme judgments is that rhyme judgments necessarily require phonological segmentation, whereas homophone judgments do not, we would claim (with obvious circularity) that sublexical segmentation is possible only on auditory input representations. It is plausible to suggest that auditorily presented phoneme strings can be segmented; any serious

Memory without rehearsal

313

proposals for the operation of a sublexical procedure for converting input phonology into output phonology must incorporate as an early stage a segmentation process. It is even possible that the PSTS normally codes items either in a segmented form, or in the form in which segmentation is marked; this would explain why, in nonsense syllables, consonants and vowels are differentially affected by phonological similarity (Cole, 1973; Drenowski, 1980). Furthermore, much of auditory input is fragmentary due to noise and other kinds of interference; to deal with this kind of partial information the input system must be able to manipulate segmented information. These results are in accordance with Allport's (1984) and Monsell's (1987) suggestions that there are separate input and output stores. If, as Shallice and Butterworth (1977) and Allport (1984) maintain, an intact (phonological) output store is necessary for fluent, grammatically well-formed speech production, MK must have (at least) reasonable capacity at this level. That he can perform homophone judgments shows that he can hold output phonological representations for at least a single word (or nonword). His difficulty appears to lie in transferring information between the input and output stores.

12.5.3. MK's performance related to the effects of articulatory suppression in normal subjects Overall, there is remarkable similarity between MK's pattern of performance over memory, rhyme, and homophone tasks, and that of normal subjects operating under articulatory suppression. Yet it is clear that for MK there is no real interference with any levels of the articulatory output process itself; his speech is fluent, contains varied syntactic structures, and is well articulated. This has led to our proposal, for which we gave two independent arguments, that articulatory suppression in normals interferes with the process of converting output phonology into input phonological forms - or phonological-to-auditory conversion. We want to summarize how this could account for the spectrum of reported effects of articulatory suppression in normals. Most of these have already been discussed in passing, so the arguments will not be given in detail here. (a) With lists of unrelated items, suppression eliminates the word length effect with auditory presentation, but leaves the phonological similarity effect unaffected. In line with Baddeley's (1986) model of working memory, we suggest that the phonological similarity effect is due to the input phonological store, which is unaffected by suppression. The word length effect is due to rehearsal, which suppression prevents. (b) Visually presented lists show word length and phonological similarity effects; suppression during presentation and recall abolishes both the effects of word length and phonological similarity. Under normal conditions, some of the recall capacity with visually presented lists is due to phonological recoding into output phonological form,

314

Howard and Franklin

and then conversion into input phonology. The representation in the input phonological store can be maintained by rehearsal; the input phonological store is responsible for the phonological similarity effect, and rehearsal for the word length effect. Suppression interferes with phonological-to-auditory conversion, which is required for access to the input phonological store, which in turn is necessary for rehearsal to play a role. Under the circumstances of suppression, subjects do not attempt to use phonological recoding but rely instead on visual short-term memory systems. (c) Suppression does not interfere with homophone judgments or with assembling an articulatory programme from visual input. Both of these tasks require access to unsegmented output phonology from visual input; suppression has no effect on access to output phonology, so performance is unaffected. (d) Suppression causes a decrement in performance in rhyme judgments with visual presentation. We have proposed that rhyme judgments, which require comparison of pairs of segmented phonological strings, can only be done with input phonological representations. If articulatory suppression completely abolished phonological-toauditory conversion, we would expect that under suppression performance would be at chance; yet normal subjects achieve respectable (though lower) levels of accuracy when required to suppress. Our account of this is as follows: Suppression only interferes with phonological-to-auditory conversion, and it is possible to do the conversion under suppression, but the process is slower and prone to error. (e) There are variable effects across experiments of suppression on pseudohomophone detection. We argued that pseudohomophone detection requires access to input phonology. If, as we suggested, the effect of suppression will be to make the phonological-to-auditory conversion process slower and more error prone, there should be an effect of suppression. Rhyme judgments will, however, be more liable to show effects of suppression, because accurate performance will require that two strings are correctly converted into their input auditory form; pseudohomophone detection, which requires conversion of only a single item, will be relatively resistant to the effects of suppression. This predicts that the effects of suppression on pseudohomophone detection should be relatively smaller in size, and harder to detect experimentally than the effects of suppression on rhyme judgments. (f) Span for pseudohomophones is greater than for control nonwords with visual presentation; this effect persists under articulatory suppression (Besner & Davelaar, 1982). Under any theory that allows interaction between lexical and sublexical levels of representation (see, e.g., Saffran & Martin, this volume, chapter 6; Campbell, chapter 11), assembled representations in an output phonological buffer will be more durable when they correspond to lexically represented phonological forms than when they do not. This effect, which does not depend on phonological-to-auditory conversion, will be unaffected by suppression. As we argued in the Introduction, evidence from the effects of a distractor task such

Memory without rehearsal

315

as articulatory suppression is not easily interpreted. We have demonstrated, however, that the position we are proposing is consistent with the existing evidence for the effects of articulatory suppression with normal subjects. As we have shown, the form of the theory we present is related to a number of other recent theortical proposals. Most of these, however, suffer from the disadvantage that they are too vague to allow us to relate deficits in lexical processing to the components of a "short-term memory" system. The position we have taken is one that is adopted by a variety of other contributors to this book: "Short-term memory" is intimately tied to the processes of word comprehension and production.

12.5.4. Summary This study of an aphasic patient who cannot rehearse has provided strong evidence in favour of a number of different conclusions about the organization of "working" memory and the role of rehearsal. Here we simply list the most important. 1. There is a prelexical phonological short-term store, whose capacity is at least three items and which, in the absence of rehearsal or lexical coding, can hold a single item for at least 3 sec. Items in this store are coded in an "auditory" or "phonological" form. 2. The process of phonological-to-auditory conversion is necessary to access this store from visual presentation of written words, and in normal subjects this process is interfered with by articulatory suppression. 3. There is a store that does not depend on rehearsal or semantic encoding and that can hold visually presented material; its capacity is three or four items. 4. Output phonological representations are sufficient to judge whether pairs of written words are homophones. It is necessary to access input auditory representations to judge whether words rhyme.

Notes 1. Both the Baddeley and Lewis (1981) and the Besner et al. (1981) experiments involve measuring the average time to make decisions on pseudohomophone detection for lists of nonwords. This is probably not a very sensitive experimental procedure. 2. We use the term phonological-to-auditory conversion to maintain terminological consistency with Patterson's (1986) revision of Morton and Patterson's (1980) model of lexical processing. This distinguishes input and output codes by using the term auditory and phonological respectively. Whether input codes are truly auditory or in some more abstract phonological form is not directly relevant to our analysis here.

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). London: Erlbaum. Baddeley, A. D. (1986). Working memory. Oxford; Oxford University Press.

316

Howard and Franklin

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. M. Perfetti (Eds.), Interactive processes in reading (pp. 108-129). Hillsdale, NJ: Erlbaum. Baddeley, A. D., & Lewis, V. J. (1982). 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, 36, 233—252. 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., & Wilson, B. (1985). Phonological coding and short-term memory in patients without speech, Journal of Memory and Language, 24, 490—502. 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. Besner, D., Davies, ]., & Daniels, S. (1981). Reading for meaning: The effects of concurrent articulation. Quarterly Journal of Experimental Psychology, 33A, 415—437. Bishop, D. V. M. (1982). TROG Test for Reception of Grammar. Newcastle upon Tyne: Medical Research Council. Bishop, D. V. M. (1985). Spelling ability in congenital dysarthria: Evidence against articulatory coding in translating between phonemes and graphemes. Cognitive Neuropsychology, 2, 229-251. Brown, G. D. A. (1987). Phonological coding in rhyme and homophony judgement. Ada Psychologica, 65, 247-262. Bub, A., Cancelliere, A., & Kertesz, A. (1985). Whole-word and analytic translation of spelling-tosound in a non-semantic reader. In K. E. Patterson, J. C. Marshall & M. Coltheart (Eds.), Surface dyslexia: Neuropsychological and cognitive analyses of phonological reading (pp. 15-34). London: Erlbaum. Byng, S. C. (1986). Sentence processing deficits in aphasia: Investigation and remediation. Unpublished Ph.D. thesis, London University. 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. Caramazza, ^., 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. Caramazza, A., Berndt, R. S., & Basili, A. G. (1983). The selective impairment of phonological processing: A case study. Brain and Language, 18, 128-174. Cole, R. A. (1973). Different memory functions for consonants and vowels. Cognitive Psychology, 4, 39-54. Coltheart, M. (1980). Analysing acquired disorders of reading. Unpublished manuscript, Birkbeck College. Coltheart, V., Avons, S., & Trollope, J. (1988). The effects of articulatory suppression on reading for meaning. Paper presented to the London meeting of the Experimental Psychological Society, January 1988. Coltheart, V., Laxon, V., Rickard, M., & Elton, C. (1987). Phonological recoding in reading for meaning by adults and children. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 387-397. Craik, F. I. M. (1971). Primary memory. British Medical Bulletin, 27, 232-236. Crowder, R. S. (1978). Memory for phonologically uniform lists. Journal of Verbal Learning and Verbal Behavior, 17, 73-89. Crowder, R. G. (1983). The purity of auditory memory. Philosophical Transactions of the Royal Society of London, B302, 251-265.

Memory without rehearsal

317

Crowder, R. G., & Morton, J. (1969). Pre-categorical acoustic storage (PAS). Perception and Psychophysics, 5, 365-373. Drenowski, A. (1980). Memory functions for vowels and consonants: A reinterpretation of acoustic similarity effects. Journal of Verbal Learning and Verbal Behavior, 19, 176-193. Funnell, E. (1983). Phonological processes in reading: New evidence from acquired dyslexia. British Journal of Psychology, 74, 159-180. Gathercole, S. E. (1986). The modality effect and articulation. Quarterly Journal of Experimental Psychology, 38A, 461-474. Greene, R. L, & Crowder, R. G. (1986). Recency effects in delayed recall of mouthed stimuli. Memory and Cognition, 14, 355—360. Howard, D., & Franklin, S. (1987). Three ways for understanding written words, and their use in two contrasting cases of surface dyslexia. In D. A. Allport, D. MacKay, W. Prinz, & E. Scheerer (Eds.), Language perception and production: Relationships between listening, speaking, reading and writing (pp. 340-366). London; Academic Press. Howard, D., & Franklin, S. (1988). Missing the meaning? A cognitive neuropsychological analysis of single word processing in an aphasic patient. Cambridge, MA: MIT Press. Johnston, R. S., & McDermott, E. A. (1986). Suppression effects in rhyme judgement tasks. Quarterly Journal of Experimental Psychology, 38A, 111-124. Jones, E. V. (1984). Word order processing in aphasia: Effect of verb semantics. In F. C. Rose (Ed.), Advances in neurology: Vol 42, Progress in aphasiology (pp. 159-181). New York: Raven Press. Klapp, S. T., Greim, D. M., & Marshburn, E. A. (1981). Buffer storage of programmed articulation and articulatory loop: Two names for the same mechanism or two distinct components of short-term memory? In J. Long & A. D. Baddeley (Eds.), Attention and performance IX (pp. 459-472). Hillsdale, NJ: Erlbaum. Kleiman, G. M. (1975). Speech recording in reading. Journal of Verbal Learning and Verbal Behavior,

24, 323-339. Monsell, S. (1987). On the relation between lexical input and output pathways for speech. In D. A. Allport, D. MacKay, W. Prinz, & E. Scheerer (Eds.), Language perception and production: Relationships between listening, speaking, reading and writing (pp. 273-311). London: 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. London: Routledge & Kegan Paul. Murray, D. J. (1968). Articulation and acoustic confusability in short-term memory. Journal of Experimental Psychology, 78, 679-684. Patterson, K. E. (1986). Lexical but non-semantic spelling? Cognitive Neuropsychology, 3, 341— 367. Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193-198. Richardson, J. T. E. (1987). Phonology and reading: The effects of articulatory suppression upon homophony and rhyme judgements. Language and Cognitive Processes, 2, 229-244. 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. Saffran, E. M., & Marin, O. S. M. (1977). Reading without phonology: evidence from aphasia. Quarterly Journal of Experimental Psychology, 29, 515-526. Schweikert, R., & Boruff, B. (1986). Short-term memory capacity: Magic number or magic spell? Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 419-425. Shallice, T., & Butterworth, B. L. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729-735. Vallar, G., & Baddeley, A. D. (1984a). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141.

318

Howard and Franklin

Vallar, G., & Baddeley, A. D. (1984b). Fractionation of working memory. Neuropsychological evidence for a phonological short-term store. Journal of Verbal Learning and Verbal Behavior, 23, 151-162. Vallar, G., & Cappa, S. F. (1987). Articulation and verbal short-term memory: Evidence from anarthria. Cognitive Neuropsychology, 4, 55—77. Waters, G. S., Komoda, M. K., & Arbuckle, T. Y. (1985). The effect of concurrent tasks on reading: Implications for phonological recoding. Journal of Memory and Language, 24, 27-45. Wilding, ]., & White, W. (1985). Impairments of rhyme judgement by silent and overt articulatory suppression. Quarterly Journal of Experimental Psychology, 37A, 95-107. Zhang, G., & Simon, H. A. (1985). STM capacity for Chinese words and idioms: Chunking and acoustical loop hypotheses. Memory and Cognition, 13, 193—201.

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

13.1. Discontinuity between episodes The extended present and the immediate past are phenomenally discontinuous. In contrast to the present, which is immediately available, to reexperience the past requires an act of episodic recollection (Tulving, 1983). Discontinuity theories in memory, notably the ones exemplified by the short-term memory/long-term memory (STM, LTM) distinction, are supported by experiments that demonstrate discontinuities in the function that relates performance on memory tasks to time elapsed since the memoranda were presented. The present chapter introduces evidence of an analogous discontinuity in the domain of estimation of time past, and of a specific pathology causing a selective loss of one phase in remembering time past. In a paradigm in which information flow to the subject is continuous, the findings isolate a time period within which information is retrievable to an extent inversely proportional to the time elapsed since it was presented. We interpret this in terms of William James's sweeping formulation of the extended present as representing "the lingerings of the past dropping successively away, and the incomings of the future making up the loss" (James, 1890, p. 606). We take this to mean that for James, the extended present is the contents of the present experience, the "episode," sliding across time. Short-term memory is usually measured in relation to an input phase, during which stimuli are presented in sufficient amount fully to engage or even overload the cognitive system. The ability to recover this information at varying times afterward is measured (e.g., Peterson & Peterson, 1959). But in the absence of informational overload, can one infer a critical period of time, beyond which experienced events are relegated to the past? This proposed discontinuity is imposed on the flow of events by the brain itself, which "parses" experience into "episodes." Thus, responding from the extended present would be based on information that is currently available, whereas responding from the past (long-term memory) would require a retrieval or reexperiencing of a bygone episode. It follows that the duration of the extended present amounts to the duration of an episode. Up to the point in time at which one has to undertake 319

320

Kinsbourne and Hicks

episodic retrieval in order to respond on the basis of the relevant information, one remains within the same episode.

13.2. Information loss within the extended present Information present in immediate experience is not necessarily as easily accessible subsequently as at the instant it is presented. One may envisage an acuity function in receding time comparable to the acuity function for objects receding in space. As time passes within the limits of the extended present it may become increasingly difficult to retrieve information even before an act of episodic retrieval is called for. Models of this type have been offered before in relation to forgetting. Bjork and Whitten (1974) used the analogy of telegraph poles receding in space to represent the difficulty in discriminating an item in a sequential array over time. Murdock (1972,1974) likened the representation of events fed into memory to the placing of suitcases on a conveyor belt. To bring a test item into attention a backward scan along the conveyor belt is needed. In other words, just as in space an item may recede from the focus of attention while still being immediately present, so items will normally recede from attention as time passes (until a deliberate shift of attention back in time to such an item is performed). Information acquired less than n seconds ago ("an object of primary memory — belonging to the rearward portion of the present space of time"; James, p. 647) is directly available, although to a decreasing extent over intraepisodic time. Consequently, within the confines of an episode, a monotonic decrement should characterize the ratio of judged duration to the clock time that has elapsed.

13.3. Neuropsychological analysis It should be possible to isolate the extended present by working with patients who have a selective deficit in long-term memory. If, on account of pathology, patients become incapable of recovering information from prior episodes, they would be confined to an existence within the present episode. Amnesic patients, deficient in episodic remembering (Kinsbourne & Wood, 1975; Kinsbourne, 1987), have often been described as living within a "permanent present" (Barbizet, 1970), with little ability to refer even to the immediate past. Asked to think about the past, such a patient described his mind as blank (Tulving, 1983). Given interpolated activity, amnesics forget stimuli within a few seconds. Luria (1973) has characterized the amnesic defect "not so much as a primary weakness of traces as of their increased inhibition by irrelevant interfering stimuli, so that even the slightest distraction inhibits existing traces" (p. 64). But suppose there is no distraction, as in time estimation? The episode might not be quite so brief if the

The extended present

321

subject maintains a consistent mental orientation. It remains to be determined how far back in time these patients' present extends.

13.4. Time estimation The duration of short-term storage might be conveniently measured by presenting information at a steady rate well within the subject's processing capacity, and requesting retrospective time judgments whose accuracy can be related to the time between the relevant input and the timing of the response. Recency and frequency estimations are based on such a steady sequential presentation of input. The most direct measure would be afforded by requesting on estimation of past time. When subjects are asked to perform tasks and warned that they will be asked to estimate how long they took in doing so (prospective time estimation), they will tend to underestimate the past time in proportion to the extent to which the task preempts their attention (Hicks, Miller, Gaes, & Beirman, 1977). This suggests that attention to time in passing provides information for prospective time judgment and incidentally explains why prospective time judgments approximate more closely to clock time than retrospective judgments (Hicks, Miller, & Kinsbourne, 1976), prior to which subjects had no inducement to attend to the dimension of time.

13.5. Experiment 1 Prospective time estimations were required from three subject groups (amnesic, nonamnesic alcoholic, and normal) for periods of time during which they were incrementing a memory drum by one step per subjective second and reading single digits as they appeared on that drum.1 Three groups of male subjects matched for age and IQ took part. They included 12 alcoholic Korsakoff syndrome patients (KS), 12 abstinent alcoholics comparable in age and IQ, and 12 age-matched normal controls. The experimental paradigm was so arranged that two dependent variables could be extracted from the subjects' performance: estimation of present time and estimation of past time. They were asked to make responses at a subjective one per second rate, and after the end of each of a number of predetermined intervals they were asked to estimate how long they had been responding. The number-reading task was designed to preclude the subject from counting in order to keep track of time past. The procedure was modified from Frankenhauser (1959). The subject cranked a roll of cash register paper on which digits 0—9 were typed in a randomized order so that a new digit was displayed every 2 cm. He cranked so as to display each successive number at a subjective one per second rate. He then read the number and so continued until the

322

Kinsbourne and Hicks Korsakoff Group

10

20

30

40 50 Interval (sec)

60

70

80

Figure 13.1. Relationship between mean estimated time and interval: Korsakoff group.

Control Group

1 0

20

30

40 50 Interval (sec)

60

70

Figure 13.2. Relationship between mean estimated time and interval: control group.

present total interval was up. At this time he was asked to estimate the duration of the trial. The following intervals were used: 12,18,26,38,54, and 80 sec. Each was presented nine times in random sequence. Thus each subject made nine judgments of present time (rate of digit reading x number of digits read) and nine judgments of pasttime for each of the six intervals. The results showed that subjects systematically underestimate time present. The patients' present timing performance was normal (Figure 13.1); that is, it did not differ significantly from that of the control (Figure 13.2) or the alcoholic contrast (Figure 13.3) group. The only significant effect in the analysis of variance was for interval, and this

323

The extended present Alcoholic Group

0

10

20

30

40 50 60 Interval (sec)

70

80

Figure 13.3. Relationship between mean estimated time and internal: alcoholic group.

Table 13.1. Time estimations (sec): Experiment 1 Duration (sec)

Control Alcoholic Korsakoff

12

18

16

38

59

80

9.2 9.3 9.1

13.7 13.6 13.1

19.9 18.7 15.3

28.8 27.3 21.4

43.4 38.9 25.1

51.3 52.8 32.0

was totally linear. This result confirms the amnesics' sustained task orientation and their knowledge of how long a second is. In contrast to episodic remembering, which is selectively impaired, information in semantic memory, such as the duration of a second, can be retrieved without difficulty by the amnesic patients. As a conscious and fully deliberate act the estimation of a 1-sec time period cannot be regarded as procedural in the sense of Cohen and Squire (1980). Rather it would fall into their declarative category, and yet contrary to their prediction, the amnesics were well able to perform this act of declarative remembering. In striking contrast was the amnesics' performance on the past time estimation task (Table 13.1), and with respect to the longer intervals used. By trend analysis only the linear component was significant for controls and alcoholics, with slopes of 0.79 and 0.67, respectively. The slope is significantly shallower for the KS than the other two groups (i.e., the Interval x [KS is (AA-C)] interaction, F[l, 33] = 15.9, p < .01). There is a significant quadratic trend for the KS group (F[l, 11] = 8.3, p < .025), but 4% more variance is accounted for by decomposing this into two linear trends, one for intervals

324

Kinsbourne and Hicks

of 12,18, and 28 sec, of slope 0.47 and the other for 38, 58, and 80 seconds, slope 0.29. The intrasubject variance of KS performance was also much greater.

13.5.1. Discussion of Experiment 1 The relationship of judged to clock time up to about 30 sec is consistent with the use of a counter or temporal register (Michon, 1975; Deutsch, 1984) by both contrast groups. The somewhat more gradual slope for alcoholics could be due to slightly less well maintained attention to time passing in this prospective time judgment paradigm than by controls. The KS group's more complex function needs an alternative explanation, as they did not maintain a consistent temporal register and thus were effectively performing retrospective time judgments, without benefit of counter. We suggest that the first limb of the curve (up to 26 sec) is consistent with a discriminability account of the perception of time passing. As in the case of long-term free recall, time periods are judged "much as a line of evenly spaced telephone poles could be approached from one direction in spatial perspective" (Crowder & Greene, 1987, p. 5). The immediate present extends over an appreciable period of time. The episode does not exit abruptly from immediate awareness; it recedes gradually into imperceptibility. Events more than 30 sec ago are no longer efficiently retrievable from immediate experience. We suggest that periods more than 30 sec in the past can be reexperienced only by means of episodic recollection. By that time, information has receded in discriminability to the point that episodic retrieval is necessary to supplement what remains within the present. Given their episodic memory deficits, amnesics would not be expected to be able to recover intact temporal durations that have receded from immediate experience. Indeed, even normal people find it hard to recollect periods of repetitive and featureless activity (see Experiment 2). In amnesics the relationship of judged to clock time should represent the fading from awareness of time past, with minimal overlap of duration retrieval from long-term memory. This appears to have been the case for the Korsakoff group, at least partially. Their performance for these longer time periods is still mostly based on retrieval from the present (episode) or experiential STM. Extrapolation suggests a total exit of an event from immediate experience over not much more than 60 sec. Interestingly, this estimate corresponds to that established by Prisko (1963) in hippocampal damage, for delayed response tasks for clicks, flashes, colors, and nonsense figures. Performance reached chance levels at about a 60-sec delay. An only slightly longer period was established by Baddeley and Warrington (1970) and Warrington (1982) for both normals' and amnesics' information loss in the Peterson paradigm. Amnesics did well in maintaining the primary set of one per second performance. But whereas all groups might have activated a "counter" (e.g., certain neurons in frontal

The extended present

325

Table 13.2. Time estimations (sec): Experiment 2 Duration (sec)

Prospective Retrospective

8

14

24

42

54

73 8.2

13.9 13.5

21.8 21.0

39.1 28.7

49.4 31.9

lobe described by Fuster & Alexander, 1971), maintaining such activity requires continual attention to time, which presumably lapsed in the amnesic group. According to our view, amnesics could nonetheless discriminate time durations up to about 30 sec because they could base this on a still immediately present time experience. It would follow that a similar, but sharper, dissociation should be found between two normal groups performing time estimations on a prospective and a retrospective basis, respectively. Up to 30 sec, foreknowledge should make little difference because the time experience is immediately at hand. Beyond 30 sec the functions relating estimated to clock time should sharply diverge.

13.6. Experiment 2 Three hundred and two young male adults were randomly allocated to two groups: "prospective" and "retrospective."2 All subjects were asked to inspect a tartan pattern in order to rate it for complexity and aesthetic value. The "prospective" subjects were also told that they would be asked to estimate how long they had inspected the pattern. The "retrospective" subjects were not told this. Subjects were randomly allocated to the following inspection durations: 8,14,24,42, and 54 sec. Each subject performed a single trial only, and yielded one data point. At a signal, each subject stopped inspection. He was then asked to estimate the duration of the inspection period. The results were consistent with those of Experiment 1 (Table 13.2). Across the three durations of less than 30 sec, the prospective and retrospective groups rendered comparable estimates (slopes 0.89 and 0.80, respectively, n.s.), as did the two contrast groups in Experiment 1. For periods longer than 30 sec, the estimates of the two groups diverged sharply. The prospective group continued to increment its estimates in constant proportion to increased clock time - slope 0.86 - just as the contrast groups did in Experiment 1. The retrospective group showed a slope of sharply decreasing increments in time estimates with increasing clock time (0.27). This function, which differed beyond the < .001 level from the corresponding value for the prospective group, closely resembled that yielded by the amnesic group in Experiment 1. It did, however, show a statistically reliable discontinuity at around 30 sec, whereas the

326

Kinsbourne and Hicks

amnesics' performance could not reliably be so dissociated over time, but could be fit by a curvilinear function. Amnesics and normals judging retrospectively cannot be formally compared, because the intervening activity was different in the two experiments. But the striking similarity in the functions relating estimated time to clock time raises certain possibilities about the determinants of time estimation. Up to 30 sec, subjects could estimate time past as well without knowing this would be required as with such foreknowledge. In other words, specific attention to time as it passes is not needed. It follows that up to about 30 sec, time duration information is directly available to immediate attention. No act of episodic recollection is called for. Beyond 30 sec, unawareness of the need to attend in time while it is passing seriously handicapped time estimation. It is evidently very difficult to recall durations during which activity was invariant. Miller, Hicks, and Willette (1978) found that retrospective time judgments are a function of the amount of information the subject remembers having processed. The highly confusable responses may have made this difficult in Experiment 1, and in the featureless period of inspection in Experiment 2. Whether one formulates the marker for retrospectively estimated time as the recollectible "content" of that duration (Ornstein, 1969), change within it (Block & Reed, 1978), or events permitting segmentation (Poynter, 1983), both paradigms left subjects singularly devoid of such assistance. It is all the more remarkable (and a novel finding) that for up to half a minute, retrospective judgments approximate prospective judgments, even in such paradigms. Clearly subjects use some other set of cues for their retrospective judgments of intervals of less than 30 sec. The retrospective normal group essentially did no better than the amnesics, even though episodic remembering was available to the former. If the comparison of the two experiments is valid, the defective performance of amnesics beyond 30 sec is therefore accounted for by their forgetting of the instruction that directed their attention to the passage of time.

13.7. Time estimation and the STM-LTM distinction The discontinuity in time estimation differs from that usually considered in current STM and LTM theory. In the time estimation paradigm, differential coding (Craik & Lockhart, 1972) and interpolated activity (Postman & Phillips, 1965) do not seem capable of explaining the discontinuity. Indeed, their absence might account for the relatively long (30-sec) duration of the initial phase. The extended present-past dichotomy survives the absence of these factors. Representations not currently active will be potentially recollectible if their encoding is sufficiently distinctive and contextually marked, and actually recollected if the context and cuing conditions are sufficiently similar (Tulving's synergistic ecphory). The length of time for which representations remain active (if unrehearsed) sets a limit on

The extended present

327

immediate awareness versus reexperiencing (Kinsbourne & Wood, 1975, 1982; Tulving, 1983, 1985).

13.8. Information recovery within the episode The concept of the extended present is distinct from that of the perceptual moment (or specious present). The perceptual moment designates the duration of the present instant, and is measured in fractions of a second. If all other impressions were past, that is, no longer directly discriminate, the observer would experience an unimaginable world of perpetual fade-out and continual recollective effort. As William James explained, the subjective continuity of awareness ("stream of consciousness") indicates that the phenomenal present is a relative concept. Over short periods of time, past is a matter of degree. What has been experienced remains available to an act of attention, although to a decreasing extent. This means that, temporally ordered, percepts remain accessible to direct inspection, for about half a minute, although they diminish in legibility over that period of time. Information is continuously lost, until the item recedes into indiscriminability, as the sliding window of the extended present travels on. The present model for the extended present STM assumes two successive mental operations. The first is the (automatic) registration (representation) of the input. The second is an act of discrimination among inputs within the boundaries of the present. This could be a serial search backward in time. Relative recency judgments support this interpretation. When subjects are asked which of two test items in a list was most recently presented, their response latency is proportional to the ordinal position back in time of the more recent item, suggesting a systematic self-terminating backward scan in time (Muter, 1979; Hacker, 1980). The length of time during which information remains legible within the present will generally be less than the duration of the present. How short it falls presumably depends on the richness of its encoding and its discriminability from interfering stimulation, as well as the identification criterion set by the experimenter. Thus, at a time still within the present, accuracy may fall short of a criterion calling for fully correct identification, but some information transmission may still be demonstrable within a forced-choice or multiple recall paradigm. Conversely, even if more than one item is correctly specified, this does not prove that they are all equally available. A shorter latency to respond may indicate the one more readily returned to attention.

13.9. STM deficits The amnesic is generally credited with a selective LTM deficit, with STM relatively intact (Talland, 1965; Baddeley & Warrington, 1970; Warrington, 1982). But has an

328

Kinsbourne and Hicks

STM deficit been described that complements the LTM amnesia, that is, an STM deficit with generality across modalities and codes? If STM were the persistence of information in the phenomenal present, what form would its impairment take? Theoretically, one might look for a patient who continually loses sight of what has just happened in time (like a visuospatial agnosic loses sight of what he has just seen in space adjacent to his current object of fixation). No such patient has come to notice. What, then, do the present findings imply for the topic of this book, STM deficits? A constricted extended present could relegate the earlier span items to the past, imposing the need for episodic retrieval. This would predict information loss selective to the primary segment of the span, which has not been reported. Alternatively, either the information becomes too rapidly illegible within the extended present, or the representations fail adequately to activate a sequence of corresponding responses. In certain cases the latter postulate is clearly preferable. Cases JO and JT (Kinsbourne, 1972) had spans limited to one item (of two presented). Other than with the simplest monosyllabic words, errors intruded even with simple items. Yet they could successively perform same—different judgments for digit strings up to eight in number. In direct repetition, they would respond correctly with the first of two words. If their extended present were constricted, the opposite would be expected. If input representations lose legibility, either the first or the second response could be lost on different trials. The fact that increasing interstimulus interval improved performance favors functional disconnection between representation and response programming. This hypothesis, which by no means implies a structural discontinuity (e.g., of arcuate bundle, as in traditional connectionistic models), invokes an impairment of concurrent activation of the areas that are responsible for representation of input and of response. An impairment of response programming is indicated by JT's and JO's prolonged latency even of correct responses. Shallice and Vallar (this volume, chapter 1) argue that other cases cannot be so accounted for, for instance, in view of their impairment on a probe digit task calling for a simple yes or no response. If so, they should be distinguishable by an unimpaired reaction time for correct single imitative response. At present, the absence of any case report that can be viewed as demonstrating a selective curtailment of the episode is uninterpretable, as this hypothetical deficit appears not to have been searched for. If such a patient never does come to light, we are dealing either with a distributed representation not vulnerable to a focal lesion or with a concept that needs revision.

Notes 1. An extended report of this experiment has been submitted by R.E. Hicks and M. Kinsbourne for publication. It is entitled "Effect of episodic memory loss in Korsakoff psychosis on time judgments."

The extended present

329

2. An extended report of this experiment is in preparation by M. Kinsbourne and R. E. Hicks.

References 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. Barbizet, J. (1970). Human memory and its pathology. San Francisco: Freeman. Bjork, R. A., & Whitten, W. B. (1974). Recency-sensitive retrieval processes. Cognitive Psychology, 6, 173-189. Block, R. H., & Reed, M. (1978). Remembered duration: Evidence for a contextual change hypothesis. Journal of Experimental Psychology: Human Learning and Memory, 4, 658-665. Cohen, N. J. & Squire, L. R. (1980). Preserved learning and retention of pattern analyzing skill in amnesia: Dissociation of knowing how and knowing that. Science, 210, 204-209. 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. Crowder, R. G., & Greene, R. L. (1987). On the remembrance of times past: The irregular list technique. Journal of Experimental Psychology: General, 116, 265-278. Deutsch, J. A. (1984). Chromomnemonics and amnesia. In L. R. Squire & N. Butters (Eds.), Neuropsychology of memory (pp. 157-169). New York: Guilford Press. Frankenhauser, M. (1959). Estimation of time: An experimental study. Stockholm: Almquist & Wiksell. Fuster, J. M., & Alexander, G. E. (1971). Neuron activity related to short-term memory. Science, 173, 652-65A. Hacker, M. J. (1980). Speed and accuracy of recency judgements for events in short-term memory. Journal of Experimental Psychology: Human Learning and Memory, 6, 651—676. Hicks, R. E., Miller, G. W., & Kinsbourne, M. (1976). Prospective and retrospective judgments of time as a function of amount of information processed. American Journal of Psychology, 89, 719-730. Hicks, R. E., Miller, G. W., Gaes, G., & Beirman, K. (1977). Concurrent processing demands and the experience of time-in-passing. American Journal of Psychology, 90, 431—446. James, W. (1890). Principles of psychology. New York: Holt, Rinehart & Winston. Kinsbourne, M. (1972). Behavioral analysis of the repetition deficit in conduction aphasia. Neurology, 22, 1126-1132. Kinsbourne, M. (1987). Brain mechanisms and memory. Human Neurobiology, 6, 81-92. Kinsbourne, M., & Wood, F. (1975). Short-term memory processes and the amnesic syndrome. In J. A. Deutsch (Ed.), Short-term memory (pp. 258-291). New York: Academic Press. Kinsbourne, M., & Wood, F. (1982). Theoretical considerations regarding the episodic-semantic distinction. In L. Cermak (Ed.), Human memory and amnesia (pp. 195—218). Hillsdale, NJ: Erlbaum. Luria, A. R. (1973). The working brain. New York: Penguin Books. Michon, J. A. (1975). Time experience and memory processes. In J. T. Fraser & N. Lawrence (Eds.), The study of time. New York: Springer-Verlag. Miller, G. W., Hicks, R. E., & Willette, M. (1978). Effects of concurrent verbal rehearsal and temporal set upon judgements of temporal duration. Acta Psychologica, 42, 173—179. Murdock, B. B., Jr. (1972). Short-term memory. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in Research and Theory (Vol. 5). New York: Academic Press. Murdock, B. B., Jr. (1974). Human memory: Theories and data. Potomac, MD: Erlabaum. Muter. P. (1979). Response latencies in discriminations of recency. Journal of Experimental Psychology: Human Learning and Memory, 5, 160-169. Ornstein, R. E. (1969). On the experience of time. Harmondsworth: Penguin.

330

Kinsbourne and Hicks

Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items, journal of Experimental Psychology, 58, 193-198. Postman, L, & Phillips, L. (1965). Short-term temporal changes in free recall. Quarterly Journal of Experimental Psychology, 17, 132-138. Poynter, W. D. (1983). Duration judgement and the segmentation of experience. Memory and Cognition, 11, 77-82. Prisko, L. (1963). Short-term memory in focal cerebral damage. Unpublished doctoral dissertation, McGill University, Montreal. Talland, G. (1965). Deranged memory: A psychonomic study of the amnesic syndrome. New York: Academic Press. Tulving, E. (1983). Elements of episodic memory. New York: Oxford University Press. Tulving, E. (1985). Time and consciousness. Canadian Psychologist, 26, 1-12. Warrington, E. K. (1982). The double dissociation of short- and long-term memory deficits. In L. Cermak (Ed.), Human memory and amnesia (pp. 61-76). Hillsdale, NJ: Erlbaum.

Part IV. Phonological short-term memory and sentence comprehension

The final part of this book includes five chapters dealing with the very contentious issue of the putative role of phonological memory in sentence comprehension. A first and general difficulty for the neuropsychological analysis of this question, which is independent of the specific problem at issue, is that claims for a positive role are based on an association and not on a dissociation of deficits (see Shallice, 1988, for discussion). This makes it difficult to determine the precise nature of the putative causal role of immediate memory deficits in the genesis of the comprehension impairment in patients with a specific span defect. Consider, for instance, the observations that short-term memory patients, even though their comprehension is comparatively preserved at the clinical level, typically show some impairment in tasks such as the Token Test and that a classical dissociation (i.e., patients with a defective auditory-verbal span and an entirely preserved sentence comprehension) had not been reported so far (see reviews in Caplan & Waters, chapter 14, and Vallar, Basso, & Bottini, chapter 17). This pattern prima facie suggests a role of phonological memory in speech comprehension. However, an association of neuropsychological symptoms produced by brain damage might arise from the anatomical contiguity of the neural structures involved in immediate retention of verbal material and in speech comprehension. After all, the neurological correlate of auditory—verbal short-term memory deficits is a lesion of a left inferoposterior parietal region, which is a part of the language areas. The hypothesis that short-term phonological memory has a specific role in speech comprehension would need to be corroborated by sets of experiments from which it is possible to rule out a simple interpretation in terms of anatomical proximity. A second problem is that, unlike the situation for the functional architecture of verbal memory systems (see Shallice & Vallar, chapter 1), the available evidence from normal subjects concerning the possible role of immediate phonological memory in speech comprehension does not appear to provide immediately relevant information (see Caplan & Waters, chapter 14; Martin, chapter 15). For instance, it is not clear whether syntactic parsing systems include an immediate memory component having functional characteristics broadly comparable to those of the phonological store. Finally, a great 331

332

Part IV

deal of evidence indicates that some syntactic and lexical—semantic processes occur online, probably with no or minimal need for immediate storage (see a review in Martin, chapter 15). The chapters in this part are primarily concerned with neuropsychological results. Two contributions (Caplan & Waters, chapter 14; Vallar et al, chapter 17) provide reviews of the pattern of comprehension impairment occurring in span-impaired patients. Caplan and Waters's review is much the more extensive and detailed and gives all the relevant information available for each patient (phonological processing, shortterm memory, comprehension), presented in tabular form. Vallar et al., in the introductory pages of their case study, focus on the characteristics of the sentence comprehension impairment of patients with a defective phonological input store. Data from brain-damaged patients with lesions acquired in adult age are presented in three chapters in this part (see also the results reported by McCarthy & Warrington, chapter 7, and by Butterworth, Shallice, & Watson, chapter 8). Martin (chapter 15) investigates the sentence comprehension deficit of fluent and nonfluent aphasics (according to the traditional aphasiological taxonomy) with a defective memory span. According to Martin, there is some correspondence between the fluent-nonfluent distinction and components of the sort of functional architecture outlined by Shallice and Vallar (chapter 1) and Baddeley (chapter 2), in that rehearsal is preserved in fluent, but not in nonfluent patients. Saffran and Martin (chapter 16, case TI) and Vallar et al. (chapter 17, case ER) investigate the immediate memory and comprehension performance of two patients having a defective phonological store. Both cases seem likely to suffer from associated additional deficits: visual store and phonological processing impairments, respectively, in patients TI and ER. In both patients, however, the functional locus of the deficit involves input components and cannot be traced back to output difficulties (cf. Howard & Franklin, chapter 12). Crain, Shankweiler, Macaruso, and Bar-Shalom (chapter 18) assess the role of working memory in sentence comprehension in a different pathological population, namely, children suffering from developmental dyslexia. Developmental dyslexics have long been known to have a much reduced span (for a review see Perfetti, 1985). Crain et al. conceive verbal working memory as a language-specific system, which has the primary function of facilitating the extraction of a meaning that corresponds to the linguistic input. They distinguish two components: a phonological storage buffer, where material may be maintained by continuous rehearsal, and a control mechanism, which relays the results of low-level analysis upwards through the language apparatus (see also the short-term memory models outlined by Shallice & Vallar, chapter 1, and by Baddeley, chapter 2; the control short-term memory components of McCarthy & Warrington, chapter 7, and of Craik, Morris, & Gick, chapter 10). The authors of the five chapters in this part make specific suggestions concerning the

Short-term memory and sentence comprehension

333

role of phonological short-term memory in sentence comprehension (as do McCarthy & Warrington, chapter 7, and Butterworth et al, chapter 8). Their proposals differ in a number of important respects. In none of them, however, at variance with the earlier models of Clark and Clark (1977) and Caramazza and Berndt (1985), is the phonological store involved in span performance regarded as the working space of the syntactic parser. Conversely, all authors agree that syntactic processing per se does not necessarily require the availability of a phonological short-term memory component. Martin (chapter 15) suggests two possible specific roles for phonological short-term memory in sentence comprehension. According to the "downstream phonological buffer hypothesis," the store could be conceived as a preparsing buffer for maintaining subsequent information in a sentence, while the analysis of an earlier portion is completed. Second, noting that phonological memory obviously contributes to the verbatim repetition of sentences, she suggests a possible contribution when the derived meaning is difficult to retain (e.g., the Token Test), since the material approaches random lists, where few strong and predictable relationships among the words exist (cf. McCarthy & Warrington, chapter 7, and Butterworth et al., chapter 8, for related arguments concerning the role of phonblogical short-term memory in the comprehension of Token Test-like sentences). Finally, drawing on experiments with garden path sentences in span-impaired patients, she concludes that the store is not involved as a backup system in the process of reanalysing an originally misinterpreted sentence (see McCarthy & Warrington, chapter 7, for a discussion of the backup role of the phonological store, and Butterworth et al., chapter 8, for related results). These conclusions fit with Crain et al.'s observation (chapter 18) that poor readers with defective working memory are riot disproportionately affected by garden path effects. The contributions by Caplan and Waters (chapter 14), Saffran and Martin (chapter 16), Vallar et al. (chapter 17), and Crain et al. (chapter 18) contain specific proposals concerning the role of phonological memory in sentence comprehension. These different views have a common factor, namely, that verbal short-term memory is thought to be a system that allows integration of and/or interaction between different levels of analysis and representation. According to Caplan and Waters (chapter 14), phonological memory may be useful when automatic on-line processes do not provide an adequate interpretation of a given sentence, for instance, if syntactic, heuristic, and lexicopragmatic processes do not all provide compatible accounts, as might happen in the case of semantically reversible sentences. Under these circumstances some postinterpretive processes should adjudicate between the available readings. The check of the alternative available interpretations, necessary to provide a final decision, would take place using the verbatim record provided by the phonological store. Saffran and Martin (chapter 16) suggest that phonological memory is involved in the mapping between structural information and thematic roles. This interpretation

334

Part IV

assumes that structural and lexical information are represented separately, and in Saffran and Martin's metaphor the phonological record serves as a glue that binds together the various representations constructed in sentence processing. Difficulties would not be expected when lexical input can be assimilated immediately into a semantic structure. They would emerge, however, when interpretations depend crucially on information that comes later in the sentence (but see Martin, chapter, 15) and when semantic operations are overloaded. Vallar et al. (chapter 17) suggest that the phonological short-term store is involved in the mapping of the syntactic description of a given sentence onto its representation based on lexical—semantic processes. They assume that the syntactic representation is confined to a structural description and that lexical—semantic processes do not primarily take into account information carried by the linear arrangement of words (see Saffran & Martin, chapter 16, for a related account). The contribution of the phonological store becomes crucial when the mapping process is not constrained by semantic factors, such as in the case of semantically reversible sentences and items made semantically anomalous by a word reversal. This type of material, Vallar et al. note, represents in fact a source of major difficulty for patients with a defective phonological memory (but see McCarthy & Warrington, chapter 7). Crain et al. (chapter 18) view the control mechanism of their working memory system as a structure that carries out a series of successive translations between different levels of representation, from input sentences to the semantic representations or plans that they evoke. Situations in which the translation from source to target language (i.e., the semantic representation) is comparatively simple would require a minor involvement of working memory. The process of "sequential lookup and concatenation," however, may not be straightforward, calling instead for additional computations, which would tax the resources of working memory. In a series of experiments Crain et al. show that sentence comprehension of poor readers is disproportionately impaired when the preceding pragmatic context is not advantageous (no-felicity vs. felicity condition). Poor readers are also defective in the processing of sentences containing temporal terms, when the conceptual order conflicts with the linear arrangement of the constituent words. In this respect, Crain et al.'s position is related to McCarthy and Warrington's (chapter 7) suggestion that backtracking to the phonological record may be required when extralinguistic assumptions bias the interpretation of the spoken message. The specific suggestions discussed so far assign more or less important roles to phonological memory in aspects of sentence comprehension. At a more general level, however, different views exist as to the ecological utility of this storage system, Caplan and Waters (chapter 14) conclude their detailed review by stating that, at least in adult humans, phonological short-term memory has a marginal role in normal language processing (see also Martin, chapter 15). According to other contributors, verbal short-

Short-term memory and sentence comprehension

335

term memory has a more central position in mental function, either as a languagespecific component (the working memory of Crain et al, chapter 18) or as a multipurpose short-term retention system (Baddeley, chapter 2; McCarthy & Warrington, chapter 7; Vallar et alv chapter 17). On this latter view, the cognitive role of verbal short-term memory is not confined to aspects of speech comprehension, important as they are, extending instead to other mental activities such as a language learning, mental arithmetic, and reasoning.

References Clark, H. H., & Clark, E. (1977), Psychology and language. New York: Harcourt Brace Jovanovich. 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. Perfetti, C. A. (1985). Reading ability. New York: Oxford University Press. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press.

14. Short-term memory and language comprehension: a critical review of the neuropsychological literature DAVID CAPLAN AND GLORIA S. WATERS

14.1. Introduction Research on memory has provided considerable evidence for a verbal short-term memory (STM) system that is involved in memory tasks, such as span, free recall, probe recognition, and the Brown—Peterson paradigm, in which subjects must retain small amounts of linguistic information over brief periods of time. There is evidence that the representations in STM include phonological forms (Conrad, 1964) and that these representations are maintained in STM in part through a process that involves articulatory rehearsal (Baddeley, Thomson, & Buchanan, 1975). The specific character of the STM system is the subject of investigation and debate. One view (see Shallice & Vallar, chapter 1, and Baddeley, chapter 2, this volume) maintains that auditorily presented items are entered into a phonological store (PS) directly (Salame & Baddeley, 1982; Baddeley, Lewis, & Vallar, 1984; Greene & Crowder, 1984), while printed items are entered, at least in part, through a controlled process of subvocal rehearsal (Murray, 1968; Levy, 1971; Peterson & Johnson, 1971; Estes, 1973; Baddeley et al., 1984). There appears to be a progressive diminishing of the strength of the phonological representations in the phonological store over a period that has been variously estimated to last between 2 and 20 sec (Crowder & Morton, 1969; Wickelgren, 1969; Darwin, Turvey, & Crowder, 1972). However, the strength of these representations can be increased through the controlled use of articulatory rehearsal processes (the articulatory loop [AL]: Baddeley et al., 1984; Baddeley and Lewis, 1984). Whether the articulatory-based processes that are responsible for the entry of printed items into the PS are the same as those involved in maintaining the strength of phonological We would like to thank Giuseppe Vallar, Tim Shallice, Brian Butterworth, and Stephen Crain for their detailed comments on an earlier version of this chapter. We would also like to thank the authors of many of the case studies reported on here for verifying the information about their patients presented in the tables. Any errors that may remain are the responsibility of the authors. This research was supported by a Medical Research Council of Canada grant (#MA8602) and by a Chercheur Boursier award from the de la Recherche en Sante du Quebec to David Caplan and by a Natural Sciences and Engineering Research Council of Canada University Research Fellowship and grant (#U0468) to Gloria Waters.

337

33S

Caplan and Waters

representations in this store, and whether nonarticulatory mechanisms are also involved in both these processes, are also issues that are the subject of research and disagreement (see Howard & Franklin, this volume, chapter 12). Despite these uncertainties surrounding the detailed structure and processes involved in the STM system, there is good agreement that the STM system does involve a phonological store and an articulatory loop, characterized by passive decay and rehearsal functions, respectively. In this review, we shall focus on the role of this system in input-side language processing: recognition and comprehension processes. Many psychologists — and especially neuropsychologists — have argued that a short-term memory system with these components plays a role in language processing. Baddeley and his colleagues have concluded that the PS is necessary for certain aspects of normal auditory sentence comprehension (Vallar & Baddeley, 1984a) and that the AL plays a role in written sentence comprehension (Baddeley, 1979; Baddeley & Lewis, 1981). Caramazza and Berndt (1985) state that "morphologically interpreted phonological representations are placed in a phonological working memory for syntactic parsing" (p. 46) and claim that "the evidence that a phonological working memory system is implicated in syntactic parsing comes from studies of both normal and pathological populations" (p. 48). Other authors maintain that a phonological short-term memory system and the process of articulatory-based rehearsal play important roles in reading single words and the comprehension of text (e.g., Kleiman, 1975). Some of these hypotheses have come under fire recently, such as the view that STM is involved in syntactic comprehension (Caplan, Vanier, & Baker, 1986b; Martin, 1987; McCarthy and Warrington, 1987a; see chapters in this volume by Martin, chapter 15; McCarthy & Warrington, chapter 7; Howard & Franklin, chapter 12). The role the STM system plays in language processing is thus now far from clear. In this chapter, we review the neuropsychological evidence relating STM and language comprehension. We will advance a specific hypothesis regarding the role the STM system plays in language comprehension on the basis of our interpretation of these data. This hypothesis is broad, inasmuch as it applies to all aspects of language processing. To the best of our knowledge, it is compatible with data currently available from studies of normal as well as pathological subjects, although many data relevant to the hypothesis are not available.

14.2. A component-based classification of STM impairments The main criterion for recognizing a disturbance of the auditory—verbal STM system is the presence of a severely reduced memory span for all types of auditorily presented verbal material. In addition, there must be evidence that the deficit does not result simply from impaired speech perception or from the speech production requirements of

STM and comprehension: neuropsychological studies

339

the span task. Thus, the patient must either demonstrate normal spontaneous speech, single word repetition, and confrontation naming, or reduced span on probe recognition and pointing tasks where verbal output is minimal or nonexistent. Some authors have claimed that memory span for visually presented material should be normal or less reduced than that for auditorily presented material and that there should be evidence for intact long-term memory (LTM) functions. Although these criteria demonstrate that a patient has a selective auditory-verbal short-term memory impairment, in some cases patients have been found to also have visual short-term verbal memory impairments (e.g., cases TB, TI, BO, discussed later) or long-term memory impairments (e.g., cases TB, EDE, also discussed later), which have been argued to be independent of their auditory-verbal short-term memory impairments. Variation within the reduced STM span as a function of lexical variables has been taken as an indication of reliance on LTS and thus as additional evidence for an STS disturbance (Saffran & Marin, 1975; Caramazza, Berndt, & Basili, 1983). In what follows, we shall allocate STM patients to one of four groups: (1) patients with impaired phonemic processing; (2) patients with an impaired PS despite good phonemic processing; (3) patients with a disturbed AL; and (4) patients with abnormalities in metalinguistic tasks requiring manipulation of phonological representations. We feel it is important to keep these sources of STM impairments separate when approaching the question of the relationship of STM to language processing, for several reasons. First, these different sources of STM impairments may themselves have consequences for language processing. For instance, major disturbances affecting phonemic discrimination and identification may affect auditory single word comprehension (Saffran, Marin, & Yeni-Komshian, 1976). In contrast, a primary disturbance of the AL or PS would not be expected to lead per se to a disturbance of lexical access or single word comprehension and thus, in a pure AL or PS case, any language-processing impairment beyond the single word level could be more easily related to a disruption of AL or PS functions. Second, different sources of STM impairments may be expected to affect different language functions. For instance, disturbances of the AL might be expected to affect processing of written language more than spoken language, since, according to the researchers cited earlier, the AL is thought to be involved in entering printed items in the PS, while auditorily presented items are thought to enter the PS directly. Third, it cannot be assumed that the impairments of STM functions created by these different sources are identical. Acoustic-phonetic processing impairments may lead to disruptions of PS functions that are quite different from those occasioned by a patient's disturbance in manipulation of phonological segments or by some primary disturbance of the PS. Despite the desirability of keeping these sources of STM impairments separate, allocating patients to groups on the basis of disturbances of each of these mechanisms is not simple. Patients with low-level phonological processing and metalinguistic

340

Caplan and Waters

impairments are the easiest to identify, since they can be distinguished from other STM patients on the basis of their performance on nonmemory tasks. An impairment in acoustic-phonetic or other auditory processes has been inferred from abnormal performances in phoneme identification and discrimination tasks, and tasks that require discrimination and identification of synthetic speech sounds that vary in phonetically salient dimensions such as voice onset time (VOT). Vallar and Baddeley (1984a) have argued that these tasks involve memory components, and the conservative stance is therefore to require normal performance on some other task with comparable memory requirements. Berndt and Mitchum (this volume, chapter 5) have suggested that the presence of phonemic paraphasias in single word repetition may also be taken as evidence for a disturbance of acoustic-phonetic processes. Many STM patients make a few phonemic paraphasias, and so could be included in this group according to this criterion. However, this criterion is clearly too lax, as many phonemic paraphasias unquestionably arise in the process of planning and executing speech output (Caplan et al., 1986a; Bub, Black, Howell, & Kertesz, 1987). We thus assume here that it is misleading to include in this group cases whose only evidence for low-level sublexical phonological processing disorders is the existence of phonemic paraphasias in single word repetition. It should be noted, in addition, that the entire range of tests that are available from normal studies to assess acoustic-phonetic processing fully have not yet been applied to patients, and that we may be seriously underestimating the extent of disturbances of acoustic-phonetic processing in these cases (and in aphasia in general). In practical terms, there is no agreement on how complete the assessment of acoustic-phonetic processing must be to exclude a disturbance in a patient or whether abnormalities that affect acoustic-phonetic functions only in extreme conditions (e.g., discrimination in noise; Martin, this volume, chapter 15) should serve to classify a patient in this group. We resolve this question by including in this group all patients with any disturbance of input acoustic-phonetic processing, however documented. Patients with "central" phonological processing impairments or abnormalities in metalinguistic tasks requiring manipulation of phonological representations can also be distinguished from other short-term memory patients on the basis of their performance on nonmemory tasks. These patients have impaired performance on metalinguistic tasks that require the manipulation of segmental phonological representations despite intact low-level phonological processing. Thus, the criteria for inclusion in this group include adequate performance on tests of low-level phonological processing coupled with impairments on metalinguistic tasks, such as ones in which the patient must create spoonerisms or form a word from the initial segments of a series of sequentially presented words. The allocation of patients to the PS and AL subgroups is much more difficult. The majority of STM patients reported on in the literature have been claimed to have impairments of the PS. In addition to the general requirements noted earlier for the

STM and comprehension: neuropsychological studies

341

establishment of an auditory-verbal STM impairment, other criteria that have been claimed to establish the disturbance as one of the PS include a reduced recency effect and increased forgetting in the Brown-Peterson filled delay paradigm for lists within span. One might also expect these patients to show a reduced or absent phonological similarity effect with auditory presentation, given that the presence of this effect has been claimed to be a function of the phonological store. However, some authors (Shallice and Vallar, personal communication) have claimed that patients with PS disturbances may simply have a PS with a reduced capacity, which would result in the presence of an auditory-phonological similarity effect at a lower level of performance. Several different sets of criteria have been used to classify STM impairments as resulting from a disturbance of the AL. In some cases (e.g., BO), a disturbance of the AL has been claimed on the basis of a pattern of performance on measures of the effect of phonological similarity and word length on auditory and visual span that is similar to that found for normal subjects under conditions of articulatory suppression (i.e., normal effect of phonological similarity on auditory span but reduced or absent effect of phonological similarity on visual span and of word length on both auditory and visual span). In other cases (e.g., MK), patients have been claimed to have an AL impairment on the basis of a pattern of deficits that would make it impossible for them to use the AL (e.g., inability to convert output to input phonology coupled with an inability to directly convert input to output phonological representations and a disrupted auditory input lexicon). It should be noted that although particular patients have been claimed to have disturbances of the PS or the AL on the basis of their STM performance, what is striking to us is the similarity of the performance of all STM patients assessed to date on tests of STM function. All but four patients (two patients, TB and TI in the PS subgroup and two patients, RE and EDE, in the central phonological processing subgroup) have been found to show normal effects of phonological similarity on auditory span, and all patients who have been tested have shown absent or significantly reduced effects of phonological similarity on visual span and of word length on both auditory and visual span. In addition, all but one patient (JS) have shown a reduced recency effect, and all patients tested have shown increased forgetting after a delay with auditory presentation in the Brown-Peterson paradigm. The similarity of the patients in these four subgroups on tests of STM function may result from the fact that several of these mechanisms are not as theoretically distinct as we have indicated. For instance, disturbances of manipulation of phonemic segments in metalinguistic tasks and disturbances of the PS itself may not be theoretically distinguishable; it is possible to claim, for example, that the former result from the latter. In addition, it is very likely that many STM cases in the literature have more than one primary impairment. The variable empirical criteria adopted by different researchers for establishing the source of an STM deficit make it extremely difficult to classify a

342

Caplan and Waters

particular patient with certainty and may contribute to the failure of any coherent pattern of STM impairment emerging for these four groups. For instance, PV (Vallar and Baddeley, 1984a) is analyzed as a patient with a PS disturbance despite the fact that her performance on span tasks is similar to that of normal subjects under conditions of articulatory suppression because her span is more reduced and she has a larger visual-auditory presentation advantage than is seen in normals under conditions of articulatory suppression. Vallar and Baddeley argue that PV's performance with respect to the phonological similarity and length effects is due to the strategic nonuse of the AL. In contrast, EDE (Berndt & Mitchum, this volume, chapter 5) shows the pattern of recall performance expected of a patient with a disturbance of the PS - that is, no effects of either phonological similarity or word length with either auditory or visual list presentation. The evidence for her having a PS deficit is thus much more direct. Similarly, the criteria used to decide that MK (Howard & Franklin, 1987, and this volume, chapter 12) has an AL deficit are very different from those used to reach the same conclusion in case BO (Hildebrandt, Waters, & Caplan, submitted). These variable criteria contribute to difficulty in classification. Finally, we must take seriously the possibility that the model of the internal structure of the auditory—verbal STM system we have adopted is incorrect, and that similarities in patients' performances reflect damage to an as-yet-underspecified component in the auditory-verbal STM system (for discussion of the structure of the auditory—verbal STM system, see Shallice & Vallar, chapter 1; Baddeley, chapter 2). Despite these uncertainties and difficulties in classification, the need remains to subdivide patients because of the potentially different effects of different STM disturbances on language comprehension functions. For our purposes here, patients have been classified into the four subgroups indicated earlier on the basis of the original classification by the authors. In those cases where it was not clear from the original report how a particular patient would be classified we have used the criteria outlined previously. We now provide a framework for viewing language processing, and then present the data we have been able to glean from the literature on the phonological processing, STM, and language functions of the patients in each of these four groups separately.

14.3. A framework for viewing language processing Language consists of a large number of types of representations (phonological, morphological, syntactic, semantic), and its processing is correspondingly complex. In our discussion, we shall consider only the major elements of language comprehension: processing of sublexical elements of word form (phonemes, letters, graphemes), recognition of words, accessing of literal meanings of words, construction of sentence form, assignment of the literal meanings of sentences, and assignment of discourse

STM and comprehension: neuropsychological studies

343

structure. We need to draw a few distinctions regarding several aspects of these processes that are relevant to the role STM may play in language comprehension. First, we need to distinguish between processes that utilize phonological representations directly and processes that do not directly utilize such representations. For instance, lexical access for auditorily presented words directly involves phonological representations. Similarly, although not obvious on a priori grounds, lexical access for low-frequency written words also seems to involve phonological representations; the activation of these representations is the basis for spelling-sound regularity effects in word naming and lexical decision tasks (Seidenberg, Waters, Barnes, & Tanenhaus, 1984; Waters & Seidenberg, 1985). In contrast, once the identity of lexical items has been ascertained, assigning the syntactic structure of a sentence and using that structure to determine aspects of the meaning of the sentence cannot use segmental or lexical phonological representations directly, but is based on syntactic, prosodic, and semantic information. One question we shall address is whether phonological representations, maintained in the STM system, are used in some indirect fashion in processes that cannot use them directly. Second, we distinguish between processes that require look back and those that do not. For instance, the assignment of reference to pronouns, anaphors (reflexives, reciprocals), and phonologically empty, referentially dependent categories (as in assigning the agent of to jump in Peter promised Henry to jump, or the object of push in The girl who the child saw the boy push fell down) requires look back to previous nouns in a discourse. With written presentation, the physical form of the input is present to be reviewed, but with auditory presentation (and with experimental visual presentation techniques such as rapid serial visual presentation RSVP), such input cannot be reviewed in physical form and this lookback requires a memory system. We shall review the neuropsychological evidence regarding the role of the STM system in these processes: Is STM the sole memory system obligatorily involved, is it optionally involved, or is it not involved in processes requiring review of a previously presented input? Third, we distinguish between what we call first-pass language processing and processes that reconsider the output of first-pass processing (second-pass processes). We consider that first-pass language processing involves the recognition of word and sentence form, the assignment of the literal meaning of words and sentences, and the construction of a coherent discourse structure. The output of first-pass processing may be reconsidered in cases of ambiguity and/or incoherence. For instance, at the level of lexical processing, it appears that all senses of lexically semantically ambiguous words are accessed in first-pass processing even in previously disambiguating contexts and that the context-unpreferred reading decays in strength over a short period of time (Swinney, 1979; Tanenhaus, Leiman, & Seidenberg, 1979). Reconsideration of the contextually unpreferred reading after this period of time has passed is a second-pass process.

344

Caplan and Waters

At the sentence level, being garden-pathed (e.g., misanalyzing sentences such as The boat floated down the river sank) reflects the operation of first-pass processing of syntax, prosody, and lexical meaning; recovery from a garden path is a second-pass process that involves reconsideration of the structures already assigned. At the discourse level, assignment of the reference of a pronoun is determined by features of discourse structure, such as the existence of focused nouns; reassignment of such reference involves reconsideration of the reference previously assigned in a second-pass process. The distinction between first- and second-pass processing is important to consider in evaluating the role of STM in language comprehension for several reasons. First, it relates to the distinction between language-processing operations that require lookback and those that do not. All second-pass processes require the review of previously presented input; only some first-pass processes - those that involve relating an element to a previously presented element - require such review. Second, some first-pass processes may involve look ahead. For instance, some deterministic parsers (e.g., Marcus, 1980; Berwick & Weinberg, 1984) maintain lexical items along with their syntactic categories in a "lookahead," "preparsing" buffer prior to the construction of phrase markers; these lexical representations may be phonological and their maintenance may involve STM (see Baddeley, Vallar, & Wilson, 1987, who also cite Clark & Clark, 1977, in reference to similar claims). Second-pass processes do not involve lookahead. Third, we think first- and second-pass processing can involve different modes of processing — automatic and controlled modes — which may involve STM in different ways. We now briefly discuss automatic and controlled processing as they may relate to both STM and language processing.

14.4. Automatic and controlled processing in STM and language If we consider the different sources of STM impairments, we believe that there is a grouping of these sources of STM impairments that is of potential interest with respect to the relationship of STM to modes of language processing. Two sources of STM dysfunction - disturbances of phonetic processing and impairment to the PS itself - affect what may be called "automatic" aspects of phonological processing. Two sources of STM impairments — disturbances affecting higher-order manipulation of phonological segments and disturbances of the AL - affect what may be considered to be "controlled" aspects of phonological processing. Automatic and controlled processing are two types of processing that have been postulated in a theory of attention developed by Shiffrin and Schneider (1977; Schneider & Shiffrin, 1977). Automatic processing occurs rapidly, without effort, and requires little processing capacity. Controlled processing occurs slowly and requires effort and processing capacity. Baddeley's view (see Baddeley, chapter 2) that auditorily presented phonological material enters the PS automatically, while written material requires the

STM and comprehension: neuropsychological studies

345

optional use of the AL, is consistent with this view of the nature of the different sources of STM disruptions. Several other authors (Butterworth, Campbell, & Howard, 1986; Friedrich, chapter 3; McCarthy & Warrington, 1987a, chapter 7; Howard & Franklin, chapter 12) make suggestions that are similar to this division of the sources of STM impairments. The division of STM impairments into those that reflect disruptions of automatic and controlled processing of phonological representations has a certain a priori appeal in attempting to relate STM functions and language processing. Although the concepts of automatic and controlled processing have been most extensively explored in experimental paradigms (primarily priming and Stroop tasks) that have little direct contact with normal psycholinguistic processing, a considerable literature has grown up that divides psycholinguistic tasks into those that are automatic and those that involve controlled processes. With respect to the processing involved in language comprehension, many psycholinguists maintain that on-line first-pass processes are involuntary and automatic (e.g., Fodor, 1983). Second-pass reconsideration of the outputs of these processes when ambiguities and incongruities arise may be automatic in some circumstances, but often involves controlled processing. Most of normal language processing is likely to involve automatic first-pass processing and possibly automatic second-pass processing. It is therefore of greatest interest to evaluate the role of STM in these types of processes. Both automatic and controlled tasks may or may not involve phonological representations. As discussed earlier, lexical access for both auditory and written words is an automatic first-pass process that involves phonological representations; much of parsing is an automatic first-pass task that cannot use phonological representations directly. Metalinguistic tasks involving operations on or comparisons of phonological forms (rhyme judgment, pseudohomophone recognition, written homophone judgment, spoonerism creation, forming a word from the initial segment of a series of words, etc.) require controlled processing of phonological form. Other controlled processes (e.g., recovery of the contextually unpreferred reading of an ambiguous word) do not necessarily require recourse to phonological representations, and it is an empirical matter whether or not they do in fact involve such representations. A cautionary note is needed regarding the assessment of automatic functions. We must distinguish between the automaticity of psycholinguistic functions and that of the tests used to assess their integrity. For instance, on the view of psycholinguistic automaticity described earlier, acoustic—phonetic processing and phonemic identification are automatic processes. However, as Vallar and Baddeley (1984b) point out in their discussion of Allport's tests of these functions in STM patients (Allport, 1984a), testing these functions with phoneme discrimination tests introduces a memory component into the task the patient must perform, and failure to perform normally on phoneme discrimination tests may reflect a memory, not a discrimination, impairment.

346

Caplan and Waters

Since there are few automatic tests of automatic functions that have been used with neuropsychological cases (such as pseudohomophone Stroop effects or rhyme priming; evoked potentials have the promise of providing other methods: see Garnsey, Tanenhaus, & Chapman, 1987; Starr and Barrett, 1987; see also Starr et al, this volume, chapter 4), we can at best reason as follows. If a patient performs normally on a controlled task that adequately assesses an automatic function, that automatic function is normal (as are the relevant control processes). If a patient fails on such a task, but can accomplish a task requiring comparable control mechanisms (e.g., failure on phoneme discrimination but success on written lexical decision), we can conclude that the automatic function is not normal. If the patient fails on a controlled task that assesses on automatic function and no other evidence of his being able to accomplish the controlled processing necessary for that task is available, we cannot unequivocally locate the locus of his impairment.

14.5. A digression regarding sentence comprehension Parsing auditorily presented sentences is a task in which a role has been suggested for phonological representations in a first-pass automatic process that logically does not require such representations and in which such representations cannot be used directly. Since this is an area in which STM patients have been extensively tested, we will briefly digress here to discuss the issues surrounding the possible role of phonological representations in parsing and the criteria for providing relevant neuropsychological evidence for such a role. First, we assume on the basis of work such as that by Frazier and Fodor (1978), Frazier, Clifton, and Randall (1983), Ferreira and Clifton (1986), Carlson and Tanenhaus (1987), Stowe (1989), and others that, during the process of comprehension, syntactic structures are assigned automatically on-line to sentences and used to determine aspects of semantic representations such as thematic roles, scope of quantification, attribution of modification, and coreference. Disorders of this process are explored in Caplan and Hildebrandt (1988). As noted earlier, the parser cannot itself make direct use of the phonological form of words once lexical access has been achieved, but operates on the syntactic properties of lexical items (category and subcategorization information). It may use phonological representations in lookahead or lookback processes, and such representations may be maintained in the STM system. Researchers have used the co-occurrence of deficits in STM with (alleged) deficits in parsing to argue for a role of STM in parsing. Such co-occurrences of deficits do not prove that the two functions are related (see Caplan, 1987; Bub & Bub, 1988), but are compatible with their being related. Conversely, the failure of parsing deficits to occur in patients with STM limitations would show that STM is not involved in parsing. To assess whether STM and parsing deficits co-occur, we must be able to identify deficits in

STM and comprehension: neuropsychological studies

347

both functions. We thus present the criteria we think need to be met to document the existence of a deficit in parsing. These criteria are related to the way that sentential semantic features (e.g., thematic roles) are extracted from sentences. Simultaneously with their recovery through the use of a parser, some sentential semantic features can be assigned by processes that do not use syntactic structures. These processes access lexical meanings and combine them pragmatically and in relationship to discourse structure to assign some semantic features. The interaction of syntactic complexity and semantic reversibility in determining reaction times in sentence-picture matching verification tasks (Slobin, 1966) provides evidence that both the syntactic and the lexicopragmatic routes to meaning are normally operative. There is excellent evidence that this "lexicopragmatic" route to meaning is frequently preserved in patients with damage to the syntactic route (Caramazza & Zurif, 1976; see Caplan & Hildebrandt, 1988, for discussion). In addition, some readings can be derived by heuristics (see Bever, 1970; Saffran & Marin, 1975; Caplan & Hildebrandt, 1986, 1988). These heuristics can be based on the linear sequence of lexical categories in a sentence and/or on specific lexical items such as the passive morphology or the agentive preposition by in English. Heuristics produce syntactic representations that have less hierarchical structure and fewer categories than those provided by the parser (e.g., phonologically null categories appear to be omitted in structures created by heuristics); several authors even refuse to term these representations "real" syntax (see Caplan & Futter, 1986; Caplan & Hildebrandt, 1986, 1988; Grodzinsky, 1986, for discussion of these heuristics). If the syntactic, heuristic, and lexicopragmatic routes to meaning are all normally operative, sentences such as (la) would receive two immediate readings: (lb) is derived syntactically, lexicopragmatically, and by a heuristic that assigns agency to an NP in a by phrase; and (lc) is derived lexicopragmatically and by a heuristic that assigns agency to the sentence initial or the preverbal NP: la. The boy was kissed by the girl. lb. The girl kissed the boy. lc. The boy kissed the girl.

Some postinterpretive process must then apply to adjudicate between these readings. The fact that some sentential semantic features can be recovered lexicopragmatically and through the use of heuristics sets requirements that must be met to document that a patient has a disturbance of syntactic comprehension. To prove that a patient has such a disturbance, it is not enough to show that a patient has a disturbance interpreting semantically reversible sentences but not semantically irreversible sentences, since the former yield two plausible readings (one syntactically and one lexicopragmatically) and the patient may have trouble with a second-pass process: adjudication between these two readings. To show that a patient has a disturbance of syntactic comprehension, one must demonstrate either a complexity effect in syntactic comprehension (i.e., more complex structures must be harder for the patient) or a deficit specific to particular

348

Caplan and Waters

syntactic structures or parsing operations. Relative complexity of different sentence types can be determined on the basis of linguistic and/or parsing theory, or derived from empirical studies of both brain-damaged and normal subjects (Cook, 1975; Caplan, Baker, & Dehaut, 1985). Parsing operations are suggested by psycholinguistic research (see Dowty, Karttunen, & Zwicky, 1985, for examples). Caplan and Hildebrandt (1988) discuss the documentation of parsing disorders in detail. Additional evidence is required to make a case that a parsing deficit is related to an STM impairment. Two pieces of evidence that would relate an STM limitation to a syntactic comprehension disorder have been suggested in the literature. One is that the syntactic comprehension deficit is worse with longer (padded) sentences (Saffran, 1985; Baddeley et al. 1987). As Baddeley et al. (1987) point out, length is often confounded with syntactic complexity, but the two can be separately varied in carefully constructed material. However, even if additional length does not impose additional parsing demands, it always adds to the complexity of the propositional content of a sentence, which is maintained in long-term memory. Thus, from a pragmatic point of view, it must be ascertained that LTM is intact in order for poorer performance on longer sentences to be interpretable as indicating that an STM limitation and a syntactic comprehension disorder are related. In fact, in the cases in which such an interaction appears to be present, LTM has been abnormal, vitiating the conclusion the authors attempt to draw (e.g., TB, discussed later). A second criterion that has been applied to link STM and syntactic comprehension deficits is the effect on the comprehension deficit of unlimited visual presentation of the stimulus sentences. It has been suggested that the disappearance of a syntactic deficit under conditions of unlimited viewing implies a relation between the deficit and an STM limitation (Martin, 1987). This seems reasonable, on the grounds that such a difference in performance must reflect some feature of the modality of presentation and the obvious candidate for this feature is the impersistence of the acoustic record and the consequent inability to maintain a phonological representation because of the STM impairment. However, once again, other aspects of memory (e.g., visual STM, LTM) must be intact for this analysis to be plausible.

14.6. Summary We have presented a framework for analyzing STM disturbances, and have drawn several distinctions in language-processing operations that may be relevant to the role of STM in comprehension. We have also discussed the role of STM in parsing, because this is an area in which considerable data from patients exist, the interpretation of the data is controversial, and establishing a role for STM in parsing would show that a firstpass automatic language comprehension process that does not logically require and cannot directly use lexical and segmental phonological representations nonetheless

STM and comprehension: neuropsychological studies

349

makes use of such representations in STM. We now turn to the neuropsychological data regarding language comprehension in different groups of STM patients to explore the consequences of deficits in STM on language comprehension.

14.7. Case studies We shall now present the data on acoustic-phonetic processing, metalinguistic phonological processing, STM, and language functions in patients with STM impairments, classified into the four groups discussed earlier. We shall interpret the patients' performances as we proceed. As the reader may suspect, we disagree with the analyses of patients' performances presented by the authors of some of these reports. This is especially true in the interpretation of impairments on tests of syntactic comprehension, where we believe some authors are too quick to infer a disturbance in this sphere on the basis of a patient's difficulty in comprehension or repetition of semantically reversible sentences. We state our views of the significance of a patient's performance on particular types of materials as we go through the cases. We believe that when the data are correctly interpreted, there is a generalization that captures the relationship between automatic and controlled aspects of STM and automatic and controlled psycholinguistic functions. We end by presenting this generalization and its broader implications for language processing.

14.7.1. Cases with low-level phonological processing impairments By the criteria outlined earlier we have been able to identify five patients as having lowlevel phonological processing impairments: JB (Warrington, Logue, & Pratt, 1971; Shallice & Butterworth, 1977; Shallice & Warrington, 1977; Allport, 1984a, b; Butterworth, Shallice, & Watson, chapter 8); KC and AL (Allport, 1984a, b); DB (Berndt, 1985); and EA (Friedrich, Glenn, & Marin, 1984; Friedrich, Martin, & Kemper, 1985; Martin, 1987). We also discuss JS (Martin & Caramazza, 1982; Caramazza et al, 1983), even though this patient appears to have a visual rather than an auditory short-term memory impairment, since this case raises interesting questions about the relationship between low-level auditory processing problems, lexical processes, and auditoryverbal short-term memory. We present the relevant data from the six clearest cases in Table 14.1, and comment on these data in the text.1 JB, originally described by Warrington et al. (1971), had a very reduced digit span and good LTM functions. Allport (1984a, b) documented disturbances in JB in making same-different judgments about pairs of nonword syllables (with 1- and 5-sec filled and unfilled interstimulus intervals), and on a lexical decision task in which foils consisted of words in which one consonant was mispronounced so that it differed from the correct pronunciation by either one, two, or three articulatory features. Many authors

Table 14.1. Characteristics of patients with low-level phonological processing impairments JB

KC

AL

DB

EA

JS

_

_ +

_ + +

Spontaneous speech characteristics

Nonfluent Word finding difficulties Phonemic paraphasias

_ —

_ + +

_

+ —

+ +

—

Single word repetition

Impaired word repetition Impaired nonword repetition Phonemic paraphasias

+

+

+

+

+

-* +

+

—

+

+

Confrontation naming

Word finding difficulties Impaired naming from description Phonemic paraphasias

— — +

+

+

+

+

+

+

+

+

Phonemic processing

Impaired phoneme discrimination Natural speech Synthetic speech Impaired phoneme identification Natural speech Synthetic speech Impaired word discrimination Discrimination poor in noise "Tapping task" impaired

+

— + + +

+ -

Auditory lexical access

Impaired auditory lexical decision

+

+

+

-*

+

-

-

-

-

-

Auditory single word comprehension

Impaired picture-word matching Impaired word categorization STA4 characteristics

Auditory-visual discrepancy Reduced span on matching and/or pointing task Span affected by lexical variables Reduced effect of phonological similarity on auditory span Reduced effect of phonological similarity on visual span Reduced effect of word length on auditory span Reduced effect of word length on visual span Reduced recency effect Increased forgetting with auditory presentation and (filled) delay Impaired nonverbal STM LTM

+ + * +

+

+

+ +

+

+ + -

characteristics

Impaired paired associate learning Impaired story recall

+ +

— -

-

+

+

+

+ +

+

—

—

+ +

+ +

+

-

-

Table 14.1. (cont.) JB

KC

AL

DB

EA

JS

-

+

+

+ +

-

+ + +

+

+

+

Impaired word learning Impaired Rivermead Test performance

-

Oral reading (word naming)

Impaired word naming Impaired nonword naming Impaired reading of pseudohomophones Presence of phonemic paraphasias Sentence reading impaired Spelling

Impaired word spelling Impaired nonword spelling

-

+

Written word lexical decision

Impaired written lexical decision Absence of effects of spelling-sound regularity

-

Written word comprehension impaired

-

-

Auditory phonological metalinguistic tasks

Impaired auditory rhyme judgment Impaired auditory homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Sound analysis, blending, mimicry impaired

—

-h

—

Written phonological metalinguistic tasks

Impaired written rhyme judgment Impaired written homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Impaired judgments of pseudohomophones Impaired stress location

-f +

Auditory sentence comprehensionimpaired performance on

Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Grammaticality judgments impaired Parisi-Pizzamiglio Test impaired Recovery from "garden path" impaired

+ +*

-

-

+ +

+ +

+

+

-

+

+

Effects of syntactic complexity

+

Specific parsing impairment

+

+

+

352

Caplan and Waters

Table 14.1. (cont.) JB Sentence repetition impaired

KC

AL

+

DB

EA

+

+

JS +

Discourse comprehension Impaired well-formedness judgments Impaired inferencing Impaired propositional recall Impaired when pragmatics incongruous Written sentence comprehensionimpaired performance on Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Unlimited viewing Grammaticality judgments

+ + +

+

-

+ +

Anagram solution

+

Effects of syntactic complexity Specific parsing impairment WAIS (WISC) impaired Ravens impaired — + + blank *

Indicates feature is not present. Indicates feature is present. Borderline performance or otherwise complex performance. Indicates data not available. See discussion in text.

(e.g., Vallar & Baddeley, 1984a) have argued that JB's poor performance on the phonemic and lexical encoding tasks results from her STM impairment. According to this view, JB's poor performance on the phoneme discrimination test may have been largely due to the fact that there was a 1-5-sec interval between the presentation of the stimulus items. However, this account would predict that JB's performance on the phoneme discrimination task should have been much poorer in the 5-sec than in the 1sec delay condition, which was not the case. JB's performance on tests of STM is very similar to that of patients with PS impairments. There was a large discrepancy between her performance on tests of

STM and comprehension: neuropsychological studies

353

auditory and visual short-term memory, with her span for auditorily presented material being approximately three items (Warrington et al., 1971). In contrast, her memory for auditorily presented nonverbal material, such as environmental sounds, was normal (Shallice & Warrington, 1974). She showed a reduced recency effect and increased forgetting with auditory as compared to visual presentation with a filled delay in the Brown—Peterson paradigm when tested on lists that were subspan (two items; Warrington et al., 1971). In the auditory modality her span was affected by lexical-semantic variables such as frequency and imageability but not word length or syntactic category (Allport, 1984a). As with normals, JB recalled more words when they were embedded in sentences than when they were presented in lists (Butterworth et al., chapter 8). JB's reduced span did not seem to be due to a speech output disturbance, since her performance on tests of confrontation naming and naming from description was excellent (Warrington et al., 1971). In addition, her spontaneous speech was normal in terms of number of words, pause time per word, and paraphasic errors (Shallice & Butterworth, 1977), but showed a high incidence of phonemic paraphasias when the items to be named or repeated were abstract or low in frequency. JB's comprehension seemed spared for the auditorily presented words on which lexical decision was impaired, as tested on a word—picture matching task. Lexical access and comprehension of written words was not tested. JB scored in the normal ranges on tests of single word oral reading and spelling (Schonnell word lists) (Warrington et al., 1971). Sentence repetition was tested with the materials from Saffran and Marin (1975). JB was able to repeat correctly only 3/20 sentences; however, her incorrect repetitions tended to maintain the meaning of the presented sentence. Her errors mainly consisted of omissions, function word errors, and content word errors (Shallice & Butterworth, 1977). Further testing showed that her ability to recall the gist of these sentences was not affected by a filled delay, although her ability to retain the exact wording of the sentences was affected (Butterworth et al., chapter 8). JB performed within the normal range on a grammaticality judgment task (Butterworth et al., chapter 8). However, she had difficulty with the repetition of garden path sentences (e.g., The horse raced past the barn fell).

Performance on the Token Test was below normal (11/15) (Warrington et al., 1971). The Token Test requires the manipulation of geometrically shaped colored objects in response to verbal commands. Its final section includes a variety of syntactic structures, and frequently requires that subjects accomplish actions in the reverse order from that in which they are spoken. The early sections of the Token Test are "semantically reversible" in the sense that the attributes syntactically related to a noun in a given command can be predicated of any item in the array. Thus, failure on the early sections of the Token Test may reflect a subject's inability to adjudicate between a syntactically derived and an erroneous lexicopragmatically derived reading of the command. It may also reflect a subject's inability to retain the encoded semantic representation

354

Caplan and Waters

corresponding to each modifier-noun sequence, especially when the commands involve two adjective adjective-noun sequences. Although failure on the sixth section of this test may indicate comprehension failure, it also may indicate an inability to retain a proposition in memory long enough to execute another action before the action specified in that proposition is accomplished. Thus, there are many reasons for failure on the Token Test other than a disturbance in syntactic comprehension, or even a disturbance in comprehension. If poor performance on the Token Test does reflect comprehension failure, the specific aspects of comprehension that are impaired cannot be delineated because of the heterogeneity of structures on the test and the small number of examples of each structure. Allport (1984a, b) has described two additional patients (AL and KC) whose STM impairments appear to be secondary to disturbances in phonological processing. These patients performed extremely poorly on the phonemic encoding and lexical encoding tests outlined earlier. Both subjects had reduced oral matching spans of three. The language comprehension abilities of these patients were not tested in depth, but both patients showed evidence of well-preserved single word comprehension on the Peabody Picture Vocabulary Test (Dunn & Dunn, 1981) (AL, 130/150) and the auditory single word comprehension test used with JS. DB's (Berndt, 1985) auditory processing deficit was evident on two same-different judgment tasks — a phoneme discrimination task {2/30) and the Wepman Auditory Discrimination Test, a test of word discrimination (4/40). His digit span was two and was influenced by part of speech (more difficulty with function words than nouns) - a somewhat different pattern from that of the patients outlined previously. DB's auditory single word comprehension was good as assessed by the Peabody Picture Vocabulary Test (87/104) and an experimental picture-word matching task (119/120). Unlike the other patients with auditory processing problems, DB performed extremely well on an auditory lexical decision task that Berndt (1985) argues was much easier than the one on which JB, KC, and AL were tested. Written word lexical decision was not tested; however, single word oral reading and spelling were very good. In contrast, nonword oral reading and spelling were very poor, suggesting a problem in the use of grapheme-phoneme correspondences. Sentence comprehension was tested with a forced-choice sentence-picture matching test in which a variety of syntactic constructions were tested. DB performed very well on nonreversible active and passive sentences and on reversible active sentences, but quite poorly on reversible passive sentences. Passives were also harder than actives with dative verbs. These data indicate a syntactic comprehension disorder. It is not clear from the data, however, that such a deficit is related to DB's impaired STM, since the deficit occurred with sentences of varying lengths (transitive and dative passives). EA (Friedrich et al., 1984, 1985; Martin, 1987) showed poor discrimination of stop consonants but not vowels on a phoneme discrimination task. EA did not show

STM. and comprehension: neuropsychological studies

355

evidence of categorical perception of synthetic stimuli - she consistently labeled the items at the extremes of the continuum but did not categorize items in the middle range of VOT consistently. She also experienced difficulty with two auditory phonological metalinguistic tasks. In one task she was required to identify the beginning, middle, or end sound in a nonword string. In the other (Lindamood Auditory Conceptualization Task) she was required to represent phonemic sequences with colored blocks. Auditory homophone judgments were perfect. As with other STM patients, EA showed poorer recall for auditorily than for visually presented stimuli, a lack of a recency effect, and effects of lexical familiarity on span. She showed an effect of phonological similarity with auditory but not visual presentation. The effect of word length on span was not tested in the visual modality, but in the auditory modality there was no difference in her memory for one- versus two-syllable words. Auditory single word comprehension and lexical access were not assessed, but single written word comprehension was excellent. Oral reading of single words was unimpaired, although reading of text was slow and inaccurate. Oral and written spelling were poor, especially for low-frequency words. EA made 25% errors on reversible locative sentences and 33% errors on reversible passive sentences with auditory presentation and 12% and 19% errors on these sentence types with visual presentation, when tested in 1985. When tested 2 years later, she performed at an overall level of 88% on these sentences, apparently with no difference between active and passive sentences or as a function of modality (Martin, 1987, note 4). This pattern suggests that the original deficit was largely due to a parsing disturbance, unrelated to STM functions, which improved. EA performs abnormally on subordinate passive and object-relativized relative clauses in subject position. On the assumptions that thematic roles are assigned as soon as possible and that the noncanonical passive form is harder to process than the active, the relative difficulty of the relative clause sentences correlates perfectly with the processing load imposed by the processing of the relative clause. As noted earlier, performance with unlimited visual presentation or the documentation of increased effects of these syntactic structures in longer sentences would be helpful in helping to establish a relationship between this syntactic comprehension impairment and EA's STM limitations. JS (Martin & Caramazza, 1982; Caramazza et al., 1983) is a patient whose acoustic-phonetic processing is so impaired that it affects his ability to recognize and comprehend auditorily presented words. His auditory STM functions are not reported, presumably because they are virtually nonexistent. JS was not able to identify natural or synthetic speech sounds or to discriminate synthetic speech sounds along a VOT continuum. Interestingly, he had no difficulty on consonant and vowel discrimination tasks with natural speech, raising the question of how he compensated for his acoustic-phonetic disturbances. The data suggests that he had difficulty abstracting

3>56 Caplan and Waters phonetic information from acoustic signals. JS also had difficulty with metalinguistic tasks that required the manipulation of phonological form such as judgments about whether the names of two pictures rhyme. JS's span for visually presented material was severely reduced and affected by part of speech. Several converging pieces of evidence suggested that JS did not use a phonological code to hold the visually presented words in memory. Unlike normals, JS showed a reaction time advantage for physical over name matches at both short and long delays on Posner and Keele's (1967) test of name and physical matches for letters; on a probe recognition memory test JS showed effects of visual and semantic distractors but unlike normals did not show effects of phonemic distractors; and memory for serial order information, which has been shown to rely on the use of phonological codes, was extremely poor. Single word auditory comprehension was not assessed formally but auditory-lexical decision performance was very poor. Written word comprehension was good as assessed by a synonymy matching test, a category judgment test, and by the Peabody Picture Vocabulary Test when the items were presented in print (127/150). Lexical decision accuracy for printed words was excellent, and reaction times were slow but in the normal range. However, JS showed a severe impairment in the ability to abstract phonology from print. Oral reading of words and nonwords was extremely poor. Rhyme judgments for pairs of written words were extremely poor, as was performance on a picture-pseudohomophone judgment task. JS scored 11/31 on the Token Test, the significance of which was discussed earlier. He did well in rejecting semantic foils for locative sentences and in a grammaticality judgment task, except for detection of syntactic errors. He scored 16/24 in picking syntactic foils for semantically reversible locatives with unlimited visual presentation. He solved anagram tasks for active sentences but not for passive sentences even when they were semantically irreversible. The authors emphasize his difficulties detecting "word order" violations, constructing passive sentences, and interpreting semantically reversible sentences. It is clear that tasks such as anagram construction need not bear on comprehension failure. It is unclear whether to attribute his comprehension failure with unlimited visual presentation to a separate parsing impairment, an impairment of visual STM, or the nonuse of phonological representations in auditory—verbal STM. In summary, patients with low-level phonological processing impairments show a fairly consistent pattern of performance in terms of the characteristics of their shortterm memory performance and their language comprehension abilities. Although the performance of these subjects on STM tasks has not been tested in detail, the available data suggest that the short-term memory impairments seen in these patients are similar to those of patients whose memory impairments are attributable to a disturbance of the PS. In general, these patients show a discrepancy in their memory span for auditorily and visually presented items, with a markedly reduced span with auditory presentation,

STM and comprehension: neuropsychological studies

357

and effects of lexical variables, such as frequency, on auditory span. This latter finding suggests that information is being recalled from LTS or some store other than STS. Those few patients who have been tested on the relevant tasks provide evidence that their reduced memory span is associated with an inability to use phonologically based codes in STM. One might expect that low-level phonological processing impairments would affect lexical access and single word comprehension. The majority of these patients performed extremely poorly on tests of auditory-lexical decision. Despite their poor performance on lexical decision, some of these patients showed excellent single word comprehension when comprehension was tested with a picture-pointing task. This finding is consistent with Blumstein, Baker, and Goodglass's (1977) report that phoneme discrimination disturbances do not correlate well with auditory single word comprehension decrements; however, they are inconsistent with other reports linking pure word deafness with low-level phonological processing disturbances (Saffran et al., 1976). The relationship between low-level phonological processing disturbances, lexical access, and semantic access remains largely mysterious. Only one patient (JS) has been tested on written word lexical decision, and his performance was unimpaired. Written word comprehension was also excellent in this patient. However, no consistent pattern emerges in terms of patients' ability to read single words aloud. Some patients are unimpaired, others can read words but not nonwords, and others are impaired on both word and nonword reading. Patients with impairments in phonological decoding might also be expected to have disturbances in sentence comprehension if such disturbances affect lexical access and thus make for incorrect word recognition. One would expect that a phonological decoding deficit that led to impairments of lexical access would lead to a syntactic comprehension deficit when syntactically crucial lexical items were improperly recognized, but whether a disturbance of phonological decoding that does not lead to incorrect lexical identification ever leads to a disturbance of syntactic comprehension remains unclear at this time. In addition, whenever decoding deficits slow down the rate of word identification to a point that words in sentences are not identified as they occur in the speech stream, prelexical segmental phonological values and even representations of acoustic features of utterances have to be retained in memory to allow lexical access to occur after a word has been "passed by." This retention of acoustic and lexical phonological features of utterances would be expected to occupy language-processing resources, and could well lead to impairments in assigning and interpreting syntactic structures, which might require processing resources no longer available because of the increased demands made by the impaired lexical access process. Two patients have evidence for a syntactic comprehension problem (DB and EA), but in neither case are the criteria for linking this disturbance to the patient's STM impairment met (DB does not show length effects; EA was not tested with unlimited visual presentation, which, in her

35 8

Caplan and Waters

case, could indicate a parsing problem). Thus the role of an impairment in phonological decoding in creating a disturbance of syntactic comprehension is unproven on the basis of these cases.

14.7.2. Cases with disturbances of the PS Using the criteria outlined earlier, and excluding the cases that fall into Groups 1 and 4, the following 11 patients have PS disturbances: KF (Warrington & Shallice, 1969,1972; Warrington et al., 1971; Shallice & Warrington, 1970, 1977; Shallice & Butterworth, 1977); WH (Warrington et al., 1971; Shallice & Warrington, 1977); IL (Saffran & Marin, 1975); MC (Caramazza, Basili, Roller, & Berndt, 1981); TB (Baddeley et al, 1987); ER (Vallar, Basso, & Bottini, chapter 17); TI (Saffran, 1985; Saffran & Martin, chapter 16); GI (Martin, Jerger, & Breedin, 1987; Martin, chapter 15); RAN and NHA (McCarthy & Warrington, 1987a, b; chapter 7); and PV (Basso, Spinnler, Vallar, & Zanobio, 1982; Vallar & Baddeley, 1984a,b, 1987). The criteria for assigning PV a deficit in the PS are complex and bear on the analysis of BO (Hildebrandt et al., submitted); we will discuss them separately. Other cases (JT and JO: Kinsbourne, 1972; some of the cases in Tzortis & Albert, 1974, and in Martin, 1987; LS: Strub & Gardner, 1974, and possibly cases from the earlier literature) may be relevant but not enough is known about their language performances to warrant our discussing them here. We summarize the language and memory impairments in these eleven patients in Table 14.2 and comment on these data in what follows. KF remains perhaps the prototypical STM patient, fulfilling all of the criteria for an STM impairment for which he was tested. Although data on KF's memory span for phonologically similar and dissimilar strings are not reported, analysis of his errors on span tasks shows an increase in acoustic confusions with auditory but not visual presentation. KF does poorly on several language tests: oral reading, spelling, and the Token Test. In oral reading, the majority of KF's errors were visual, and phonemic errors did not occur (Shallice & Warrington, 1975). This result, as well as the finding that he was unable to sound out even a single letter, led to the claim that he is a phonological dyslexic (Shallice & Warrington, 1980). No data are available on KF's ability to perform metalinguistic tasks involving phonology, except a "tapping test" on which he performed well. The authors take this as evidence for intact low-level phonemic processing; it may also indicate some preserved metalinguistic processing capacity. KF's poor performance on the Token Test has been noted by many authors but cannot be interpreted unequivocally for the reasons presented earlier. WH (Warrington et al., 1971; Shallice & Warrington, 1977) performs very similarly to KF, with the exception that his reading and writing are somewhat better than those of KF. IL (Saffran & Marin, 1975) is also similar to KF with respect to span reduction and the

STM and comprehension: neuropsychological studies

359

lack of recency effect. His span was larger for familiar material, such as digits and highfrequency words, than for less familiar material, such as low-frequency words and nonwords, indicating reliance on LTM in these tasks. IL was not tested on a sentence comprehension test directly. Rather, he was asked to repeat sentences, and it was observed that he frequently produced a semantic paraphrase of a sentence rather than a verbatim repetition. Because of this feature of his response on this task, the authors used his performance in repetition as an index of his comprehension. The authors point out that IL's responses sometimes reflected syntactic ambiguities in the presented sentences. For instance, he paraphrased a sentence that mentioned sailors and "amusing girls" by saying that the sailors had fun with the lively girls. They also note that IL repeated 6 of 10 passive sentences correctly, reversed the voice and thematic roles in 2, and had incomplete and unscoreable responses in the remaining 2 sentences. The authors interpret this performance as consistent with IL's having a deficit in interpreting passive constructions, and they argue that patients with short-term memory impairments have trouble with interpretation of complex (or "tortuous") syntactic structures. The data supporting the hypothesis that a disturbance of short-term auditory-verbal memory impairs IL's ability to interpret complex syntactic structures, especially noncanonical word orders, are weak. IL showed considerable sensitivity to syntactic structure, including appreciation of lexical category ambiguities. He made frequent errors only under the special circumstance when the passive-voice main clause of a sentence was preceded by a subordinate clause or a prepositional phrase of considerable length that was semantically unrelated to the main clause. Since IL's task was verbatim sentence repetition, and these lengthy sentences were considerably beyond his shortterm memory span, the fact that he failed to repeat passive clauses correctly when his memory abilities were taxed in this fashion does not indicate that he could not interpret syntactic structures when this difficult concurrent task was not imposed. Saffran (personal communication, 1988) suggests that IL may be showing an effect of length on syntactic comprehension, but the task used involved verbatim repetition, making it impossible to attribute the effect to comprehension per se. MC (Caramazza et al., 1981) performed very similarly to KF and IL on a variety of memory tasks (auditory span of one, lack of recency effect, span affected by meaningfulness of material). Probe recognition was below normal for four-item lists, with a recency effect of one item. Brown-Peterson forgetting was greater than normal by 12 sec, but not as marked as in other cases. However, unlike other cases, MC also had a very reduced span (two to three) with visual presentation. MC repeated 9/10 words on the Boston Diagnostic Aphasia Exam (BDAE) normally, and auditory single word comprehension was good, suggesting no serious impairment in automatic phonological processing of auditory stimuli. He was able to read 30/30 words and 10/10 sentences correctly, but only 60% of nonwords. He recognized 50%

^

Table 14.2. Characteristics of patients with disturbances of the phonological store

8

= KF

WH

IL

MC

PV

l

_ + +

_ + +

_

—

-

-

TB

ER

TI

GI

RAN

NHA

Spontaneous speech characteristics

Nonfluent Word finding difficulties Phonemic paraphasias

_ + -h

_ +

_ + +

_ _ — +

_

_ ± +

_ -

± —

+

+ + —

Single word repetition

Impaired word repetition Impaired nonword repetition Phonemic paraphasias Confrontation

-

-

+

+

+ i + —

+ +

-

+ i

+ —

naming

Word finding difficulties Impaired naming from description Phonemic paraphasias

+ +

+

-

-h

-

Phonemic processing

Impaired phoneme discrimination Natural speech Synthetic speech Impaired phoneme identification Natural speech Synthetic speech Impaired word discrimination Discrimination poor in noise "Tapping task" impaired Auditory

-

-

+

-

— -

-

-h -

-

lexical access

Impaired auditory lexical decision

-h

Auditory single word comprehension

Impaired picture-word matching Impaired word categorization

~ ~

+

— +

-

STM characteristics

u» M

Auditory—visual discrepancy Reduced span on matching and/or pointing task Span affected by lexical variables Reduced effect of phonological similarity on auditory span Reduced effect of phonological similarity on visual span Reduced effect of word length on auditory span Reduced effect of word length on visual span Reduced recency effect Increased forgetting with auditory presentation and (filled) delay Impaired nonverbal STM

+ + +

+ + +

+

+

+

+ + +

* * +

+

+ —

+

— -

— -

+ +

LTM characteristics

Impaired Impaired Impaired Impaired

paired associate learning story recall word learning Rivermead Test performance

Oral reading (word naming)

Impaired word naming Impaired nonword naming Impaired reading of pseudohomophones Presence of phonemic paraphasias Sentence reading impaired

+ +

—

— + + —

Spelling

Impaired word spelling Impaired nonword spelling Written word lexical decision

Impaired written lexical decision Absence of effects of spelling-sound regularity Written word comprehension impaired

+

+

+

Table 14.2. (cont.) KF Auditory phonological metalinguistic

WH

IL

MC

PV

TB

ER

tasks

Impaired auditory rhyme judgment Impaired auditory homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Sound analysis, blending, mimicry impaired Written phonological metalinguistic

tasks

Impaired written rhyme judgment Impaired written homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Impaired judgments of pseudohomophones Impaired stress location Auditory sentence performance on

comprehension-impaired

Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Grammatically judgments impaired Parisi-Pizzamiglio Test impaired Recovery from "garden path" impaired Effects of syntactic Specific parsing

complexity

impairment

+*

±*

TI

GI

RAN

NHA

Sentence repetition impaired Discourse Impaired Impaired Impaired Impaired

+*

+

comprehension well-formedness judgments inferencing propositional recall when pragmatics incongruous

+

+

—* —* +*

Written sentence comprehension—impaired performance on Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Unlimited viewing Grammaticality judgments Anagram solution

—

+ + + +

+*

+ + +*

+ +

Effects of syntactic complexity Specific parsing impairment WAIS (WISC) impaired

±

Ravens impaired

+

+ + blank *

Indicates feature is not present. Indicates feature is present. Borderline performance or otherwise complex performance. Indicates data not available. See discussion in text.

—

—

364

Caplan and Waters

of pseudohomophones as such. These data do not permit clear determination of the presence of automatic phonological effects in reading, but indicate that he can extract some phonology from print, although not as well as normals. MC was tested on a sentence—picture matching test using semantically reversible, syntactically complex sentences, and failed to choose the correct picture over lexical and syntactic foils on tests of active, passive, and center-embedded clauses with auditorily presented sentences. There are no data regarding his performance on these different sentence types; thus we cannot conclude that he had a syntactic comprehension disturbance (no documented complexity or specific parsing effects). MC did, however, show a decline in his overall error rate from about 50% with auditory presentation to about 25% with visual presentation, and almost all his errors with visual presentation were syntactic. The increased number of errors on these sentences with auditory presentation was interpreted by the authors as an indication that auditory-verbal short-term memory is required for auditory sentence comprehension when interpretation of syntactic form is necessary. However, this is an overinterpretation of the data. MC's increased impairment with auditorily presented sentences came about not because of increased selection of syntactic foils but because of an increased selection of lexical foils. MC had evolved from an initial stage of severe word deafness, and it is quite possible that his poor performance with auditory presentation reflected some degree of residual word deafness. The magnitude of the increase of selection of lexical foils from visual to auditory presentation is almost the same for syntactically complex, semantically reversible sentences as for simple single word-picture matching. If MC has a syntactic comprehension deficit, the auditory-visual performance difference does not indicate that it is related to auditory-verbal STM limitations. PV (Basso et al, 1982; Vallar & Baddeley, 1984a, b, 1987) has an auditory span of two and a visual span of over three; there was no recency effect with either auditory or visual free recall; span did not improve with pointing; and she showed rapid forgetting of one auditory item after a 3-sec filled delay and roughly normal rates of forgetting from a poor immediate recall baseline of two and three auditorily presented items after 3-, 9-, and 15-sec unfilled delays. PV showed no effect on her span of word length and a normal effect of phonological similarity on span with auditorily presented material, but not with visually presented material. Articulatory suppression had no effect on visual span performances. Despite this pattern, which is apparently more consistent with a disorder of the articulatory loop than of the PS, the authors argue that PV has a reduced capacity of the PS. They base their analysis on (a) PV's rapid forgetting rate; (b) the fact the PV's span was more reduced than that seen in normals under articulatory suppression; (c) the discrepancy between PV's auditory and visual spans, which they argue cannot be accounted for by an impairment in the articulatory loop; and (d) PV's normal articulatory rate, which correlates with span in normals and is related to the operation of the articulatory loop. Vallar and Baddeley thus suggest that the

STM and comprehension: neuropsychological studies

365

impairment in PV lies in the PS and is such as to render the operation of the articulatory loop ineffective. PV is good at phoneme discrimination in nonsense CVs, in placing stress on written words, and in picture—picture, word—word, and word—picture rhyme judgment tasks. These tests suggest that both automatic and some controlled processing involving phonological representations are good. Vallar and Baddeley (1984b) report that PV showed evidence of retained ability to assign and interpret syntactic structure, performing within normal limits on a test of syntactic comprehension devised by Parisi and Pizzamiglid (1970). This battery tests comprehension of passive sentences, subject-verb agreement, adjective-noun agreement, and other syntactic features of Italian, in sentences whose length greatly exceeded PV's span. PV also did well on several other specially designed tests of sentence comprehension. She did fail on one experimental condition, in which she was required to judge whether sentences such as (2) were true: 2. The world divides the equator into two hemispheres, the northern and the southern.

Vallar and Baddeley (1987) present further data on auditory comprehension in PV. She is excellent in syntactic comprehension, verification of lexical-semantic anomalies, and grammaticality judgments in short sentences. She is near perfect on grammaticality judgments in longer sentences in which the error-producing segments are separated by more words than her span. She is able to recall information and make inferences in passages of several sentences. PV makes a few errors in detecting anomalies in sentences when different error types (number and gender agreement, word order inversion, semantic) are presented in mixed lists. She does worse in detection of number and gender anomalies when they are separated by several sentences in a passage; here, too, her error rates are often close to normal. She makes no errors on permuted word anomalies in these passages. Vallar and Baddeley (1984b, 1987) discuss these performances. We have major disagreements with their interpretations of their findings. Vallar and Baddeley (1984b) interpret the first of these abnormal performances — PV's failure on sentences like (2) — as evidence for a disturbance of syntactic comprehension; specifically an inability to interpret "word order," which they idiosyncratically use to refer to inversion of two nouns within a sentence to yield a semantically false or anomalous reading. However, as discussed earlier, the deficit may lie at the stage of processing at which PV must choose between the syntactically and lexicopragmatically derived meanings of sentences like (2). This would constitute a disturbance that arises at a postinterpretive stage of processing. This analysis is consistent with PV's good performances on other tests of syntactic comprehension. Vallar and Baddeley (1987) discuss all three abnormalities in PV's sentence comprehension, noting that she is not normal (a) on sentences like (2) with syntactically

366

Caplan and Waters

encoded semantic anomalies, (b) in grammatical judgments in passages of prose, and (c) in grammatical and semantic judgments within sentences in a mixed presentation condition. They conclude that the PS is needed "(1) when interpretation of lexical semantic and syntactic structure is needed for comprehension; and (2) when a relatively complete analysis of multiple aspects of the material is needed" (p. 435). Condition 1 accounts for PV's poor performance on (a) and (b), and Condition 2 accounts for her performance on (c). We find these suggestions vague and are not sure that we understand the authors' intented analyses. We believe that, with respect to Condition 1, Vallar and Baddeley are arguing that some aspect of on-line, first-pass sentence and discourse interpretation involves combining semantic and syntactic information, and that this aspect of sentence processing is disturbed in PV. We do not agree that the data support such an interpretation. We presented our views regarding PV's sentence anomaly judgment performance earlier in this chapter. Ascribing this abnormal performance to a disturbance that arises in adjudicating between syntactically and lexicopragmatically derived meanings of a sentence is one possible interpretation of Vallar and Baddeley's first condition, but we doubt it is one that they intend, since it does not posit that PV's deficit is an aspect of online, first-pass sentence comprehension. It is our hypothesis that PV has trouble with a postinterpretive process that involves the PS. Next, consider PV's second abnormal performance - in detecting anomalies based on gender and number in passages. An example passage from Vallar and Baddeley (1987) is the following: [The businessman from Milan was travelling in his luxurious car. The car was very speedy and silent. Fortunately, that day the motorway was empty. *They/He had to arrive in Rome... ] In this passage, they is a semantically legitimate, but discourse-inappropriate, word (one can invent an antecedent for they). PV's failure to reject these passages might indicate that she does not keep discourse structure in mind, that she does not recognize discourse anomalies, or that she is not prepared to base anomaly judgments on discourse incongruities. The first possibility is rendered extremely unlikely by her good performance in recalling and drawing inferences from discourse. The second account is more likely, and raises interesting possibilities regarding the role of phonological representations in discourse processing. Recognition of discourse incongruity cannot require the maintenance of a phonological representation of all the items in a preceding passage in the PS, since it is clear that verbatim retention of sentences falls off sharply even at clause, let alone sentence, boundaries (Jarvella, 1970, 1971; Caplan, 1972). Other possibilities come to mind, in particular, the possibility that discourse-focused nouns may be maintained in phonological form (along with selected other items; see later discussion of BO below), and referred to in cases of discourse-incongruent reference. If

STM and comprehension: neuropsychological studies

367

so, PV's abnormal performance on these passages has an interesting resemblance to her failure on sentences like (2): Both occur when an initially assigned reading must be reconsidered. This pattern is therefore consistent with PV's having a postinterpretive processing disturbance. Unfortunately, we cannot take PV's failure to perform normally on the discourse grammaticality task as clear evidence for such a deficit because the third possible account of her performance - that it reflects a failure to base her judgments on such anomalies - must be considered seriously: PV also adopts an unusual decision criterion for acceptability that leads to many false positives in a grammaticality judgment task, as we shall discuss immediately. Finally, consider PV's third performance - (c) above: poor performance on grammatical and semantic judgments within sentences in a mixed presentation condition. First, we note that PV has a very minor deficit on this task, if indeed she has a deficit at all. PV has a response bias (22/98 false alarms) and makes only 8/98 errors. Her error rates are all within two items of normal performance on this task and her overall A' score (Grier, 1971) is .91, hardly enough to warrant imputing a major role to the PS in the accomplishment of this task. PV's performance on this task does, however, suggest that she has trouble adjusting her criteria for acceptance of a sentence when anomalies can be of several sorts, and adopts a strategy of accepting incorrect sentences under these conditions: The most striking feature of her performance on this test is her high false alarm rate. Her trouble setting criteria for well-formedness is of significance in interpreting her performance in judging well-formedness of discourse, as discussed earlier. Even setting aside this disagreement about the significance of PV's error rate, poor performance on a grammaticality judgment task need not reflect poor parsing ability but may simply reflect an inability to deal normally with the metalinguistic aspects of the task (Linebarger, Schwartz, & Saffran, 1983; Zurif & Grodzinsky, 1983). Overall, PV shows interesting performances that suggest she is having problems with postinterpretive aspects of sentence and discourse processing. The evidence for this analysis is not strong, however, because other factors — such as idiosyncratic setting of criteria for well-formedness on judgment tasks — may also have led to impaired performances. However, there is no evidence for a disturbance of first-pass automatic aspects of sentence or discourse processing in PV's performances.2 TB's (Baddeley et al., 1987) STM impairment is less pure than that of the cases described earlier, but quite severe. TB has a moderate LTM impairment but a very severe STM impairment. He has a digit span of two items with both auditory and visual presentation, a sentence span of three items, no recency effect in free recall, and he shows no effect of phonological similarity or word length on memory span for either auditorily or visually presented items. TB's single word repetition is reported to be unimpaired (Baddeley, personal communication), but he failed on two metalinguistic tasks - one in which he was required to produce spoonerisms and another in which he was required to form words from the initial sounds of three words.

368

Caplan and Waters

TB's performance on several subsections of the Test for the Reception of Grammar (TROG; Bishop, 1982) is not good. Baddeley et al. (1987) try to distinguish length from syntactic complexity effects in this performance. They claim that the sentence types on which TB had difficulty are those that require holding several concepts simultaneously in memory; they specifically deny that semantic reversibility of a sentence is the factor that determine complexity for TB. However, Baddeley et al. misanalyze their data regarding TB. All the sentence types on which TB does poorly are reversible. The two sentence types that Baddeley et al. claim are not reversible on which TB does poorly are of the form Not only the bird but also the flower is yellow and Neither the dog nor the ball is brown. Both these sentence types require interpretation of the syntactic structure to be understood correctly, since they are ambiguous on purely lexicopragmatic grounds. In fact, TB shows a syntactic complexity effect in error rates with auditory presentation. He achieves a good score but has very long reaction times for some sentences with unlimited visual presentation; these reaction time (RT) data show some effects of syntactic complexity. Baddeley et al. (1987) conclude that STM is important in maintaining lexical items in serial order in a "mnemonic window" that "facilitates the semantic processing of the material [and] which might be assumed to involve constructing some form of mental model or representation (cf. Johnson-Laird, 1983)." Again, we cannot comment directly on this vague hypothesis. Our analysis is that, since there are syntactic complexity effects, TB has a syntactic comprehension disorder. The interaction of reversibility, complexity, and length effects with auditory presentation cannot be interpreted as evidence for a relationship between this disorder and the STM impairment, because of TB's ppor LTM, as discussed earlier. Despite his accurate performance, TB's RT data with unlimited visual presentation indicate that he has trouble with syntactic comprehension for visually presented sentences. This cannot be interpreted as evidence for an independent parsing disturbance because of his LTM deficit, but it provides no evidence for any relation of his syntactic comprehension deficit and his auditory-verbal STM disturbance. ER (Vallar et al., chapter 17) showed good auditory and written single word comprehension and good oral reading but poor repetition. She scored 49/60 on phoneme discrimination. This score suggests some impairment but is better than that of the patients whom we have classified as having low-level phonological processing impairments. Performance on metalinguistic tasks involving phonological representations was good: ER could read words and nonwords, place stress on regular and irregular words, and do rhyme matches for written words, pictures, and word—picture pairs. Span was two items auditorily and three or more visually: imageability and, to a lesser extent, word frequency affected span. There was an auditory but not a visual phonological similarity effect and no word length effect with either modality of list presentation. In apparently direct contradiction to their analysis of PV, the authors

STM and comprehension: neuropsychological studies

369

claim that ER had a disturbance of both the PS and the AL (as opposed to the strategic nonuse of the AL), as well as a low-level phonological processing deficit. ER scored 78/80 on the Parisi and Pizzamiglio (1970) test of syntactic comprehension but poorly on semantically reversible active, passive, dative, and relative clause sentences on a sentence—picture matching task with three distractors representing the nouns in the sentences in various relationships (one a thematic role reversal). The stimuli for this task consisted of both short and long sentences. However, since the data are not presented separately for these two types of stimuli, no direct assessment of the effect of length on performance can be made. She also did poorly in detecting syntactically encoded semantic anomalies, especially in longer sentences. The authors repeat their analysis of PV in relation to this case. We have extensively discussed the significance of nondetection of syntactically encoded semantic anomalies, and need only comment that the ability to pick the correct picture of two but not four items clearly indicates a disturbance at a postinterpretive stage of sentence processing. TI (Saffran, 1985; Saffran & Martin, this volume, chapter 16) showed good phoneme discrimination, and perfect single word auditory comprehension and oral reading on the BDAE. He was not tested on metalinguistic phonological tests. Saffran primarily discussed his sentence comprehension performance. TI failed to understand semantically reversible passive and center-embedded relative clauses and did better on passives and locatives, thus providing evidence for a syntactic comprehension disturbance. Saffran argues that aspects of TI's sentence comprehension performance suggest that his disturbances in this task are not secondary to his shortterm memory impairments. For instance, sentence length did not affect TI's performances. TI performed as poorly on "truncated" passives without a by phrase, such as (3), as on full passives: 3. The dog was chased.

Saffran argues that, if TFs short-term memory impairment underlies his syntactic comprehension problem, performance should improve on short sentences that are within his memory span of three digits. TI performed extremely well in the grammaticality judgment task, and showed no effect of the number of words intervening between the critical items determining grammaticality. Saffran (1985) thus concluded that TI could assign syntactic structure, but could not interpret it. She suggested that auditory-verbal short-term memory functions are not involved in assigning syntactic structure, but might be involved in interpreting these structures. GI (Martin et al, 1987; Martin, this volume, chapter 15) was an 11.6-year-old epileptic boy with ictal expressive language disturbances. He had a phonemic discrimination problem that occurred only in noise and an STM disturbance attributed to the PS (no recency or phonemic similarity effects). He showed an impairment in auditory comprehension of semantically reversible sentences with object-relativized

370

Caplan and Waters

and passive but not active relative clauses in subject position, thus indicating a disturbance of syntactic comprehension. He also performed poorly on complements of the verbs tell and promise. Comprehension of these sentences with unlimited visual presentation was good. Martin (1987) attributes this syntactic comprehension problem to GI's PS limitations, arguing that the PS is used by the parser as a lookahead buffer and that more items must be stored in this buffer if earlier items cannot be parsed quickly. However, she now (Martin, chapter 15) is reluctant to attribute this problem to his PS limitations because other cases (she cites RE) with PS disturbances do not have such problems with comparable sentences. She argues that GI's subtle phonemic processing disturbance may have affected lexical access or that he may have an independent syntactic comprehension problem. GI has a disturbance of syntactic comprehension by our criteria, showing clear complexity effects. These effects are easily related to the memory load imposed by the structure in the relative clause, as discussed earlier (case EA). By the aforementioned logic, the fact that GI performed normally with unlimited visual presentation certainly strengthens the case for a relationship between his PS and syntactic comprehension impairments. However, the fact that some PS cases (as well as other cases with equally reduced spans, such as BO — see later discussions) are unimpaired on these sentences proves that the two are not necessarily related. Although we do not wish to claim that the fact that GI is a developmental case necessarily sets him apart (see our later discussion of objections voiced by Vallar & Baddeley, 1987, to the significance of case RE, Butterworth et al, 1986), we suggest that one possible way to attempt to resolve these contradictory data is to explore the possibility that the relationship between the PS or the entire STM system and syntactic comprehension changes during language development. Although normal 11-year-olds performed well on the sentences used by Martin, GI was not a normal child and the more complicated sentences may have been at the limits of his parsing capacities. The PS or the STM system may play a greater role in the developing than in the fully developed parser and may be particularly important in sentences with structures at the limit of the developing parser's capacities. McCarthy and Warrington (1987a, b; this volume, chapter 7) present two STM cases, RAN and NHA with PS disturbances. They are very similar and will be discussed together. RAN showed excellent phonemic discrimination; results are not available for NHA. Auditory span was reduced below visual span, and forgetting for one auditorily presented digit was over 75% in 5 sec. No information is given regarding direct measures of lexical access (such as lexical decision), phonologically mediated effects in written word processing, or metalinguistic phonological tasks. Sentence processing is described in some detail. Both cases did poorly on versions of the Token Test presented auditorily, with the later sections being most difficult. Both showed a disturbance in carrying out a reversible command in an unconventional pragmatic setting (e.g., Put the ball on the top of the ladder, when the ball is

STM and comprehension: neuropsychological studies

371

fixed in the array - Huttenlocher & Strauss, 1968). Neither could respond correctly to simple questions that required them "to utilize spoken information in additional constructive and reconstructive operations," such as Which is red, a poppy or a lettuce? On the other hand, both were able to name objects on the basis of triplets of words (e.g., small, yellow, bird) and both were quite good in comprehending reversible active and passive sentences. Sentence repetition was better than word repetition (though not normal) and was marked by attempts to provide appropriate completions for incomplete sentences. McCarthy and Warrington argue that these patients retained a "dynamic, integrative memory system that underlies sentence comprehension," and that "they are specifically impaired in their recourse to systems which are required for the monitoring and control of language processing when resources based upon anticipatory hypotheses are inadequate." These are all second-pass processes by the criteria we have described. RAN's and NHA's comprehension difficulties on the Token Test are also attributable to such processes, as described earlier. In summary, none of these 11 cases has been tested in enough detail to determine whether automatic aspects of phonological processing of auditorily presented words are entirely normal. Several cases have normal or near-normal auditory single word comprehension (MC, ER, TI). Several can read single words aloud normally (MC, ER, TI), and at least one shows good nonword reading (ER). These results suggest that automatic processing of phonological representations of both auditory and written single words can occur with major disturbances of the PS. The data from patients with PS impairments provide evidence that phonological representations are not involved in parsing. Patients with severe limitations on this memory system have been shown to be capable of comprehending many aspects of syntactic form (PV, RAN, NHA). In many cases, authors have overinterpreted their data, attributing syntactic comprehension deficits to patients without adequate evidence (MC). In many cases, a disturbance in processing syntactic structure appeared only when the memory requirements of the task were increased, such as the requirement for verbatim repetition (IL). When a syntactic comprehension disturbance has been documented, it has not been related to the STM limitation (e.g., TB), except in one developmental case, GI. The discrepancy between GI and the PS cases with good syntactic comprehension can best be explored by considering the role of STM in the developing parser. There are also limited data on aspects of processing beyond the extraction of propositional meaning. McCarthy and Warrington have shown that PS cases can respond normally on tasks in which discourse-relevant features such as the focused item in a proposition are congruent with the pragmatics of a test situation but are impaired when these features are incongruent. We have interpreted the data on PV provided in Vallar and Baddeley (1987) along the same lines. In terms of the distinctions we drew

Table 14.3. Characteristics of patients with disturbances of the articulatory loop MK

RL

BO

— + +

— ± +

+ i +

i

i

i

-

+

Spontaneous speech characteristics

Nonfluent Word finding difficulties Phonemic paraphasias Single word repetition

Impaired word repetition Impaired nonword repetition Phonemic paraphasias Confrontation naming

Word finding difficulties Impaired naming from description Phonemic paraphasias Phonemic processing

Impaired phoneme discrimination Natural speech Synthetic speech Impaired phoneme identification Natural speech Synthetic speech Impaired word discrimination Discrimination poor in noise "Tapping task" impaired

-

-

—

—

+

—

—

—

Auditory lexical access

Impaired auditory lexical decision Auditory single word comprehension

Impaired picture—word matching Impaired word categorization STM characteristics

Auditory-visual discrepancy Reduced span on matching and/or pointing task Span affected by lexical variables Reduced effect of phonological similarity on auditory span Reduced effect of phonological similarity on visual span Reduced effect of word length on auditory span Reduced effect of word length on visual span Reduced recency effect Increased forgetting with auditory presentation and (filled) delay Impaired nonverbal STM LTM characteristics

Impaired Impaired Impaired Impaired 372

paired associate learning story recall word learning Rivermead Test performance

+ i -

i

i

-

-

+ + + +

+ + + +

-f

Table 14.3. (cont.)

Oral reading (word naming) Impaired word naming Impaired nonword naming Impaired reading of pseudohomophones Presence of phonemic paraphasias Sentence reading impaired Spelling Impaired word spelling Impaired nonword spelling Written word lexical decision Impaired written lexical decision Absence of effects of spelling sound regularity

MK

RL

+ + +

+ +

Written phonological metalinguistic tasks Impaired written rhyme judgment Impaired written homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Impaired judgments of pseudohomophones Impaired stress location Auditory sentence comprehension - impaired performance on Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Grammaticality judgments impaired Parisi-Pizzamiglio Test impaired

— ±

+

+ +

+

Written word comprehension impaired Auditory phonological metalinguistic tasks Impaired auditory rhyme judgment Impaired auditory homophone judgment Impaired performance on Spoonerisms task Impaired word formation from sequential initial segments Sound analysis, blending, mimicry impaired

BO

i

~

-

-

+ —

+

4-

+ +

+

—

+

+*

-

+ +" + +

Recovery from "garden path" impaired Effects of syntactic complexity

~

+*

Specific parsing impairment

—

-f * —

Sentence repetition impaired

+

373

-

374

Caplan and Waters

Table 14.3. (cont.) MK Discourse Impaired Impaired Impaired Impaired

RL

BO

comprehension well-formedness judgments inferencing propositional recall when pragmatics incongruous

Written sentence comprehension - impaired performance on Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Unlimited viewing Grammaticality judgments

+ +

— —

+ +* + H-

Anagram solution Effects of syntactic complexity

+

Specific parsing impairment

H~

~~

WAIS (WISC) impaired Raven impaired + + blank *

Indicates feature is not present. Indicates feature is present. Borderline performance or otherwise complex performance, Indicates data not available. See discussion in text.

between first-pass and second-pass language processing, all these results suggest that the PS is not involved in first-pass automatic language processing, but is involved in later, postinterpretive processes subject to control.

14.7.3. Cases with disturbances of the AL Three patients have been analyzed as having disturbances of articulatory-based rehearsal (the AL): MK (Howard and Franklin, 1987; this volume, chapter 12); RL (Caplan et al, 1986a, b); and BO (Hildebrandt et al., submitted). PV may also have an AL impairment, as discussed above. Table 14.3 presents data on these cases. MK is a Wernicke's aphasic with significant disturbances of lexical decision, single word auditory comprehension, and naming. However, elementary phonemic discrimination is good. Howard and Franklin argue that MK has three deficits that make it

SThA and comprehension: neuropsychological studies

375

impossible for him to use the AL: (a) an inability to convert output to input phonological representations; (b) an inability to convert input to output phonological representations directly; and (c) a disrupted auditory input lexicon. We note that these are very different criteria for an AL deficit from those discussed by Vallar and Baddeley (1984a, b) and Hildebrandt et al. (submitted), and do not require nonfluency as a criterion for its diagnosis. MK's STM performance is different from that of the PS cases described earlier. He has an oral and pointing span of one with auditorily presented digits and of three with visually presented digits. However, matching span is between three and five items with auditorily presented digits, far longer than any other case reported thus far. He shows phonological similarity effects in auditory but not visual matching span, but no length effects in either modality. This is consistent with an impairment of the AL. MK is likely impaired in automatic processes involving phonological representations of auditorily presented single words (e.g., poor lexical decision), but this appears to be due to a primary disturbance of lexical phonological representations, as the authors indicate. There is some evidence that his single word reading involves phonological processes insofar as he shows a strong spelling-sound regularity effect in reading aloud with many regularization errors and can read nonwords. On the other hand, some metalinguistic tasks involving phonological representations derived from print are poorly performed, notably pseudohomophone definition and rhyme judgments. MK's sentence comprehension is very poor. He fails on comprehension, grammaticality judgment, and anomaly detection tests with both auditory and written sentence presentation on a wide variety of sentence materials. However, no firm conclusions can be drawn about the relationship of these failures to his AL deficit because his lexical access failure obviously contributes heavily to this problem. The authors leave open the possibility that articulatory rehearsal may play a role in the comprehension of complex written sentences. RL is a reproduction conduction aphasic patient whose repetition span was two to three items, but whose pointing, probe recognition, and serial probe recognition performances were considerably better (roughly similar to those of MK). On the basis of this output disorder, the authors suggest that RL may have had an impairment in the AL; no further evidence for such an impairment is provided. Sentence comprehension was examined in an object manipulation task. RL showed difficulty on subject-object relative sentences (The goat that the horse chased pushed the rabbit) with auditory presentation. With written presentation, RL treated conjoined and relative clauses with relativized subjects as though they consisted entirely of the nonhiearchically organized major lexical category sequence N-V-N-V-N. These data suggested a reliance on intonational cues to recognize and structure relative clauses, and a difficulty maintaining the head of a subject—object relative clause in memory with auditory presentation, which the authors suggested may be related to RL's putative difficulty with rehearsal.

376

Caplan and Waters

BO was diagnosed as mildly dysarthric, dyspraxic, and anomic. Auditory comprehension of single words was intact. Phoneme discrimination, auditory word discrimination, auditory lexical decision, and receptive vocabulary (Peabody Picture Vocabulary Test score of 139/150) were intact. Written lexical decision showed long RTs but was basically similar to that of normals in relationship to spelling-sound regularity effects in low-frequency words, although her latency data are mixed, likely due to a speed-accuracy trade-off. Written synonymy judgments were normal. Pseudohomophone recognition (85%), homophone judgment (79%), and stress identity judgment were poorer than for control subjects. Thus, although the data are not completely clear because of a probable speed—accuracy trade-off in written word lexical decision, BO shows normal automatic phonological effects in word processing but some impairment on metalinguistic tasks that involve single word phonological representations. BO's auditory pointing span is three for digits and high-frequency words and two for low-frequency words: two for both high- and low-frequency words in print. BO shows a phonological similarity effect with auditory but not visual list presentation and no length effects. This pattern is consistent with an AL impairment, an analysis appropriate to her very reduced articulatory rate. This analysis clearly runs counter to that of Vallar and Baddeley in case PV, who showed virtually identical STM performance. However, BO shows much less forgetting in the Brown—Peterson paradigm than PV, and her pointing span is similar to that of normals under conditions of articulatory suppression, indicating that her deficit is not the same as PV's. BO was tested on auditory and written versions of the Token Test and the TROG. She performed extremely well on the latter. On the Token Test, her performance worsened through the penultimate sections (reflecting the increased STM demands of each successive section) but improved on the syntactically more difficult final section (indicating no specific difficulty with syntactic comprehension). BO's auditory and written syntactic comprehension also was tested in considerable detail on the battery devised by Caplan and Hildebrandt (1988). BO showed a negligible effect of syntactic complexity on error rate and no disturbance that can be related to a specific parsing operation, except for some difficulty with subject-object relative clauses with written but not auditory presentation. In contrast to BO's lack of syntactic complexity effects, there is evidence that BO has particular difficulty with sentences containing three proper nouns. This pattern suggests that BO is not having difficulty in parsing a sentence, but in holding proper nouns in memory after a sentence has been parsed. If this difficulty with proper nouns were occurring at a syntactic level, BO would be expected to have much more difficulty achieving the correct parse, and hence would be expected to show effects of syntactic complexity with these items. Therefore, it is hypothesized that she has difficulty maintaining proper nouns with the correct thematic role assignment at the level of the proposition or mixes up transferring the thematic

STM and comprehension: neuropsychological studies

377

roles from arguments in the proposition to actual referents in the array in the process of acting out the sentence. We note that Perfetti and McCutchen (1982) have suggested that proper nouns are maintained in phonological form in sentential semantic representations because they are less semantically specified than common nouns, and that they, like nouns that serve as the foci of discourse representations (see the earlier discussion of PV), may be part of a small set of lexical items held in interpreted structures in phonological form. In summary, these three cases suggest that a disturbance of the AL may impair STM functions with auditory list presentation slightly, but perhaps more than is apparent from the literature on articulatory suppression in normal subjects (which, as we have noted, may not totally eliminate the AL). These impairments are not necessarily associated with disturbances in automatic auditory phonological processing on either the input-side (BO) or the output-side (MK), though an impairment of the link from output to input phonological representations may be required for an AL disturbance in cases where the latter is normal. Automatic processes involving phonological representations in written word recognition seem to be intact with AL disturbances, but various controlled processes have been disrupted. Syntactic comprehension for both auditory and written sentences can be intact (BO), but postinterpretive processes (presumably requiring control) can be affected.3

14.7.4. Cases with disturbances of central phonological processing Several cases (JS, MK, BO) have disturbances affecting their ability to accomplish some metalinguistic tasks that require manipulation of segmental phonological representations. In some cases (e.g., JS), some parts of the patient's STM impairment have been attributed by the authors to this metalinguistic phonological impairment. However, we have discussed these cases elsewhere because a co-occurring disturbance of low-level phonological processing or of the AL may have produced the impairment on metalinguistic phonological tasks. In two cases without such concomitant disturbances - RE (Campbell & Butterworth, 1985; Butterworth et al., 1986) and EDE (Berndt, 1985, where the patient is labeled "EE", and Berndt & Mitchum, this volume, chapter 5) - the STM disturbance may be linked to an inability to manipulate segmental phonological representations. Table 14.4 presents the data on these cases. RE is a developmental phonological dyslexic and dysgraphic with excellent overt language processing. Her phoneme discrimination is normal, and repetition of single multisyllabic words and nonwords is good. However, she shows numerous disturbances in many tasks that require phonological segmentation: reading nonwords, rhyme judgment (oral as well as written), counting phonemes in written words with graphemic bigrams, creating spoonerisms, creating words from the initial phonemes of a series of words, homophone and pseudohomophone detection. In all metalinguistic

Table 14.4. Characteristics of patients with disturbances of central phonological processing RE

EDE

-

+ +

+ -

+

— -

+

-

-

-

-

-

+

-

+

+ + + + + 4+

+ +

Spontaneous speech characteristics

Nonfluent Word finding difficulties Phonemic paraphasias Single word repetition

Impaired word repetition Impaired nonword repetition Phonemic paraphasias Confrontation naming

Word finding difficulties Impaired naming from description Phonemic paraphasias

+

Phonemic processing

Impaired phoneme discrimination Natural speech Synthetic speech Impaired phoneme identification Natural speech Synthetic speech Impaired word discrimination Discrimination poor in noise "Tapping task" impaired

-

Auditory lexical access

Impaired auditory lexical decision Auditory single word comprehension

Impaired picture-word matching Impaired word categorization STM characteristics

Auditory-visual discrepancy Reduced span on matching and/or pointing task Span affected by lexical variables Reduced effect of phonological similarity on auditory span Reduced effect of phonological similarity on visual span Reduced effect of word length on auditory span Reduced effect of word length on visual span Reducing recency effect Increased forgetting with auditory presentation and (filled) delay Impaired nonverbal STM

+ + + + + +

LTM characteristics

Impaired Impaired Impaired Impaired

paired associate learning story recall word learning Rivermead Test performance

+

Oral reading (word naming)

Impaired word naming Impaired nonword naming

+

4-

Table 14.4. (cont.) RE Impaired reading of pseudohomophones Presence of phonemic paraphasias Sentence reading impaired

EDE

+ -

Spelling

Impaired word spelling Impaired nonword spelling

+

Written word lexical decision

Impaired written lexical decision Absence of effects of spelling-sound regularity

— +

Written word comprehension impaired

—

Auditory phonological metalinguistic tasks

Impaired auditory rhyme judgment Impaired auditory homophone judgment Impaired performance on spoonerisms task Impaired word formation from sequential initial segments Sound analysis, blending, mimicry impaired Written phonological metalinguistic

+ + + + -

+

tasks

Impaired written rhyme judgment Impaired written homophone judgment Impaired performance on spoonerisms task Impaired word formation from sequential initial segments Impaired judgments of pseudohomophones Impaired stress location

+ + + + + -

Auditory sentence comprehension - impaired performance on

Short irreversible sentences Long irreversible sentences Short reversible sentences Long reversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Grammaticality judgments impaired Parisi-Pizzamiglio Test impaired Recovery from "garden path" impaired Effects of syntactic Specific parsing

— -

+ + + +*

~~

+

+

complexity

impairment

Sentence repetition impaired Discourse comprehension

Impaired Impaired Impaired Impaired

well-formedness judgments inferencing propositional recall when pragmatics incongruous

~~

380

Caplan and Waters

Table 14.4. (cont.) RE

EDE

Written sentence comprehension - impaired performance on

Short irreversible sentences Long irreversible sentences Short reversible sentences Long irreversible sentences Syntactically encoded semantic anomalies Token Test TROG Test Unlimited viewing Grammaticality judgments Anagram solution Effects of syntactic complexity Specific parsing impairment WAIS (WISC) impaired Raven impaired -f + blank *

Indicates feature is not present. Indicates feature is present. Borderline performance or otherwise complex performance. Indicates data not available. See discussion in text.

orthographic tasks she relied on the written form of the word (e.g., judging lemon and demon to rhyme). RE has a span of four items or less, better with visual than auditory list presentation. There are no recency effects, phonemic similarity effects, or effects of articulatory suppression on span, consistent with a disturbance of the PS. The authors suggest that her reduced span may be related to her "central" "phonological processing" disturbance. RE performed normally on the Token Test and the TROG, in comprehension of reversible locative sentences, in grammaticality judgments of sentences in which the number of words between the anomalous lexical items exceeded her span, and in comprehension of passages. She segmented passages normally for self-paced recall, but performed below normal on verbatim sentence repetition. The authors interpret these findings in relation to a distinction between "passive" and "controlled" short-term store (STS), arguing that "different processes - comprehension and recall - [have] differential access to a phonological representation held in a STS" (Butterworth et al., p.734). They conclude that the PS in not necessary for sentence comprehension.

STM and comprehension: neuropsychological studies

381

Vallar and Baddeley (1987) dismiss this case on two grounds: RE's span is greater than that of other cases and new adaptive mechanisms may have developed in a developmental case without a focal neurological lesion. As evidence of the latter, they point out that, unlike normals, RE fails to show a significant deterioration on Part 4 of the Token Test under conditions of articulatory suppression. This hardly seems adequate evidence for this analysis, especially since RE otherwise behaves normally on this task. It is certainly the case that adaptive mechanisms mediate responses in developmental cases, but they do so as well in acquired disorders (PV, for instance, has had years to develop adaptive heuristic mechanisms). To make this point seriously, Vallar and Baddeley would have to demonstrate the differences between RE's adaptive mechanisms and those of acquired cases. RE's span is not as reduced as many PS cases, and roughly comparable to the AL cases described earlier. We noted in the Introduction that different sources of PS disturbances may limit the PS in different ways and to different degrees. However, RE clearly has a limited PS (perhaps secondary to her phonological processing impairment), and this limited PS is compatible with normal comprehension of sentences and passages. EDE showed excellent phoneme discrimination. She had a large positive bias in auditory lexical decision, much better word than nonword repetition, and "relatively spared" auditory single word comprehension (Peabody Picture Vocabulary Test score of 115). EDE is capable of some phonological processing of print: 46/60 nonwords are read correctly and most errors are lexicalizations (indicating that some phonological information is derived from print). Thus there is evidence for an output-side phonological disorder but better input-side phonological processing. Berndt (Berndt & Mitchum, this volume, chapter 5) suggests that EDE has difficulty maintaining lexicophonological (Monsell's [1984] Stage II) or "morphonolexical" (Barnard, 1985) representations in memory and relies on earlier "acoustic" representations to accomplish some single word and nonword processing tasks. EDE has a better visual than auditory span, shows a small phonological similarity effect with auditory and none with visual lists, and no word length effects. The recency effect is reduced. There is a discrepancy between function word and noun lists. These data are evidence for a PS disturbance. EDE achieved only 59% on the TROG. She shows a clear syntactic complexity effect in sentence—picture matching (passives are harder than actives with both transitive and dative verbs), establishing the existence of a syntactic comprehension disorder, and some evidence for a highly specific length effect: Performance is poorer on locative sentences with an added adjective in the subject, but not object, noun phrase when reversible distractor pictures were used. There is no evidence regarding performance with unlimited viewing of written sentences. The fact that EDE's deficit occurs equally with sentences of varying lengths (transitive and dative passives) may be taken as evidence against a connection between her syntactic comprehension deficit and her

382

Caplan and Waters

STM disturbance. The "length" effect of an adjective in subject but no postverbal PP position with locatives is hard to relate to the functions of the PS, which is responsible for recency but not primacy effects. A reasonable guess is that it is hard for EDE to maintain NPs with adjectival modifiers in a propositional representation. The possibility that it is hard for her to construct, interpret, or transfer them to such a representation is rendered less likely by the fact that she does well on sentences with adjectivally modified NPs in post-verbal PP position. Although a disturbance of the initial transfer of a semantically complete NP to a propositional representation would be an impairment in automatic processing, EDE does not provide good evidence linking the PS with these processes.

14.8. Discussion Our review of the neuropsychological literature shows that significant reductions in STM functions have very limited effects on language processing. There are no instances of clear effects of STM limitations of any sort on on-line automatic processes, even when these processes make use of phonological representations, except in cases in which an STM impairment is itself secondary to a low-level acoustic—phonetic processing disturbance. On the other hand, language processing is not normal in STM patients. The word-level disturbances that have been documented in STM patients all involve metalinguistic tasks that require controlled manipulation of segmental phonological representations. The sentence and discourse comprehension impairments that have been documented in STM cases can all be attributed to an impairment in the ability to check semantic readings derived lexicopragmatically and syntactically against the presented sentence or to operate on selected items (such as proper nouns and possibly nouns in the focus of discourse) in propositional and discourse representations. The STM system thus appears to be involved in a number of processes that involve checking or manipulation of representations derived by automatic firstpass processing. Similar conclusions have been drawn by several investigators (e.g., McCarthy & Warrington, 1987b, this volume, chapter 7) with respect to sentence and discourse-level processing. A broad generalization consistent with the entire set of neuropsychological studies dealing with word and sentence comprehension in STM patients is that STM is not utilized in automatic first-pass psycholinguistic processes — lexical access, accessing word meaning, parsing and sentence comprehension, and structuring discourse - but is invoked by second-pass processes (e.g., judgments based on phonological representations derived from written stimuli and checking sentence meaning against sentences form). If this view of the role of STM in language processing is correct, it suggests that language processing has fundamentally different operational characteristics with respect to the representations used in automatic first-pass and controlled second- and

STM and comprehension: neuropsychological studies

383

subsequent-pass language processing. Although there are some differences between our views, the division we are about to suggest overlaps to some degree with the division of mental processes into "domain-specific/encapsulated" and "isotropic/ Quinean" sets suggested by Fodor (1983). The data we have reviewed are consistent with the proposition that each stage of automatic first-pass language processing makes use of the fewest representational types possible. For instance, auditory lexical access makes use of phonological representations only in its first (contact) stage (Frauenfelder & Tyler, 1987). In Marslen-Wilson's (1987) view, context only serves to reduce the level of activation of the lexical items activated bottom-up. Written lexical access also only involves aspects of word form, although these seem to include phonological as well as orthographic representations (perhaps a reflection of lexical representations of form being a more abstract, modality-neutral representation of word form). Parsing and sentence interpretation, which cannot operate on phonological forms, do not appear to invoke mechanisms that involve phonological representations. This is true not only with regard to their nonreliance on mechanisms that involve controlled use of phonological representations, such as the AL, but also with regard to their nonuse of automatic mechanisms involving phonological representations, such as the PS. The structuring and interpretation of discourse may well refer not to syntactic structure per se but only to propositional content and semantic features such as focus; it is well established that syntactic information is not available in recall tasks even short periods after sentences have been presented. The subcomponents of the first-pass language comprehension system appear to use the minimal number of representational types needed to accomplish their individual objectives. That is, they are not only domain specific and primarily driven by bottom-up information, but they also are "Markovian," in the sense that representations computed early in the comprehension process and no longer relevant to later processes are not used by these later processes. Our discussion of case GI raises the possibility that this feature holds only of the adult and not the developing system, which may make more use of phonological representations. In noted contrast, those processes that review and operate on the results of automatic first-pass processing — all of which involve controlled operations — have access to whatever representations are available at the point such a process is engaged. Some such tasks (e.g., rhyme judgments) require phonological representations. Others do not but nonetheless can make use of them. For instance, phonological representations maintained in STM (presumably through the operation of AL or the PS) can be referred to by postinterpretive processes that can only make use of these representations indirectly (such as those that reinitiate parsing to check the thematic roles actually assigned by the parser against sets of thematic roles held in a representation of the propositional content of a sentence). On this view, it is to be expected that there will be very large number of task- and individual-related parameters that would determine

384

Caplan and Waters

which representations are referred to in second-pass tasks, as seems to be the case. There is some neuropsychological evidence that this is the case: EA, for instance, is reported to be able to recover the unpreferred sense of a lexical semantic anomaly (Martin, this volume, chapter 15) - a capacity that requires review of the presented sentence but that appears to be accomplishable without normal PS functions. If this is the case, it is reasonable to conclude that phonological representations play no necessary role in second-pass language processing, only an optional role that varies with their availability, which is subject to task and subject variables. The strongest conclusion one can reach - and one we feel is strongly supported by available evidence - is that phonological representations play obligatory roles only in tasks that absolutely require them. In first-pass automatic processing of language, this appears to be lexical access. In metalinguistic tasks, those referring to phonological form require such representations. Phonological representations maintained in STM itself play no role in first-pass on-line language processes beyond lexical access, and no obligatory role in second-pass language processing.

Notes 1. The tables include the information we have been able to abstract from the published reports as well as additional information about the patients that was not a part of the original published reports but that in some cases has been provided by the authors of the case reports. 2. Baddeley et al. (1987) review this case, and indicate that PV's comprehension is "relatively good." However, despite the fact that Vallar and Baddeley (1984 a, b) consider PV to be a prototypical example of a PS case, Baddeley et al. (1987) hedge regarding the implications of this case for the relationship between STM deficits and comprehension impairments. They do so because another measure of STM function — "sentence span" — is six words in PV. They suggest that "sentence span" rather than the tests assessing the subcomponents of STM is the relevant measure of short-term memory function. There are many problems with this analysis and these suggestions. First, "sentence span" is undefined, so it cannot be used as a measure of STM ability. The ability to repeat sentences verbatim probably varies as function of lexical content, syntactic structure, and other features of sentences. In general, subjects can repeat sentences verbatim better than lists. This fact, however, does not license the use of "sentence span" as a measure of STM function. On the contrary, this fact itself requires an explanation, and the most obvious explanation is that subjects structure sentences prosodically, syntactically, and semantically and use these structures to improve memory. If this is the case, one cannot use sentence span as a measure of STM function, because sentence span reflects sentence comprehension function(s) as well as STM function(s). The use of a "sentence span" deficit as a measure of STM deficit thus avoids the issue of whether the memory functions involved in STM tasks related to maintaining and rehearsing the phonological form of lexical items are used in sentence comprehension. Its use for this purpose is circular. 3. We note in passing that the considerable discussion that has sprung up in connection with AL disturbances and articulatory suppression effects (Howard & Franklin, chapter 12; Monsell, 1984; Besner, 1987) regarding the role of input and output phonological representations in tasks involving homophones, rhymes, and pseudohomophones, as well as the discussion of whether different controlled tasks (span, pseudohomophone judgment, homophone judg-

STM and comprehension: neuropsychological studies

385

ment, etc.) use the same or different phonological representations, seem to us to be largely irrelevant to the most important questions regarding language processing, which deal with automatic first-pass processing. All these tasks are controlled ones that involve different operations on phonological representations (storage, segmentation, comparison, etc.). It is to be expected that they would show w-way dissociations in patients due to disturbances in different components of these tasks in different patients. Specific roles for the AL and PS in these tasks are suggested by experiments on normals, but not proved, as controls for the general demands of dual-task situations and other factors have not been provided (Waters, Komoda, & Arbuckle, 1985).

References Allport, D. A. (1984a). Auditory—verbal short-term memory and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 313-325). London: Erlbaum. Allport, D. A. (1984b). Speech production and comprehension: One lexicon or two? In W. Prinz & A. F. Sanders (Eds.), Cognition and motor process (pp. 209-228). Heidelberg: SpringerVerlag. Baddeley, A. D. (1979). Working memory and reading. In P. A. Kolers, M E. Wrolstad, & H. Bouma (Eds.), Processing of visible language (pp. 327-350). New York: Plenum. 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. 108-129). Hillsdale, NJ: Erlbaum. Baddeley, A. D., & Lewis, V. J. (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, 36, 233-252. Baddeley, A. D., Thompson, 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). Comprehension and the articulatory loop: Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and performance XII (pp. 509-529). London: Erlbaum. Barnard, T. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). London: Erlbaum. Basso, A., Spinnler, H., Vallar, G., & Zanobio, E. (1982). Left hemisphere damage and selective impairment of auditory verbal short-term memory: A case study. Neuropsychologia, 20, 263-274. Berndt, R. (1985). Working memory and auditory sentence comprehension. Paper presented at the Venice meeting on cognitive neuropsychology. Berwick, R. C., & Weinberg, A. (1984). The grammatical basis of linguistic performance: Language use and acquistion. Cambridge, MA: MIT Press. Besner, D. (1987). Phonology, lexical access in reading, and articulatory suppression: A critical review. Quarterly Journal of Experimental Psychology, 39, 467-478. 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. Bishop, D. V. M. (1982). TROG Test for Reception of Grammar. Available from the author at the Department of Psychology, University of Manchester, England. Blumstein, S., Baker, E., & Goodglass, H. (1977). Phonological factors and auditory comprehension in aphasia. Neuropsychologia, 15, 19-30. Bub, D., Black, S., Howell, J., & Kertesz, A. (1987). Damage to input and output buffers: What's a

386

Caplan and Waters

lexicality effect doing in a place like that? In E. Keller & M. Gopnick (Eds.), Motor and sensory processes of language. Hillsdale, NJ: Erlbaum. Bub, J., & Bub, D. (1988). On methodology of single case studies in cognitive neuropsychology. Cognitive Neuropsychology, 5, 565—582. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38, 705-737. Campbell, R., & Butterworth, B. (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, 37, 435—475. Caplan, D. (1972). Closed boundaries and recognition latencies for words and sentences. Perception and Psychophysics, 12, 73-76. Caplan, D. (1987). Neurolinguistics and linguistic aphasiology. An introduction. Cambridge: Cambridge University Press. Caplan, D., Baker, C, & Dehaut, F. (1985). Syntactic determinants of sentence comprehension in aphasia. Cognition, 21, 117-175. Caplan, D., & Futter, C. (1986). Assisgnment of thematic roles to nouns in sentence comprehension by an agrammatic patient. Brain and Language, 27, 117—13A. Caplan, D., & Hildebrandt, N. (1986). Language deficits and the theory of syntax. A reply to Grodzinsky. Brain and Language, 27, 168-177. Caplan, D., & Hildebrandt, N. (1988). Disorders of syntactic comprehension. Cambridge, MA: Bradford Books. Caplan, D., Vanier, M , & Baker, C. (1986a). A case study of reproduction conduction aphasia: I. Word production. Cognitive Neuropsychology, 3, 99—128. Caplan, D., Vanier, M., & Baker, C. (1986b). A case study of reproduction conduction aphasia: II. Sentence comprehension. Cognitive Neuropsychology, 3, 129-146. Caramazza, A., Basili, A., Koller, J. J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 234-271. Caramazza, A., Berndt, R., & Basili, A. G. (1983). The selective impairment of phonological processing: A case study. Brain and Language, 18, 128—17'4. Caramazza, A., & Berndt, R. S. (1985). A multicomponent view of agrammatic Broca's aphasia. In M. L. Kean (Ed.), Agrammatism (pp. 27-63). New York: Academic Press. Caramazza, A., & Zurif, E. (1976). Dissociation of algorithmic and heuristic processes in language comprehension: Evidence from aphasia. Brain and Language, 3, 572—582. Carlson, G. N., & Tanenhaus, M. K. (1987). Thematic roles of language comprehension. Cognitive Science Technical Report URCS 42. Rochester, NY: University of Rochester. Clark, H., & Clark, E. (1977), Psychology and language. New York: Harcourt Brace Jovanovich. Conrad, R. (1964). Acoustic confusions in immediate memory. British Journal of Psychology, 55, 75-84. Cook, V. J. (1975). Strategies in the comprehension of relative clauses. Language and Speech, 18, 204-218. Crowder, R. G., & Morton, J. (1969) Precategorical acoustic storage (PAS). Perception and Psychophysics, 5, 365-373. Darwin, C. J., Turvey, M. T., & Crowder, R. G. (1972). An auditory analogue of the Sperling partial report procedure: Evidence for brief auditory storage. Cognitive Psychology, 3,

255-267. Dowty, D. R., Karttunen, L, & Zwicky, A. M. (Eds.) (1985). Natural language parsing. Cambridge: Cambridge University Press. Dunn, L. M., & Dunn, L. M. (1981). Manual for forms L and M of the Peabody Picture Vocabulary Test - Revised. Circle Pines, MN: American Guidance Service. Estes, W. K. (1973). Phonemic coding and rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior, 12, 360-372.

STM and comprehension: neuropsychological studies

387

Ferreira, F., & Clifton, C. (1986). The independence of syntactic processing. Journal of Memory and Language, 25, 348-368. Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press. Frauenfelder, A. H., & Tyler, L. (1987). The process of spoken word recognition: An introduction. Cognition, 25, 1-20. Frazier, L, Clifton, C, & Randall, J. (1983). Filling gaps: Decision principles and structure in sentence comprehension. Cognition, 13, 187-222. Frazier, L, & Fodor, J. D. (1978). The sausage machine: A new two-stage model of parsing. Cognition, 6, 291-325. Friedrich, F. J., Glenn, G, & Marin, O. S. M. (1984) Interruption of phonological coding in conduction aphasia. Brain and Language, 22, 266—291. Friedrich, F. J., Martin, R. G, & Kemper, S. J. (1985). Consequences of a phonological coding deficit on sentence processing. Cognitive Neuropsychology, 2, 385-412. Garnsey, S., Tanenhaus, M. K., & Chapman, R. (1987). Evoked potential measures of sentence comprehension. Paper presented at the Cognitive Science Society, Portland. Greene, R. L, & Crowder, R. G. (1984).'Modality and suffix effects in the absence of auditory stimulation. Journal of Verbal Learning and Verbal Behavior, 23, 371—382. Grier, J. B. (1971). Non-parametric indexes for sensitivity and bias: Computing formulas. Psychological Bulletin, 75, 424-429. Grodzinsky, Y. (1986). Language deficits and the theory of syntax. Brain and Language, 27, 135-159. Hildebrandt, N., Waters, G. S., & Caplan, D. (submitted). On the nature and functional role of verbal short-term memory in sentence comprehension: Evidence from neuropsychology. Howard, D., & Franklin, S. (1987). Three ways for understanding written words and their use in two contrasting cases of surface dyslexia. In D. A. Allport, D. Mackay, W. Prinz, & E. Scheerer (Eds.), Language perception and production: Relationships between listening, speaking, reading, and writing (pp. 340—366). London: Academic Press. Huttenlocher, J., & Strauss, S. (1968). Comprehension and statement's relations to the situation it describes. Journal of Verbal Learning and Verbal Behavior, 7, 527-530. Jarvella, R. V. (1970). Effects of syntax on running memory span for connected discourse. Psychonomic Science, 19, 235-236. Jarvella, R. V. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Behavior, 10, 409-416. Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language inference and consciousness. Cambridge: Cambridge University Press. Kinsbourne, M. (1972). Behavioral analysis of the repetition deficit in conduction aphasia. Neurology, 22, 1126-1132. Kleiman, G. (1975). Speech recoding in reading. Journal of Verbal Learning and Verbal Behavior, 14, 323-329. Levy, B. A. (1971). The role of articulation in auditory and visual short-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 123-132. Linebarger, M. G, Schwartz, M. F., & Saffran, E. M. (1983). Syntactic processing in agrammatism: A reply to Zurif and Grodzinsky. Cognition, 15, 215-226. McCarthy & Warrington (1987a). Understanding: A function of short-term memory? Brain, 110,

1565-1578. McCarthy, R, & Warrington, E. K. (1987b). The double-dissociation of short-term memory for lists and sentences: Evidence from aphasia. Brain, 110, 1545-1563. Marcus, M. P. (1980). A theory of syntactic recognition for natural language. Cambridge, MA: MIT Press. Marslen-Wilson, W. D. (1987). Functional parallelism in spoken word recognition. Cognition, 25, 71-102.

388

Caplan and Waters

Martin, R. C. (1987). Articulatory and phonological deficits in short-term memory and their relation to syntactic processing. Brain and Language, 32, 159-192. Martin, R. C, & Caramazza, A. (1982). Short-term memory performance in the absence of phonological coding. Brain and Cognition, 1, 50-70. Martin, R. C, Jerger, S., & Breedin, S. (1987). Syntactic processing for auditory and visual sentences in a learning disabled child: Relation to short-term memory. Developmental Neuropsychology, 3, 129-152. Monsell, S. (1984). Components of working memory verbal skills: A "distributed capacities" view. In. H. Bouma & B. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 327-350). Hillsdale, NJ: Erlbaum. Murray, D. J. (1968). Articulation and acoustic confusability in short-term memory. Journal of Experimental Psychology, 78, 67'9-684. Parisi, D., & Pizzamiglio, L. (1970). Syntactic comprehension in aphasia. Cortex, 6, 204-15. Perfetti, C, & McCutchen, D. (1982). Speech processes in reading. In N. Lass (Ed.), Advances in speech and language (Vol. 7). New York. Academic Press. 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. Posner, M. I., & Keele, S. (1967). Decay of visual information from a single letter. Science, 158, 137-139. Saffran, E. M. (1985). Short-term memory impairment and language comprehension: Specifying the nature of the interaction, Paper presented at the Venice meeting on cognitive neuropsychology. 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—33. Saffran, E. M., Marin, O., & Yeni-Komshian, G. H. (1976). An analysis of speech perception in word deafness. Brain and Language, 3, 209—229. Salame, P., & Baddeley, A. (1982). Disruption of short-term memory by unattended speech: Implications for the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 21, 150-164. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1-66. Seidenberg, M. S., Waters, G. S., Barnes, M., & Tanenhaus, M. (1984). When does irregular spelling or pronunciation influence word recognition? Journal of Verbal Learning and Verbal Behavior, 23, 383-404. 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. Shallice, T., & Warrington, E. K. (1975). Word recognition in a phonemic dyslexic patient. Quarterly Journal of Experimental Psychology, 27, 187-199. 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. (1980). Single and multiple component central dyslexic syndromes. In M. Coltheart & J. Marshall (Eds.), Deep dyslexia (pp. 119-145). London: Routledge & Kegan Paul. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 82, 127-190. Slobin, D. (1966). Grammatical transformations and sentence comprehension in childhood and adulthood. Journal of Verbal Learning and Verbal Behavior, 2, 219-227.

STM and comprehension: neuropsychological studies

389

Starr, A., & Barrett, G. (1987). Disordered auditory short-term memory in man and event-related potentials. Brain, 110, 935-959. Stowe, L. A. (1989). Thematic structures and sentence comprehension. In G. Carlson and M. Tanenhaus (Eds.), Linguistic Structure in Language Processing (pp. 319-358). Drodrecht: Kluwer. Strub, R. L, & Gardner, H. (1974). The repetition defect in conduction aphasia: Mnestic or linguistic? Brain and Language, 1, 241-255. Swinney, D. A. (1979). Lexical access during sentence comprehension: (Re: consideration of context effects). Journal of Verbal Learning and Verbal Behavior, 18, 645-660. Tanenhaus, M. K., Leiman, J. L, & Seidenberg, M. S. (1979). Evidence for multiple stages in the processing of ambiguous words in syntactic context. Journal of Verbal Learning and Verbal Behavior, 18, 427-441. Tzortis, C, & Albert, M. (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, 151-61. Vallar, G., & Baddeley, A. D. (1984b). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-41. Vallar, G., & Baddeley, A. D. (1987). Phonological short-term store and sentence processing. Cognitive Neuropsychology, 4, 417-438. 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. Warrington, E. K., & Shallice, T. (1972). Neuropsychological evidence of visual storage in shortterm memory tasks. Quarterly Journal of Experimental Psychology, 24, 30—40. Waters, G., Komoda, M., & Arbuckle, T. (1985). The effects of concurrent tasks on reading: Implications for phonological recoding: Journal of Memory and Language, 24, 27-45. Waters, G. S., & Seidenberg, M. S. (1985). Spelling-sound effects in reading: Time course and decision criteria. Memory and Cognition, 13, 557-572. Wickelgren, W. A. (1969). Auditory or articulatory coding in verbal short-term memory. Psychological Review, 76, 232-235. Zurif, E., & Grodzinsky, Y. (1983). Sensitivity to grammatical structure in agrammatic aphasics: A reply to Linebarger, Schwartz, and Saffran. Cognition, 15, 207-214.

15. Neuropsychological evidence on the role of short-term memory in sentence processing RANDI C. MARTIN

Many claims have been made attributing comprehension deficits in brain-damaged subjects to their short-term memory deficits (e.g., Saffran & Marin, 1975; Caramazza, Basili, Koller, & Berndt, 1981; Vallar & Baddeley, 1984a; Friedrich, Martin, & Kemper, 1985; Martin, Jerger, & Breedin, 1987). However, among the patients with similar restrictions in memory span, different levels of comprehension have been found. In fact, some recent studies have demonstrated impressive sentence-processing abilities in individuals with very restricted memory spans (Vallar & Baddeley, 1984a; Butterworth, Campbell, & Howard, 1986; Martin, 1987; McCarthy & Warrington, 1987a, b; Caplan & Hildebrandt, 1988). Before dealing in detail with the empirical evidence on patients' short-term memory and comprehension abilities, this chapter addresses current theories of the comprehension process with an emphasis on the possible points at which various types of memory storage might be involved. This discussion is followed by a consideration of evidence regarding the types of memory storage that appear to be involved in typical short-term memory tasks and how these might overlap with those involved with comprehension. With this background in mind, the patterns of associations and dissociations in brain-damaged individuals will be brought to bear in determining what connections between memory and comprehension appear consistent with current evidence.

15.1. What has to be remembered during sentence comprehension? The goal of sentence processing is to arrive at a representation of the meaning of the sentence. In this discussion it will be assumed that meaning can be represented as a set of propositions specifying semantic relationships among constituents, such as role relationships about the verb, the existence or properties of objects, and temporal relations among objects or events (Clark & Clark, 1977; Kintsch & van Dijk, 1978; Just & Carpenter, 1987). During auditory sentence comprehension, in order to arrive at the The preparation of this manuscript was supported by NIH Grant 19652 to Rice University.

390

The role of STM in sentence processing

391

propositional representation, acoustic information arriving at the ear of the listener must be transformed into a phonological form in order to access the lexicon. Once lexical entries have been located, the syntactic and semantic features of the words become available to feed into parsing and semantic interpretation mechanisms. The output of these syntactic and semantic processes is the propositional representation. Although the involvement of phonological, lexical, semantic, and syntactic processes and representations in comprehension is not controversial, the timing of the application of these processes is of direct relevance to issues concerning the role of memory in comprehension. To the extent that sentence processing must depend on phonological memory, processing would be limited to the few items that can be maintained simultaneously in such a format. However, there is evidence that memory capacity increases if syntactic structure can be assigned to a group of words (Tejirian, 1968). Further, there is evidence that representations of meaning (i.e., propositional representations) are less subject to decay or interference than either phonological or syntactic representations (Sachs, 1967). Thus, the more quickly that propositional representations can be developed from the input, the less the dependence on rapidly fading surface representations. The question, then, is the point at which syntactic structure is assigned to the input and the point at which propositions are derived from syntactic structure and lexical identities. Early psycholinguistic theories taking a transformational approach assumed that a surface syntactic representation was developed first that was used to derive the deep structure representation. Semantic interpretation could only follow derivation of the deep structure. Since the deep structure had to be derived from the surface representation of an entire sentence (or at least a clause), it would be necessary to hold onto the entire syntactic representation of the sentence before propositions could be developed. In addition to the syntactic representation, it would be necessary to preserve the lexical identity of the words in the sentence, and thus some type of lexical memory system (perhaps phonological in nature) would also be taxed prior to the development of propositions. Most recent theories of comprehension do not assume that an entire sentence or clause must be held in a surface form prior to the development of propositions, but rather assume that some smaller word group serves as the unit on which semantic interpretation is attempted (Berwick & Weinberg, 1984; Crain & Steedman, 1985; Frazier, 1985; Just & Carpenter, 1987). The READER model of Just and Carpenter (Just & Carpenter, 1987; Thibadeau, Just, and Carpenter, 1982) represents one example of the extreme case in which as much semantic and syntactic processing as is possible is carried out on each word as it is identified. In their model, syntactic categories are identified one word at a time, and hypotheses generated about successive categories. For example, after encountering the determiner the, a subsequent noun is expected. Larger constituents such as noun phrase and verb phrase are identified and completed as soon as possible. Semantic processing is assumed

392

R. Martin

to occur in parallel with syntactic processing and to make use of syntactic information as it becomes available. Also, syntactic parsing decisions can make use of the semantic information available up to that point in the sentence. Another example of a model that uses a somewhat larger unit for semantic analysis is the parsing model of Berwick and Weinberg (1984). Like READER, the Berwick-Weinberg parser develops syntactic structure on a word-by-word basis. However, the Berwick-Weinberg parser uses a lookahead buffer of about two or three words so that if the structure to be developed for a certain word is ambiguous, the next few words can be analyzed to determine the correct structural choice. This model assumes that semantic interpretation is carried out on phrasal units of different sizes (e.g., noun phrases, verb phrases, sentences) after these units are completed in the syntactic analysis. Once the unit is completed, the propositions derived from the unit are constructed and the detailed syntactic information below the highest node for that unit is lost. Thus, unlike the READER model, semantic interpretation lags behind syntactic analysis to some degree because of the requirement that phrasal units be completed before propositions are derived. Also, initial decisions about syntactic structure are not influenced by semantic information. The assumption of immediate syntactic processing that is common to both models appears justified by several lines of research (Tyler & Marslen-Wilson, 1977; Just & Carpenter, 1980; Frazier & Rayner, 1982; Rayner, Carlson, & Frazier, 1983). For example, Frazier and Rayner (1982) and Rayner et al. (1983) have shown that readers spend a longer time fixating a word if it is inconsistent with the syntactic structure that has been developed up to that point than if it is consistent. For example, in a sentence such as The performer sent the flowers was very pleased, these studies indicate that readers

are likely to interpret sent as main clause active verb and are therefore surprised when encountering the was. Rayner et al. (1983) found that the time spent fixating the was was longer in such a reduced relative clause sentence than in a corresponding sentence with a nonreduced relative (i.e., The performer that was sent the flowers was very pleased).

The longer time indicates that readers were constructing the syntactic structure on a word-by-word basis because they detected the syntactic incongruity on the first word that did not fit with the previously developed structure. The data also indicate that although the structure to be developed on encountering sent was ambiguous, a choice had been made, at least by the point that the main clause verb was read. A series of studies by Just and Carpenter and colleagues (Just & Carpenter, 1980; Carpenter & Just, 1981; Carpenter & Daneman, 1981) indicates that semantic processing also appears to be carried out on a word-by-word basis. These studies have shown that readers spend longer times fixating a disambiguating word if they had initially misinterpreted the meaning of an earlier ambiguous word (e.g., a homograph such as tears that has two distinct meanings) than if they correctly interpreted its meaning. As in the studies of the detection of syntactic incongruity, these longer times occurred on the first word indicating a semantic mismatch. These studies were not

The role of STM in sentence processing

393

designed to address Berwick and Weinberg's proposal that semantic interpretation occurs only at the end of a phrase. In most of their examples, the disambiguating word was the last word of a noun phrase or prepositional phrase, and thus one could not determine whether longer fixation times would also be observed for disambiguating words that were internal to a phrase. Because of the evidence favoring immediacy of processing, it might be thought that there would be no need to retain a verbatim representation of a sentence during comprehension. However, as will be discussed, there are some situations in which retention of at least a portion of a sentence in a verbatim form is useful. Also, although a verbatim form may be of only limited usefulness, there is a requirement in such a system for the retention of the developing semantic and syntactic structure and hypotheses about upcoming material Thus, there are memory demands for more abstract representations that have been shown to tax an immediate memory system. The following sections on auditory comprehension, reading comprehension, and repetition detail the possible situations in which different types of memory codes may be needed at the various steps from segmental speech perception to the construction of propositions to the selection of a response.

15.1.1. Auditory comprehension Speech perception and lexical access

As phonetic information is derived from the incoming auditory signal, it is necessary that the sequence of phonemes be maintained until lexical access is achieved. However, lexical access occurs very quickly (on the order of 200 msec after the onset of a word, according to Marslen-Wilson & Welsh, 1978), and thus prelexical phonological information would not have to be maintained long before it could be replaced with a lexical code. There are some situations in which a phonological string is ambiguous with regard to which word or words have been presented. For example, in the two sentences / better do my laundry and / bet her five dollars, the underlined segments might be pronounced identically and thus the segmental phonological information might have to be maintained until the disambiguating information was reached. Even in such cases, it seems evident that disambiguation will occur within a word or so beyond the ambiguous section. Thus, the capacity needed to maintain segmental phonological information even in such ambiguous cases would be limited to that contained in a few words.

Syntactic parsing

Once lexical entries have been identified, word class and selectional restrictions may be obtained to guide the syntactic parsing process. As discussed earlier with regard to the

394

R. Martin

Berwick-Weinberg model, parsing models sometimes incorporate a preparsing buffer containing a limited number of lexical items. The buffer is usually employed as a lookahead buffer so that if the parser reached a point of ambiguity based on the current word, the parser can look ahead to the next few words in the sentence to aid in the decision. The exact coding format in which these words are maintained has not been specified in these models. It is possible that a phonological code that is tied to lexical units is used (Barnard, 1985), although this is not necessarily the case. Some more abstract lexical code might also be employed. It should be noted that the results for some sentence types using the eye-fixation methodology are not supportive of the three-word lookahead buffer proposed by Berwick and Weinberg (1984). For example, for the sentence Since Jay always jogs a mile seems like a short distance to him, an incorrect structure should not be assigned to the ambiguous region (i.e., a mile) if the parser was also considering seems when making the assignment. However, according to parsing principles discussed by Frazier and Rayner (1982), if the parser makes decisions without looking ahead, the parser should prefer to structure a mile as the direct object oijogs rather than as the subject of the main clause. Frazier and Rayner compared reading times for a sentence such as this to control sentences in which the preferred structure would be correct (e.g., Since Jay always jogs a mile this seems like a short distance to him). Reading times for the words immediately following the ambiguous region were shorter for sentences that matched the parser's preferences than for those that did not. Frazier and Rayner interpreted these results as indicating that the parser did not look ahead even a short distance. A plausible argument could be made based on the eye-movement data that the parser makes a decision immediately on each word without looking ahead. Whether or not a lookahead buffer is used, parsing decisions made at one point in a sentence may later prove to be incorrect, and then the parser must have some means of reanalyzing the sentence, or at least the incorrect portion. Again there is little evidence on the nature of the memory code that might be used for the reanalysis. A persisting phonological representation of the sentence could serve as the input to the reanalysis. Another type of memory representation needed during the syntactic analysis of the sentence would be some means for retaining the results of the parsing process itself, both for the syntactic units that have been identified and for expectations for upcoming material. In several parsing models, a pushdown stack is used to keep track of phrases that have been begun or completed (e.g., Marcus, 1980; Berwick & Weinberg, 1984; Pereira, 1985). In the Berwick-Weinberg model, as each, new phrase is begun, the phrase that the parser has been working on is pushed down one level. For example, in a sentence beginning The girl who was walking the dog, a noun phrase would be begun upon processing the first word. When who was reached, the noun phrase would be pushed down one level, and a new phrase would be started for the embedded clause. Once a phrase is completed, the propositions derived from that phrase are placed on the

The role of STM in sentence processing

395

propositional list and the phrase is removed from the stack. All the remaining phrases then move up one level, and processing continues on the phrase now at the top of the stack. One might expect that the more incomplete phrases on the stack, the greater the demands on a memory for syntactic structure. Sentences that stretched the limits of this capacity should be difficult to understand. Frazier (1985) presents some evidence along these lines, although her explanation for the source of syntactic difficulty is somewhat different. She suggests that the number of nonterminal structural nodes that have to be introduced within a local region of a sentence (i.e., across three-word spans) determines the difficulty of syntactic processing. Since the number of nonterminal nodes would be closely related to the number of new phrases that had to be introduced, Frazier's measure of difficulty would be closely related to the number of phrases on the pushdown stack. The local nature of the measure could reflect the fact that for any given point in a sentence, phrases begun at a larger distance from the current word are likely to be parts of phrases that have already been completed and thus removed from the stack. The type of memory representation needed for retaining the results of the parsing processes would obviously have to be one that represented syntactic structures. Thus, the information would be in a highly abstract format rather than in some memory code closely tied to perceptual analysis. It is possible, however, that lexical items are represented by means of a phonological code in the developing syntactic representation (Saffran & Martin, this volume, chapter 16). A phonological code might be useful because it would be one means of maintaining exact lexical identities until propositions were derived. Although the number of incomplete phrases at any point relates to syntactic processing difficulty, other factors that may have memory implications also play a role. For example, the center-embedded subject relative structure The boy who pushed the girl was tired and the center-embedded object relative structure The girl who the boy pushed was tired would contain the same number of incomplete phrases until processing was, yet the object relative structure is more difficult to understand. Another important factor affecting syntactic processing complexity appears to be the need to determine long-distance dependencies (e.g., to determine the referent of a reflexive pronoun or to determine the mapping between a missing element in a relative clause and its antecedent). Caplan and Hildebrandt (1988) present an extensive discussion of different types of referential dependencies and how they are handled during parsing. Here we will only consider the case of determining the role of the head noun phrase in subject versus object relative sentences, since this will be of relevance to the later discussion of patient data. Several explanations have been offered for the greater difficulty of center-embedded object than subject relative constructions. Wanner and Maratsos (1978) have offered a

396

R. Martin

memory-related explanation. In their parsing model, upon encountering the relative pronoun, the head noun phrase is placed in a memory buffer until it can be assigned a grammatical role with respect to the embedded verb. Since the head noun phrase must be maintained in this buffer for a longer time for object than subject relative constructions, the memory demands of the object relative form are greater and cause greater comprehension difficulty. Ford (1983) has offered a processing rather than a memory explanation for the comparative difficulty of object relatives. She argues that the difficulty with the object relative form does not arise until encountering the "gap" following the embedded verb (i.e., the slot where the direct object of the verb would be expected to appear but does not). At the gap, the parser must search back over the structure that has been developed to find the appropriate noun phrase to fill the gap. In the subject relative form, if there is a gap at all (cf. Ford, p. 212), it is much closer to its antecedent. Thus, the need to search back over a greater number of words to find the filler for the gap causes a greater difficulty of the object than the subject relative sentences. A similar explanation could be derived from the assumptions of the Berwick-Weinberg parser. This model also incorporates a procedure for searching back over the syntactic structure to find the noun phrase to fill a gap. The need for potentially time-consuming searches over the syntactic structure suggests another possible use for a pre-parsing buffer. Specifically, such a buffer might be used as a means for maintaining subsequent words in a sentence while the analysis of an earlier portion is completed. In auditory comprehension, the remaining words of a sentence continue to arrive at the ear of the listener even though the analysis of the earlier portion may still be in progress. Thus, such a buffer would be useful whenever sentence processing lagged behind the input. The number of words that would have to be maintained in such a buffer would depend on how long various processes took. There appear to be little data on auditory sentence comprehension that would directly bear on this question. The gaze-duration data from reading tasks are difficult to translate into auditory comprehension times for several reasons. First, lexical access time is probably different for the two modalities. Also, in gaze-duration studies, readers sometimes initiate regressions to earlier portions of the sentence when trying to interpret a later part, whereas for auditory comprehension any "regression" would have to be carried out in the mind of the listener. The time parameters for visual and mental regressions are most likely quite different. Finally, readers often misinterpret the meaning of syntactically difficult sentences (Rayner et al., 1983), and thus gaze-duration times may not reflect the time needed to correctly analyze a sentence. The situation of completing the processing of one section while another is arriving could be considered analogous to the selective attention situation in which processing is directed to one message while another is being presented. Such studies have demonstrated that the phonological form of the unattended message remains available

The role of STM in sentence processing

397

for approximately 2 sec (Treisman, 1964). Also, there is evidence that the semantic representations of the words in the unattended message are activated (Eich, 1984). However, there is no evidence that syntactic or deeper semantic processes are carried out on the unattended message. Thus, a phonologically based short-term memory would be a likely storage system for holding the remainder of the sentence past the point of difficulty (Shallice, 1979).

Semantic interpretation The semantic propositions that are constructed as a sentence is parsed must be retained and integrated with each other. As discussed earlier, evidence indicates that propositions are retained quite well and much longer than information about the exact phonological and syntactic structure of a sentence (Sachs, 1967). Thus, it might be thought that all of the propositions derived from a sentence become part of an unlimited capacity long-term memory system and would not be forgotten. However, even in Sachs's study in which a recognition procedure was used, performance was not perfect on rejecting sentences that were semantically different from the original even at the shortest delay between presentation and test. Furthermore, performance declined over increasing delays up to the longest delay that was used (160 syllables past the end of the target sentence). The factors affecting the likelihood of retaining different propositions have been studied mainly at the text level rather than at the single sentence level (e.g., Kintsch & van Dijk, 1978). Studies of single sentence recall indicate that semantic coherence affects the probability of recall (Tejirian, 1968; Forster & Ryder, 1971). In the Forster and Ryder study, better recall was found for semantically coherent sentences (e.g., Several children raced to the burning building) than for other sentences that were meaningful but bizarre (e.g., The noise slowly deafened the pretty minister). Two memory-related issues that were raised with regard to syntactic processing would also be relevant to semantic processing. For one, there may be a need to retain a verbatim memory representation of a sentence in case a sentence needs to be reanalyzed. Because propositions appear to be derived as quickly as possible, lexically ambiguous words (e.g., case as in suitcase vs. legal case) may initially be misinterpreted. Consequently, the propositions derived from the misinterpreted word would have to be revised after the error was discovered. As was true for the recovery from the misinterpretation of syntactic structure, it is not clear what type of sentence representation is used for the recovery from a semantic misinterpretation. Gazeduration studies indicate that readers' eyes often regress to the ambiguous word upon encountering a later disambiguating word (Carpenter & Daneman, 1981). In auditory comprehension, it is possible that the listener has to refer back to a surviving

398

R. Martin

phonological representation of the ambiguous word in order to access another meaning. A second issue was the possible need for a pre-parsing buffer to maintain downstream lexical information when syntactic processing could not keep pace with the input. It is also possible that lexical and semantic processes could cause processing to lag behind the input. For example, low-frequency words might slow both syntactic processing and semantic interpretation because of the greater time needed to access word class and semantic feature information. Also, the introduction of a new concept appears to slow text processing considerably, independently of the frequency of the word or words used to express the concept (Just & Carpenter, 1980). As discussed earlier, it is possible, though not necessarily the case, that this preparsing buffer could store information in the form of lexically coded phonological representations.

Response selection

In tasks used to assess comprehension, it may be that a subject could understand a sentence as it was being spoken, but forget the derived propositions while trying to decide on a response (e.g., while viewing a set of pictures in order to choose the one matching the sentence). More forgetting might be expected the less semantically coherent the sentence. For sentences whose propositions are difficult to remember, it may be that retaining the surface form through rehearsal would aid performance. The subject could keep the sentence activated in a verbatim form and reanalyze it until a response could be selected.

15.1.2. Reading comprehension In reading, some of the memory demands of auditory comprehension are avoided. For example, there would be no need to retain a surface form of a sentence downstream from the point at which a syntactic complexity occurred because the remainder of the sentence would remain available on the printed page. Also, perceptual ambiguity and ambiguity in lexical segmentation would not be involved in reading, assuming that one is reading a clearly printed page. Thus, any memory demands that might derive from these types of ambiguity would not occur in reading. On the other hand, syntactic and lexical—semantic ambiguity would be involved in reading as much as in auditory comprehension. In fact, structural ambiguities may be more difficult to deal with in reading than listening because there would be no prosodic cues to aid parsing. (On the other hand, there may be punctuation that provides more obvious cues to segmentation than prosody.) In recovering from an error in the interpretation of ambiguous material, one has the advantage of being able to reread a section of a sentence, and subjects apparently often do so (Carpenter & Daneman, 1981; Frazier & Rayner, 1982). Even so,

The role of STM in sentence processing

399

subjects do not reread the entire sentence, but often go back immediately to the ambiguous word. Thus, even in reading, the subjects must be retaining a representation of the point at which the ambiguity occurred. As in listening, the memory demands deriving from syntactic complexity would hold for reading. That is, difficult syntactic structures (e.g., those with high local nonterminal node counts) would pose just as great a memory load irrespective of the format of input. Also, the processes demanded in determining referential dependencies would be equivalent in reading and auditory comprehension. Finally, the need to maintain propositional representations would be the same for the two modalities. Task demands requiring the subject to remember the meaning of a sentence for a period of time before making a response would not be involved in reading if the sentence remained visible until a response was made. However, if the sentence were removed from sight prior to response or if a procedure such as rapid serial visual presentation were used, similar demands for remembering the meaning while selecting a response would be involved.

15.1.3. Repetition The preceding discussion has focused on comprehension and the memory demands involved in deriving a propositional representation. Unlike comprehension, verbatim repetition of sentences is hot an activity that one typically carries out in everyday life. Nonetheless, repetition is a task often used in psychological studies, and often for the purpose of assessing comprehension. The involvement of comprehension processes in sentence repetition is evident from the fact that subjects can repeat many more words if they form sentences than if they form random lists. Early research demonstrated that the closer a word list approximated normal English prose the greater the subject's recall (Miller & Selfridge, 1950). Both semantic and syntactic factors appear to contribute to the enhanced recall (Tejirian, 1968). Although both semantic and syntactic processing are involved in sentence repetition, one might expect that a phonological representation would be an important aid to repetition in order to preserve the exact lexical representations contained in the input. It might be argued that a phonological record would be of little help in sentence repetition, since phonological representations appear to persist only on the order of 2 sec or so if not rehearsed (Treisman, 1964). However, a fairly long sentence can be spoken in under 2 sec. Also, the estimates of phonological persistence come either from recall of random lists of words or from recall of words in unattended sentences that have not benefited from extensive semantic and syntactic processing. In these situations, a more complete phonological record might be needed for accurate recall than when the words are in a sentence that has been comprehended. For an attended sentence, even if only a few phonological features persist for a given word, the syntactic and semantic

400

R. Martin

information about that word can be used in addition to the phonological information to choose the correct lexical entry. 15.2. The relationship of short-term memory and memory involved in sentence processing As discussed earlier, sentence processing involves memory for phonological, lexical, semantic, syntactic, and propositional information. The question to be addressed here is the extent to which the memory systems involved in comprehension overlap those measured by traditional short-term memory tasks. Many claims have been made attributing sentence comprehension deficits to patients' short-term memory deficits. The evidence indicating a short-term memory deficit has typically come from ordered serial recall tasks such as digit span. In order for a deficit in serial recall to have implications for sentence processing, it would have to be the case that the memory components involved in span tasks overlap with those demanded by sentence processing. What aspects of memory are involved in a memory span tasks? It used to be thought that memory span measured some general information-processing capacity (see Klatzky, 1980, for discussion). If so, then memory span could reflect a composite capacity for all the types of memory codes involved in comprehension. However, more recent evidence indicates that span should not be equated with general capacity. For one, it appears possible to carry out certain information-processing tasks while retaining a full memory load, with little or no decrement in performance (Klapp, Marshburn, & Lester, 1983). On other tasks including prose comprehension, Baddeley and Hitch (1974) found that a six-item memory load did impair performance, but did not have the devastating effect that might have been expected if the span task had used up all or most of the information-processing capacity. Also, little relation has been found between individuals' memory span and their ability to carry out complex informationprocessing tasks such as reading comprehension (Hunt, Lunneborg, & Lewis, 1975; Perfetti & Goldman, 1976). A second notion of memory span that appears valid for the most part is that span reflects the capacity for retaining verbal information in a phonological form. A great deal of evidence implicates an articulatory component to memory span: Span is greater for short than for long words, an individual's speech rate correlates highly with memory span, and articulatory suppression reduces memory span and eliminates word length effects (see Baddeley, 1986, for a review). Although articulatory suppression reduces memory span, it does not reduce it to zero. Memory for auditorily presented items under articulatory suppression is still affected by phonological similarity though not by word length (Baddeley, Lewis, & Vallar, 1984). The phonological similarity effect implies a phonological character to the residual nonarticulatory component; however,

The role of STM in sentence processing

401

the absence of an effect for word length implies that the capacity of this system is not in terms of phonemes. Perhaps this system stores lexical units coded in a phonological form (Barnard, 1985). Although an articulatory and a nonarticulatory phonological component can explain most of what is known about performance on a serial recall task, there are some findings that are not easy to accommodate within a model consisting of only these two components. First, memory span is greater for high- than for low-frequency lists, and the effect of frequency interacts with serial position (Watkins, 1977). Wright (1979) argued that these frequency effects could derive from faster rehearsal rates for highthan for low-frequency words. However, Martin (1987) found a frequency effect on span even for patients who could not rehearse. Second, memory for visually presented phonologically identical words (e.g., write, right, rite) is only slightly reduced relative to phonologically distinct words if subjects are requested to pronounce the words silently to themselves (Crowder, 1978). (A larger discrepancy was found when subjects were asked to pronounce the words aloud). Other results from Crowder's study suggested that subjects were not relying on a visual code to remember the phonologically identical items. Crowder suggested that the subjects were relying on an abstract lexical code that was not phonologically based. Other researchers have made similar suggestions concerning a lexical contribution to span (see, for example, Monsell, 1984; Saffran & Martin, this volume, chapter 6). Based on the earlier discussion of sentence processing, it should be possible to make some predictions about the effects of restricted memory span on sentence processing. If reduced memory span is due to a disruption of articulatory processes, few detrimental consequences would be predicted. The only proposed role for a rehearsal process was in keeping the verbatim representation of a sentence activated prior to making a response. It was suggested that this might be useful in cases in which the propositions derived from the sentence were difficult to remember, for example, where there was little semantic coherence among the content words in the sentences. Greater consequences would be expected from a disruption of a phonological memory system. First, phonological memory could be used for holding phonologically coded lexical items in a preparsing buffer. As discussed earlier, this preparsing buffer could be used either for the resolution of local syntactic ambiguities or for maintaining downstream information while the analysis of an earlier portion of the sentence was completed. Second, phonological memory might be needed to maintain lexical items in the syntactic representation. Third, phonological memory could provide a backup record of a sentence in case a semantically or syntactically ambiguous sentence was initially misinterpreted and later had to be reanalyzed (see Shallice, 1979, for a similar suggestion). Fourth, for sentence repetition, it was suggested that phonological memory would serve as a useful adjunct to semantic and syntactic representations in ensuring verbatim repetition. Finally, if it is assumed that an articulatory memory

402

R. Martin

system must be fed by a phonological representation, then a phonological representation would be needed to ensure rehearsal of a sentence whose propositions were difficult to remember. A disruption of memory span due to an articulatory or phonological source would be expected to have adverse consequences for the processes presumed to be dependent on that memory code. It should be noted that memory span is measured for random word lists, and thus does not reflect the ability to retain syntactic or propositional information. Consequently, there would be no reason to suspect that memory span would relate to the ability to retain syntactic structures or the ability to search over a syntactic structure to map a long-distance dependency.1 Also, there would be no reason to predit difficulty in integrating semantic information from different sections of a sentence or in understanding propositionally complex sentences. In the literature on the effects of short-term memory deficits and sentence processing, two other possible grounds for assuming a connection between phonological memory and syntactic processing have been proposed. One is that phonological memory appears particularly useful for retaining order information (Martin & Caramazza, 1982). Since order information is crucial for syntactic analyses, a disruption of phonological memory should impair syntactic analyses. However, once the parsing process begins, structure is assigned to the string of words on the basis of the order in which they are perceived. It is the assigned structure that must be maintained in order to understand the sentence, not the input order of the words. A second basis for claiming a connection between phonological memory and syntactic processing is that phonological memory would be useful for retaining inflectional morphemes and function words that have little semantic content but are essential for syntactic processing. Again, once parsing mechanisms have been applied to inflectional morphemes and function words, the semantic and syntactic information derived from these elements would be important to maintain (or the structures derived from them), not the exact phonological form of these elements. The ability of a phonological representation to preserve order information and information about elements with little semantic content would be useful characteristics for a representation that maintained information prior to parsing or for a representation that was used as a backup in case an initial analysis proved incorrect. However, as discussed earlier, the preparsing buffer is assumed to play a critical role in comprehension only under a limited set of circumstances: when a syntactic ambiguity may be resolved by looking ahead a few words or when a difficult construction causes sentence processing to lag behind the input. A backup representation would come into play only when the preferred analysis of an ambiguity had to be revised. Presumably the preferred analysis would most often be the correct one.

The role of STM

in sentence processing

403

15.3. Neuropsychological evidence on the role of short-term memory in comprehension Most brain-damaged individuals with language-related impairments perform very poorly on short-term memory tasks, and many claims have been made that comprehension deficits in various groups might be attributed to their short-term memory deficit (e.g., Saffran & Marin, 1975; Heilman, Scholes, & Watson, 1976; Vallar & Baddeley, 1984b; Ostrin & Schwartz, 1986). Most of the detailed work exploring a possible connection between memory and comprehension has been carried out on a subgroup of patients for whom the memory impairment is their most striking deficit. Several of these patients have been identified who show a common pattern in the nature of their memory deficit. Much of this pattern could be attributed to a disruption of inner rehearsal, as many of these effects mimic those obtained with normal subjects under articulatory suppression. For example, word length and phonological similarity do not affect memory span for visually presented lists (Vallar & Baddeley, 1984a; Martin et al., 1987; Martin, 1987). For auditorily presented lists, word length does not affect performance (Caramazza et al., 1981; Vallar & Baddeley, 1984a), but phonological similarity does (Vallar & Baddeley, 1984a; Martin et al., 1987). Other features of these patients' performance argue against a disruption of inner rehearsal as the sole explanation of their memory deficit. For one, these patients are often quite fluent speakers (Shallice & Butterworth, 1977; Vallar & Baddeley, 1984a; Martin et al., 1987). In one study that measured overt speech rate, the patient spoke as rapidly as normal controls (Vallar & Baddeley, 1984a). Perhaps more compelling evidence against a rehearsal deficit account is that these patients' serial recall is better with visual than auditory presentation, the reverse of the normal pattern, and they show no recency effect with auditory presentation (Warrington & Shallice, 1969; Saffran & Marin, 1975; Caramazza et al., 1981; Basso, Spinnler, Vallar, & Zanobio, 1982; Friedrich, Glenn, & Marin, 1984; Vallar & Papagno, 1986; Martin, 1987; Martin et al., 1987). The reversed modality effect and absent recency effect would not be predicted from a disruption of inner rehearsal, since articulatory suppression in normal subjects depresses performance more for visual than auditory presentation (Baddeley, Thomson, & Buchanan, 1975) and does not eliminate the recency effect with auditory presentation (Levy, 1971). Vallar and Baddeley (1984a) attributed their patients' deficit to a disruption of phonological storage. They argued that without the phonological store, subvocal rehearsal is of little use because the purpose of subvocal rehearsal is to refresh information in the phonological store. Thus, all the effects attributed to subvocal rehearsal would not be observed in a patient with a disrupted phonological store. Some possible objections might be raised to the phonological storage account. First,

404

R. Martin

an effect of phonological similarity on recall for auditory presentation would not be expected if the patient had no ability to store information in a phonological form. In fact, one case has been presented that showed no phonological similarity effect for auditory presentation (Campbell & Butterworth, 1985). Whether or not a phonological similarity effect is obtained might depend on the degree of disruption of the phonological store or on the use of alternative memory strategies by the patient. If a disruption of the phonological store is interpreted to mean that phonological information decays more rapidly than normal, then, after a given amount of time, one would still expect more distinguishing phonological information to be available to the patient for phonologically distinct items than for phonologically similar items. However, if the patient adopts some alternative coding strategy such as using visual imagery (as appeared to be the case in the patient reported by Campbell & Butterworth, 1985) or elaborating semantic relationships among items (as has been found with one of the patients tested in our lab), no phonological similarity effect would be observed. The absent recency effect in serial recall is in some ways more difficult to reconcile with a phonological store deficit account. The absence of the recency effect is usually explained on the grounds that the recency effect reflects recall from a phonologically based store. However, the recency effect with auditory presentation appears to have a specifically auditory, nonphonological component (Penney, 1975). Berndt and Mitchum (this volume, chapter 5) present a case who they claim has a phonological memory deficit but preserved auditory memory and who does show a recency effect. Penney (1989) has suggested that the patients showing the reverse modality effect and absent recency effect have both an auditory and a phonological memory deficit. However, pending any strong evidence that the patients who do not show a recency effect have memory difficulties for nonspeech auditory information, this discussion will refer to these patients as having a phonological store deficit. Nonetheless, the controversial nature of the interpretation of some of the evidence should be kept in mind.

15.3.1. Articulatory deficit and its consequences for sentence processing If an articulatory component of memory exists that can be separated from the phonological store, one might expect to see patients who have a deficit in articulatory rehearsal but who have a preserved phonological store. As discussed earlier, because articulatory rehearsal is presumed to depend on the phonological store, these patients would be expected to show many of the same memory patterns as the patients claimed to have a phonological store deficit. However, one would expect covert speech for these patients to be slow or impossible. Also, one would expect a preservation of the recency effect with auditory presentation and better performance with auditory than visual presentation. Several recent studies (Baddeley & Wilson, 1985; Baddeley, 1986;

The role of STM in sentence processing

405

Vallar & Cappa, 1987) have demonstrated that patients who are severely anarthric do not necessarily show a deficit in inner rehearsal. Thus, a disruption of overt speech abilities is not sufficient to disrupt the articulatory loop. The inner articulatory processes involved in rehearsal must occur at a more central level than those processes affected in anarthric patients with preserved language abilities. Nonfluent aphasic patients are classified as aphasic rather than dysarthric or anarthric because their nonfluency is presumed to derive from a more central deficit in language. Several studies have been carried out in this laboratory attempting to determine if nonfluent patients would show a memory pattern consistent with a disruption of inner rehearsal. All of the patients were screened to have good comprehension at the single word level. In one study of nonfluent and fluent aphasics, Martin (1987) found that with auditorily presented word lists, nonfluent aphasics were unaffected by articulatory difficulty of word lists and were less affected than fluent or normal subjects by number of syllables. In contrast to the results for the patients claimed to have a phonological store deficit, the nonfluent patients showed a recency effect and performed better with auditory than visual presentation. Feher (1987) showed that nonfluent aphasics had very slow inner speech rates relative to fluent aphasics and normal subjects in a covert articulation task, thus verifying that their nonfluency extended to covert speech. His study replicated the failure to find an effect for articulatory properties of the word lists on memory span for nonfluent patients. In addition, he showed that memory span performance decreased over an unfilled delay interval for nonfluent patients but not fluent patients, implicating a failure to rehearse in the delay interval. Feher also tested the patients on an auditory six-item recognition memory probe task, a task that appears not to draw on inner rehearsal for performance (Clifton & Tash, 1973; Chase, 1976). The nonfluent patients performed within the normal range on this task. Patients who have been claimed to have a phonological store deficit have been shown to perform poorly on a recognition probe task (Shallice & Warrington, 1970, 1974; Caramazza et al, 1981). Thus, the good performance of the nonfluent patients provides further evidence that their memory deficit involved rehearsal and not the phonological store. Some suggestions have been made that the syntactic comprehension deficits demonstrated by nonfluent agrammatic patients might be due to their short-term memory deficit (Linebarger, Schwartz, & Saffran, 1983; Ostrin & Schwartz, 1986). Linebarger et al. have argued that these patients' short-term memory deficit does not prevent syntactic parsing but does impair the mapping between syntactic structure and thematic roles. However, based on the earlier discussion of the role of memory in comprehension, if agrammatic patients' memory deficit could be attributed to an inner rehearsal deficit, few consequences for comprehension would be predicted. Martin's (1987) study examined the comprehension of sentences varying in syntactic complexity for the same patients whose memory patterns had been assessed. The sentence types included one-clause active and passive sentences (e.g., The boy pushed the

406

R. Martin Table 15.1. Relative clause sentence types {from Martin, 1987) I. II. III. IV. V.

The The The The The

woman woman woman woman woman

that that that that that

had black hair pushed the man. pushed the man had black hair. had black hair was pushed by the man. was pushed by the man had black hair. the man pushed had black hair.

girl and The boy was pushed by the girl) and center-embedded relative clause constructions. Table 15.1 provides examples of the types of relative clause constructions that were used. For each sentence type, half of the trials assessed whether the patient assigned the descriptive clause (e.g., had black hair) to the correct person or object, and half assessed whether the patient correctly assigned role relations about the action verb (e.g., whether the man pushed the woman,or vice versa). The relative clause sentences might be expected to place greater demands on memory than the one-clause sentences of several reasons. For one, the relative clause sentences were longer. Second, because all types were center-embedded constructions, the embedded clause intervened between the main clause head noun and verb, and, consequently, information about the incomplete main clause had to be retained while the embedded clause was processed. In terms of the parsing models that were discussed, a greater number of incomplete phrases would be backed up on the pushdown stack during the processing of the relative clause sentences that the one-clause sentences. Because of the intervening embedded clause, it was sometimes the case that associations of nearby words (as in the man had black hair in the Types II and IV sentences) had to be ignored in favor of longdistance dependencies (as in The woman that pushed the man had black hair). The nonfluent patients were divided into those who were agrammatic speakers and those who were not. Both groups showed similar degree of impairment on span tests and patterns consistent with a disruption of inner rehearsal. The agrammatic speakers showed considerable difficulty in understanding the role relations specified by syntactic information in even the simplest sentences. For example, AK, who had a memory span of 2.2 items, scored 57% correct on the reversible active and passive sentences and 54% overall on the relative clause sentences. However, the remaining nonfluent patients showed only a mild impairment in their ability to understand even the most syntactically complex sentences. For example, MM, who had a memory span of 2.2 items, scored 88% correct on the reversible active and passive sentences and 93% correct on the relative clause sentences. Thus, the severe deficits of the agrammatic patients could not be attributed to their memory deficit. That is, if the memory deficit of the agrammatic patients impaired the mapping between syntactic structure and thematic roles, one would have expected to see similar comprehension deficits in the other nonfluent patients with similar memory deficits. The good comprehension of the

The role of STM in sentence processing

407

other nonfluent group implies that the comprehension deficit of the agrammatic patients has to be attributed to some source other than their memory deficit. This study does not rule out the possibility that the mild degree of impairment of the other nonfluent patients could be attributed to their articulatory memory deficit. However, the evidence does not favor this interpretation. For one, some of the fluent patients who had larger memory spans and showed evidence of rehearsal demonstrated nearly identical patterns on the comprehension test. Second, the nonfluent patients showed a similar level of performance on the one-clause sentences and on the relative clause sentences. If memory were implicated in their mild comprehension impairment, worse performance might have been expected on the relative clause sentences. It was proposed that one role of articulatory rehearsal in comprehension might be to keep the verbatim form of a sentence active until a response could be made. It was hypothesized that this would be useful only in cases in which the propositions in the sentences were difficult to remember. Martin and Feher (in press) addressed this issue by comparing the effects of presentation mode on comprehension of syntactically simple sentences that varied in the number of semantic propositions and of sentences varying in syntactic complexity that were propositionally equivalent. In the sentences with varying numbers of content words, the same words were used repeatedly in different combinations across all the sentences. In the sentences varying in syntactic complexity there was much less repetition of content words. Three different presentation modes were employed: auditory, unlimited visual, and limited visual. In the unlimited visual condition, the sentence was presented on a card that remained in view until the subject made a response. In the limited visual condition, the words were presented one at a time on a computer screen at a rate of one word per second. One would expect that an ability to rehearse a verbatim form would be useful in the auditory and limited visual conditions, but not in the unlimited visual condition. Two nonfluent patients who had disrupted inner rehearsal ability and five fluent patients who showed evidence of an ability to rehearse were tested. The nonfluent patients had slow inner speech rates, showed no effect of articulatory difficulty on span, and showed decreasing performance during an unfilled delay. The fluent patients had speech rates close to those of normal subjects, performed worse on difficult-to-articulate than on easy-to-articulate items, and showed no effect of delay. The syntactically simple sentences were similar to those on the first four sections of the Token Test (DeRenzi & Vignolo, 1962). (See Table 15.2.) The subjects' task was to point to the tokens as instructed in each sentence. For the limited visual presentation mode, the tokens were not in view until the end of the sentence. For auditory presentation, the subject looked at the experimenter as the sentence was spoken and then viewed the tokens. One patient made a substantial number of errors in the unlimited visual condition, and was eliminated from further analyses. The remaining patients made few errors in the unlimited visual condition. Performance on the auditory

408

R. Martin Table 15.2. Examples of sentences with simple syntactic structure and varying numbers of content words (from Martin and Feher, in press) 1. 2. 3. 4.

Touch Touch Touch Touch

the the the the

red square. large white circle. green square and yellow circle. small yellow circle2 and the large green square.

Table 15.3. Memory performance and percentage correct on token sentences for unlimited and limited presentation conditions (based on Martin and Feher, in press) Memory spanfl

Token sentence presentation condition Unlimited

Limited b

Difference

No rehearsal

NB

JS

3.3 3.7

92 98

64 77

28 21

3.0 3.5 4.4 4.6

96 98 100 100

56 56 88 91

40 34 12 9

Rehearsal

A? AB MW WH

"Estimated set size at which the patient would score 50% lists recalled in correct order. ^Average of performance on limited visual and auditory conditions.

condition and the limited visual condition was similarly impaired. For both of these presentation modes, performance decreased with increasing numbers of content words. Averaging across the two limited conditions, the patients' scores ranged from 92% to 100% correct on the Type I sentences and from 4% to 84% correct on the Type IV sentences. Performance on the Type IV sentences was highly correlated with memory span (r = .92, p < .05). Table 15.3 shows performance on the unlimited condition and the average of performances on the limited visual and auditory conditions. As these data indicate, the difference between performance in the unlimited and limited conditions was also highly related to the patient's memory performance (r == — .94, p < .05). However, this was the case whether or not the patient's memory deficit was due to a disruption of inner rehearsal or some other factor. Two of the four fluent patients who showed evidence of rehearsal but very impaired memory performance also performed very poorly on the limited conditions relative to the unlimited condition.

The role of STM in sentence processing

409

Table 15.4. Examples of sentences with varying syntactic structure matched in content words (from Martin and Feher, in press) 1. The boy carried the girl that had red hair. 2. The boy that carried the girl had red hair. 3. The boy that the girl carried had red hair.

While the four types of token sentences differed in number of content words, they also differed in syntactic structure. Thus, it might be argued that the results were due to the confounded factor of increasing syntactic complexity rather than the increasing numbers of content words. In order for this factor to explain the correlation between memory span and the difference between performance in unlimited and limited presentation conditions, it would have to be argued that the memory demands that derived from increasing syntactic complexity were minimized by having the sentence remain in view. The results for the sentences that varied in syntactic complexity but were matched in content words should help to determine if increasing syntactic complexity could cause a difference between limited and unlimited conditions. The sentences varying in syntactic complexity but matched in content words were three types of relative clause sentences: right-branching, center-embedded subject relative, and center-embedded object relative (see Table 15.4). Although the same content words appeared in the group of three sentences matched for content words, these same words were unlikely to appear in other sentences. The subjects' task was to choose from two pictures the one matching the sentence. On half the trials, the incorrect picture depicted a reversal of agent and object with respect to the action verb. On the other half, the incorrect picture depicted the descriptive clause as describing the incorrect person. Previous evidence indicates that the right-branching sentences should be the easiest, the subject relative the next easiest, and the object relative the most difficult to understand (De Villiers, Flusberg, Hakuta, & Cohen, 1979). The greater difficulty of the center-embedded subject relative than the right-branching form might be explained on the grounds that the role of the head noun with respect to the matrix verb cannot be determined until past the embedded clause in the center-embedded form. In the rightbranching form, this role can be determined on the next word. As discussed earlier, the greater difficulty of the object than the subject relative forms can be explained on either memory (Wanner & Maratsos, 1978) or processing (Ford, 1983) grounds. Thus, the rank ordering of difficulty of these sentences might be predicted on the basis of the working memory demands involved in the processing of syntactic structure. If these working memory demands can be lessened by having the sentence remain in view, then one might expect to see better performance with the unlimited than limited modes, as

410

R. Martin

Table 15.5. Memory performance and average percentage correct on right-branching and center-embedded subject relative sentences for unlimited and limited presentation conditions {based on Martin and Feher, in press) Memory span a

Sentence presentation condition Unlimited

Limited b

Difference

No rehearsal

NB

JS

3.3 3.7

59 67

SO 77

-21 -10

3.0 3.5 4.4 4.6

67 S3 100 80

63 75 94 84

4 8 6 -4

Rehearsal

AP AB MW WH

"Estimated set size at which the patient would score 50% lists recalled in correct order. Average of performance on limited visual and auditory conditions.

was found for the token sentences. On the other hand, if the memory demands due to syntactic processing are equivalent for auditory and reading comprehension, as was argued previously, then no effect of presentation mode would be expected. Unlike what was found with the token sentences, performance with unlimited visual presentation did not approach ceiling. However, all but one of the patients scored above chance on the right-branching and subject relative forms, though several patients scored at chance on the object relative forms. Patients' performance was predicted solely by syntactic complexity. None of the patients showed any consistent pattern of decrement (or improvement) on the limited visual or auditory conditions compared to the unlimited visual condition, and there was no interaction between complexity and mode of presentation. Table 15.5 shows the average of the patients' performance on the right-branching and center-embedded relative clause sentences for the unlimited visual condition and the average of the limited visual and auditory conditions. (Performance on the centerembedded object relative clauses was not included because of the floor effect for several subjects in the unlimited visual condition.) Unlike the case for the token sentences, there was no relation between memory span and the size of the difference between unlimited and limited presentation modes. Although there appears to be some relation between memory span and overall performance on the sentences, there is an exception in that AB, who had one of the lower scores on the memory test, performed nearly as well as the patient with the highest memory span on these relative clause sentences. Also, NB scored 92% correct on the right-branching and center-embedded subject relative forms

The role of STM in sentence processing

411

with auditory presentation despite having one of the lowest memory spans. It should also be recalled that the nonfluent patients in the Martin (1987) study who were not agrammatic scored at a high level on similar relative clause sentences despite very restricted memory spans. The results of Martin and Feher's study suggest that the memory capacities tapped by a memory span task are useful for keeping in mind sentences with many content words in which there is little necessary connection among the words. This result should perhaps not be surprising, since in a memory span task, one is called on to remember lists of unrelated words. The fact that memory span rather than the ability to rehearse predicted the decrement in performance with auditory or limited visual performance suggests that whatever the composite of abilities tapped by a memory span task, all are useful for remembering these types of sentences. It might be questioned whether this relation to memory span resulted because patients who were generally more impaired performed worse on the memory span task and worse on the more difficult limited presentation modes. However, one of the fluent patients, AB, who had a very low memory span and showed a large effect of limited versus unlimited presentation mode, scored the highest of the patients on a vocabulary test (above the mean for normal subjects) and obtained one of the higher scores on the relative clause test. For the sentences varying in syntactic complexity, the difficulty for the patients was in being able to derive the correct propositions rather than in remembering the propositions, as evidenced by the difficulty the subjects had even with unlimited visual presentation. The poor performance observed for most of the patients could be attributed to a disruption of parsing mechanisms per se or to reduced working memory capacity specific to syntactic processing. The failure to find a difference between limited and unlimited presentation modes indicates that whatever propositions could be derived from these sentences, they were not subject to the same forgetting as for the token sentences. Arguably, the greater distinctiveness of the propositions across sentences made them more memorable. The absence of an effect of presentation condition is also consistent with the argument that whatever the memory demands that derive from syntactic complexity, they are the same for auditory and reading comprehension. That is, these memory demands cannot be minimized by having the opportunity to reread a sentence.

15.3.2. Phonological deficits and their consequences for sentence processing Several suggestions were made concerning the possible roles for phonological memory. Data from this laboratory have been collected relevant to many of these. One possible role is a preparsing buffer for maintaining information downstream in a sentence while the analysis of an earlier portion is completed. A second role is to serve as a backup

412

R. Martin

representation in case an ambiguity had originally been misinterpreted and the sentence had to be reanalyzed. A third is to serve as an adjunct to other memory records to aid in verbatim repetition of sentences. Finally, since articulatory rehearsal is assumed to depend on an intact phonological store (Vallar & Baddeley, 1984a), sentenceprocessing mechanisms dependent on inner rehearsal would also be indirectly dependent on phonological storage. Two other possible roles - that of a pre-parsing looking-ahead buffer for syntactic analysis and as a means of maintaining lexical items in syntactic structure - have not been directly addressed by our research. However, some of our data as well as data from others bear on these issues. A discussion of these two issues will be deferred until the end of this section. Investigations of the role of phonological memory in sentence processing have been carried out on two subjects - one an aphasic patient, EA, and the second a learning disabled child, GI. Data discussed in detail elsewhere for EA (Friedrich et al., 1984; Friedrich et al., 1985) and GI (Martin et al., 1987; Jerger, Martin, & Jerger, 1987) demonstrate that these two individuals showed memory patterns like those shown by other patients claimed to have a disruption of the phonological store. It should be noted at the outset that EA had some degree of difficulty in discriminating speech sounds differing in a single phonetic feature (Friedrich et al., 1984). GI had a more subtle speech perception deficit that could be demonstrated only for speech sounds presented in noise. In quiet, his perception was normal even for synthetically generated speech sounds differing only in voice onset time (Jerger et al., 1987).

The downstream phonological buffer hypothesis

In order to test the downstream phonological buffer hypothesis, both subjects were tested on auditory versus unlimited visual presentation of the center-embedded relative clause sentences shown in Table 15.1. If they had difficulty in retaining phonological information past the embedded clause while processing relations in the embedded clause, one would expect poorer performance on the sentences with embedded clauses that took the longest time to process. The embedded clauses were either active subject relative forms, passive subject relative forms, or object relative forms. Since passives take longer to process than corresponding actives (Forster & Olbrei, 1973), and since object relatives take longer to process than subject relatives (Ford, 1983), one would expect the types IV and V sentences to place more demands on a downstream phonological buffer than Types I—III. A critical comparison here is performance on the Type III sentences versus the Type IV sentences for auditory versus visual presentation. Both contain a passive form. However, in Type IV the passive is in the embedded clause, whereas in Type III the passive appears at the end of the sentence. The Type IV sentences should cause more difficulties than the Type III sentences because in the Type III sentences there is no

The role of STM in sentence processing

413

material past the point at which the difficult processing is carried out. Worse performance on Type IV than on Type III would be predicted only for auditory presentation, however. With visual presentation, the remainder of the sentence past the embedded passive will remain available on the page no matter how long the processing of the passive takes. It might be objected that worse performance on the Type IV than on the Type III sentences could be due to more complex processing demands for the Type IV sentences. That is, although both types involve processing a passive, in the Type IV sentences, the passive must be processed while at the same time maintaining the head noun phrase to link up with the matrix verb. Consequently, worse performance on the Type IV than on the Type III sentences might reflect restricted capacity for processing syntactic information. However, restricted capacity for syntactic processing should result in the same pattern for both auditory and visual presentation. Data discussed earlier suggested that unlimited visual presentation does not help to overcome the memory demands resulting from syntactic processing (Martin & Feher, in press). The results for GI conformed to the predictions of the downstream phonological buffer hypothesis as he performed at a normal level for Types I, II, and III but below normal level on Types IV and V with auditory presentation. With visual presentation his performance was within normal range for all sentence types (Martin et al, 1987). With auditory presentation, EA had more difficulty than GI with the Types IV and V sentences, but like GI, EA performed at a high level on the Type III sentences (Martin, 1987). She was also tested on visual versions of these sentence types, but these data were not reported in the Martin (1987) study. In addition, she was later tested on another set of sentences in which passive and object relative subordinate clauses appeared at the end of a sentence (e.g., It was the boy with red hair that was pushed by the girl, and It was the boy with red hair that the girl pushed). Good performance would be expected with this second set of sentences for both auditory and visual presentation if the downstream phonological buffer hypothesis is correct. A summary of results on all the relative clause sentence types is shown in Table 15.6. For the sentences with a passive action clause, poor performance (58%) was obtained only for the embedded passive form with auditory presentation. When the subordinate passive clause was at the end of the sentence (e.g., It was the boy with red hair that was pushed by the girl), EA performed at a higher level with auditory presentation (84% correct) and scored 100% correct with visual presentation. These results are entirely consistent with the downstream phonological buffer account. However, she performed at chance for the object relative forms for both auditory and visual presentation even when the relative clause was at the end of the sentence. Her difficulty with the object relative form suggests that she had a mild syntactic deficit that made it impossible for her to decipher the role relationships in this difficult syntactic form irrespective of the modality of presentation or the position of the embedded clause in the sentence.

414

R. Martin

Table 15.6. Percentage correct on sentence-picture matching for patient EA Example of sentence type

Auditory

Visual

Active action clause The boy that had red hair pushed the girl. The boy that pushed the girl had red hair. It was the boy with red hair that pushed the girl.

92 92 100

S3 92 100

Passive action clause The boy that had red hair was pushed by the girl. 92 The boy that was pushed by the girl had red hair. 58 It was the boy with red hair that was pushed by the girl. 84 84

92 92 100

Object relative The boy that the girl pushed had red hair. It was the boy with red hair that the girl pushed.

42 50

67 50

Table 15.7. Examples of sentences on garden path test Sensible The congressman fired all of his staff after learning about the payoffs. Anomalous They were running along a path near the grove where Sue had a checking account. Garden path I was afraid of Ali's punch because it contained too much alcohol. The lawyer decided to take the case because it was large enough to hold all his papers.

Recovering from the misinterpretation of an ambiguous word

A test employing garden path sentences like those shown in Table 15.7. was devised to test the idea that a phonological record of a sentence is used when an auditorily presented sentence has to be reanalyzed following an incorrect interpretation of an ambiguous word. In this task, EA was asked to determine whether a sentence was sensible or not. Sensible sentences that were not garden path sentences, garden path sentences, and anomalous sentences were included. The garden path sentences were lexically rather than structurally ambiguous. The practice trials included some garden path sentences in order to alert EA and the control subjects that some of the sentences might initially seem nonsensical but could have a sensible interpretation on second thought. EA scored 100% correct on the sensible sentences, 80% correct on the garden path sentences, and 80% correct on the incorrect sentences. Although her performance was below the mean of the normal controls, it was within the normal range.

The role of STM in sentence processing

415

This test represented a first attempt at examining whether recovering from a misinterpretation requires the ability to retain a verbatim phonological record of the sentence. A further test is being developed in which the distance between the ambiguous word and the disambiguating context and the strength of the context preceding the ambiguous word are varied systematically. A second test is being developed that uses syntactic rather than lexical ambiguities. These initial data suggest that a phonological record is not needed to recover from the misinterpretation of an ambiguous word. Sentence repetition

A third question investigated with EA was the role of a phonological code in repetition. An earlier study by Friedrich et al. (1985) investigated the influence of syntactic structure and semantic plausibility on EA's ability to repeat relative clause sentences. The comprehension data described earlier indicate that EA could understand reversible center-embedded subject relative forms but had difficulty understanding reversible center-embedded object relative forms. On the repetition task, EA repeated only 1 of the 24 center-embedded object relative forms verbatim. Her ability to repeat the subject relative center-embedded forms depended on the semantic constraints among the content words. On irreversible sentences (e.g., The girl that is reading the book is sleepy), she repeated 7 out of 8. verbatim. On reversible sentences (e.g., The cat that is biting the dog is black), she repeated only 3 out of 8 verbatim. Her errors included single-word substitutions and omissions, one reversal of agent and object, and one sentence for which she repeated only the first two words. On implausible sentences (e.g., The horse that is riding the man is brown), she repeated only 2 of 8 correctly. Her productions here were farther from the target, as she sometimes changed the sentence structure and often reversed the roles of agent and object to make the sentence more sensible. These data suggest that a phonological record may not be necessary for verbatim sentence repetition if certain conditions are met. First, the patient must be able to process the syntactic structure of the sentence. Second, there must be strong semantic constraints on the possible relations of the content words; and, third, these relations must be consistent with those implied by sentence structure. If any of these conditions is not met, repetition breaks down sooner than it would for subjects with a normal ability to retain phonological information. It is important to note that these same conditions need not be met to ensure comprehension as EA was able to understand reversible center-embedded subject relative forms that she had difficulty repeating. (She has not been tested on her ability to understand implausible sentences with this syntactic form.) Further testing has indicated that syntactic constraints in addition to semantic constraints influence EA's ability to repeat. Table 15.8 shows examples of the sentences that were used, and Table 15.9 shows her repetition performance for complements,

416

R. Martin Table 15.8. Examples of sentences used in repetition task Complement The mother told the boy to wash the dishes. Adverbial clause The boy did his work before watching television. Conjoined The baby fell asleep, but the noise woke her up.

Table 15.9. Percentage of sentences repeated correctly and percentage of sentences with one word or paraphrase errors for patients EA and AB Errors Correct

One word

Paraphrase

Irreversible subject relative Complements Adverbial clause Conjoined

88 10 0 0

12 70 0 0

0 10 86 63

AB Irreversible subject relative Complements Adverbial clause Conjoined

88 70 86 12

12 20 14 88

0 0 0 0

EA

adverbial clauses, and conjunctions, and also the center-embedded subject relative forms described earlier. These sentences were similar in length and all contained content words with semantic constraints on possible role relationships. Under the heading of errors inTable 15.9, one-word errors included the substitution, deletion, or addition of one word to the sentence. Errors of two or more words that were not paraphrases are not shown in the table. The likeliest response for the complements was to add a that to the sentence but otherwise repeat the sentence correctly. The likeliest response for the adverbial clause and conjoined sentences was to produce a sentence that preserved the gist of the original but that deviated substantially from the surface form of the original. For example, for the sentence The boy did his work before watching television, EA said The

boy did his work before he watched TV. As in this example, she was likely to repeat the first clause correctly and paraphrase the second.

The role of STM in sentence processing

417

These data might be explained if it is assumed that at the time of repetition EA has a propositional representation of the sentence plus a verbatim representation of the first few words. If these first few words constrain the possible syntactic structure for the remainder of the sentence (as in the center-embedded relative clauses and the complements), then EA is likely to repeat the sentence accurately. If, however, these first few words place few constraints on the possible syntactic structure of the remainder (as in the adverbial clauses and conjoined sentences), the production is more likely to deviate from the original. EA's ability to repeat the beginning but not the end of sentences is remniscent of her pattern of a primacy but no recency effect in serial recall. Based on memory data, it seems unlikely that this primacy effect in sentence repetition is based on storage in a phonological form. Because it is more difficult to defend interpretations based on associations rather than dissociations of function in neuropsychological research, one might question whether this repetition pattern has any necessary connection to EA's phonological memory deficit or whether any patient with restricted memory span, no matter what the source, would show a similar pattern. Evidence in this regard has been obtained from patient AB, who on several serial recall tests has shown a memory span very close to EA's but who shows signs of a normal phonological store. He performs better with auditory than visual presentation and shows a recency effect as well as phonological similarity effects with both auditory and visual presentation. The source of his memory span deficit is something of a puzzle, as he is a fluent speaker, sometimes shows evidence of rehearsal, and has word recognition and naming abilities above the mean for normal subjects his age. Perhaps some nonphonological lexical component has been affected in this patient. As can be seen in Table 15.9, AB showed a very different pattern than did EA on the adverbial clause and conjoined sentences. He did not paraphrase any of the sentences, but either repeated them correctly or made single word errors. The errors that he did make tended to occur in the middle of the sentence rather than at the end.

Comprehension of sentences requiring rehearsal

A final question investigated with EA was whether she would have difficulty with the auditory version of the token sentence test that was presumed to depend on an ability to rehearse the verbatim representation of the sentence. With unlimited visual presentation, EA scored 100% correct. With auditory presentation, her scores declined from 93% correct on the Type I sentences to 42% correct on the Type IV sentences, for an average of 61% correct overall. This level of performance was very similar to that observed by Martin and Feher (in press) for limited presentation modes for other patients with memory spans in the range of EA's (see Table 15.3).

418

R. Martin

Involvement of phonological memory in syntactic parsing

Two possible roles for phonological memory in syntactic parsing were suggested: one as a lookahead buffer and a second as a means of maintaining the lexical items in the parse tree. The lookahead buffer hypothesis seems difficult, although probably not impossible, to test. First, the buffer is assumed to contain only a few words in some models (e.g., Berwick & Weinberg, 1984), and consequently one would have to find a patient with a very severe phonological memory deficit in order to observe the effects of a smaller capacity. Second, one would have to provide evidence that normal subjects do use a lookahead procedure to resolve syntactic ambiguity for at least certain sentence types. Third, in order to show that patients are processing such sentences abnormally, it would probably be necessary to use some on-line reaction time or gazeduration measure rather than a sentence final measure of comprehension accuracy. That is, if patients are unable to look ahead, they would probably choose the likeliest syntactic structure at each word. If this structure later proves incorrect, they could revise the assignment using whatever means normal subjects use to revise mistaken analyses, and eventually arrive at the correct interpretation. With regard to the second hypothesis, data currently available argue against the proposal that phonological memory is used to maintain the lexical items in the parse tree. Using the Berwick-Weinberg parsing model, this hypothesis would predict that patients with phonological memory deficits should have trouble understanding any sentence in which the number of lexical items in incomplete phrases on the pushdown stack exceeded the patient's phonological memory capacity. In order to address this question, one would have to specify precisely what constitutes an incomplete or complete phrase. However, one might predict that syntactic analysis would fail completely with a patient who had a memory span of only one item. Contradicting this prediction are the data from the patient NHA reported by McCarthy and Warrington (1987a, b). NHA could recall only one-item lists with a high degree of accuracy, yet scored 94% correct on a test of reversible active and passive sentences. One might also predict that a patient with a memory span of only two items would have difficulty comprehending a complex noun phrase such as the bear that the donkey kissed, assuming that this complex noun phrase is incomplete until the end of the relative clause. However, patient BO (reported by Caplan & Hildebrandt, 1988), could reliably recall only two-word lists, yet scored 100% correct on an object manipulation task for cleftobject sentences such as It was the bear that the donkey kissed.

The role of STM. in sentence processing

419

15.3.3. Conclusions concerning the overlap of memory span and memory involved in comprehension Although there appears to be some degree of overlap between the memory capacities tapped by a span task and those involved in sentence processing, it is evident that impressive sentence-processing abilities can be observed in patients with severely restricted memory spans. In particular, the processing of syntactic structure does not appear to depend on memory span capacities. Data from our lab indicate that patients with articulatory deficits or phonological deficits in short-term memory can understand syntactically complex sentences (Martin, 1987; Martin et al., 1987). As discussed earlier, other recent studies of patients with phonological memory deficits have also demonstrated preserved syntactic processing abilities (Vallar & Baddeley, 1984b; Caplan & Hildebrandt, 1988; McCarthy & Warrington, this volume, chapter 7). The findings of earlier studies in which severe syntactic comprehension difficulties were observed in patients with phonological memory deficits (e.g., Saffran & Marin, 1975; Caramazza et al., 1981) might be attributed to several factors. First, it is possible that these patients had syntactic processing deficits per se in addition to their memory deficit. Second, it is possible that testing procedures increased the need to use phonological memory. For example, Saffran and Marin (1975) tested only repetition rather than comprehension. As evident in the data for EA, repetition of a certain sentence form may be impaired even though comprehension of that form is not. In the Caramazza et al. study, the patient had to select from four pictures in a sentence—picture matching comprehension test. In many other studies, only two picture choices have been presented (e.g., McCarthy & Warrington, 1987b; Martin, 1987). It is possible that having to select from a larger number of pictures causes a reliance on a backup phonological memory representation to keep the sentence in mind until the correct choice can be made. (See Vallar & Baddeley, 1984b, and Vallar, Basso, & Bottini, this volume, chapter 17, for related discussions.) Another aspect of sentence processing that appears not to depend on memory span capacities is the ability to reinterpret a sentence when subsequent context indicates that the interpretation of an earlier lexically ambiguous word was incorrect. It was hypothesized that a surviving phonological record might be used as input to the reanalysis. The data from EA suggest that this is not the case. The question then becomes what kind of representation is used to support the reanalysis. An argument can be made that phonological record alone would be of little use, since there would be nothing to prevent the reanalysis from coming up with the same mistaken interpretation that was derived from the first pass. The eye movement data from reading comprehension indicate that subjects' eyes regress to the ambiguous word after realizing that their initial interpretation was incorrect (Carpenter & Daneman, 1981). Thus, there must have been some tag in memory indicating which word was

420

R. Martin

ambiguous. Assuming that a similar process occurs in auditory comprehension, a second meaning for the tagged word could be accessed that would now have been primed by the subsequent context. Rather than reanalyzing the syntactic and semantic structure of the sentence, only those propositions involving the ambiguous word would have to be reinterpreted. Of course, if a structural rather than lexical ambiguity was the source of the garden path, syntactic structure would have to be recomputed. Because structurally ambiguous sentences were not tested with EA, no conclusions can be drawn about whether phonological memory would be needed for revising syntactic structure. One aspect of sentence processing that appeared strongly related to memory span was the ability to remember long sentences in which size and color adjectives were arbitrarily paired with shape nouns. It would be difficult to argue that patients could not understand these sentences, as performance was nearly perfect with unlimited visual presentation and quite good with auditory presentation if only one token was mentioned. Performance began to break down when two tokens were mentioned, each preceded by one adjective, and was very poor for two tokens, each preceded by two adjectives. The degree of breakdown was strongly correlated to memory span. The source of the reduced memory span — impaired rehearsal, a phonological store deficit, or some other factor (as for AB) - did not seem to make any difference. As mentioned earlier, since memory span measures the ability to remember random lists of unrelated words, it should perhaps not be surprising that span relates to the ability to remember these sentence types. However, it might be argued that there should be no relation, since the patients should be able to develop the propositions implied by these sentences and remember those, as for any other sentence. Since there appears to be no relation between memory span and the ability to remember propositions from other sentence types (e.g., relative clause sentences), there should be no relation here as well. The proposal being made here is that not all propositions are equally easy to retain. Because of the arbitrary nature of the relationships and because the same colors, sizes, and shapes are mentioned over and over again in different sentences, these propositions are difficult to retain without being able to hold onto a verbatim record of the sentence. This conclusion is strongly challenged by data from patient RE (reported by Campbell & Butterworth, 1985, and Butterworth et al, 1986), who was shown to perform at a normal level on the standard Token Test (DeRenzi & Vignolo, 1962) despite showing reduced memory span and evidence of a phonological store deficit. With auditory presentation, she obtained a score of 100% correct on the part 4 sentences of this test, which were identical to the most difficult sentences Martin and Feher (in press) used. It is possible that RE's good performance on this test derived from her larger memory span. RE had a digit span of four items, while EA had a digit span of two items or less (Friedrich et al., 1984). However, evidence presented by Campbell and Butterworth (1985) indicated that RE's memory was not simply a reduced version of

The role of STM

in sentence processing

All

normal memory. The memory data from RE revealed that she had no ability to retain information in a phonological or articulatory form. Thus, one would have to argue that RE used some type of memory code other than a phonological one to carry out this task. This latter possibility is supported by data presented by Butterworth et al. demonstrating that normal subjects but not RE were impaired on the Part 4 sentences with visual presentation if they were required to suppress articulation during sentence presentation. Butterworth et al. point to the fact that RE was unaffected by suppression as evidence that she does not use the articulatory loop to perform the task and that use of the loop in not necessary for good performance. However, they do not explain why the normal subjects were impaired, if the articulatory loop is not typically relied on for their performance of this task. RE's phonological store deficit did not arise from a traumatic injury but was apparently a developmental problem. Thus, she may have developed strategies for dealing with memory tasks that are unlike those used by patients who have suffered a sudden onset of their problems due to brain damage. Perhaps RE was using a visual imagery strategy in performing the task, that is, imagining shapes of the appropriate size and color when hearing or reading the sentences and using these images to aid performance. Evidence from the memory tasks given RE indicated that she used visual coding in a manner not typically seen in normal subjects who are suppressing articulation (Campbell & Butterworth, 1985). She was unable at first to decide whether verbal sequences of numbers that had identical shapes matched or did not match (e.g., nought zero nought nought — nought nought zero nought) but quickly developed a strategy of imagining a different shape for the different names. She was so dependent on visual coding for short-term memory tasks that her memory span was actually worse when she had her eyes open than when she had them closed. Although there is no direct evidence that RE used visual imagery in the Token Test, it does seem possible that someone adept at using such a strategy could have used it for this task. (See Vallar & Baddeley, 1987, for related comments.) Other data from RE present a greater challenge for the hypothesis that phonological storage is used to hold speech downstream from the point at which sentence processing is taking place. Such storage was proposed to be useful whenever sentence processing was slowed down due to syntactic complexity or some other factor, and consequently lagged behind the input. Data consistent with this hypothesis were presented for two individuals with phonological store deficits (EA and GI) who performed better on certain types of relative clause sentences for visual than auditory presentation, and who performed better when a syntactic complexity was encountered at the end of the sentence compared to the middle. However, RE made only one error on 52 trials on the Test for Reception of Grammar (Bishop, 1982), which includes relative clause sentences (e.g., The boy the dog chases is big) similar to those that caused difficulty for EA and GI. She also performed very well (95% correct) on another test that included different types

422

R. Martin

of sentences (e.g., The bus is preceded by the train which the triangle is below and The dog which the circle is above follows the bus) that might also be expected to place demands on the phonological buffer because of the complex syntactic form at the beginning of the sentence. It would be difficult to claim that RE adopted some nonstandard strategy for comprehending these sentences, since it would be hard to imagine what kind of strategy could be used if the phonological buffer hypothesis was correct. That is, if sentences such as these cause sentence-processing mechanisms to lag behind the input, the input would have to be stored in some form. Without the benefit of attention, it would seem impossible for the input to be translated from a phonological form to some other form that RE could retain. On the other hand, if the phonological buffer hypothesis is not correct, some other explanation must be offered for the comprehension patterns of GI and EA. Perhaps their difficulties in speech perception (which RE did not have) could account for their comprehension pattern. Particularly in the case of EA, speech perception may be more of a conscious problem-solving activity than is true for normal subjects. Consequently, when processing capacity must be devoted to the comprehension of a difficult syntactic structure, there may be insufficient capacity remaining to devote to perceiving the subsequent words in the sentence. Another possibility is that EA and GI have an additional deficit in syntactic processing that causes their syntactic processing mechanisms to operate more slowly than normal. Consequently, they are more dependent than normal subjects (and apparently RE) for having some means of retaining information past a syntactic complexity until it can be processed. Thus, their comprehension patterns derive from an interaction of their syntactic and memory deficits. It is possible that data from other patients strongly suggesting a role for shortterm memory in syntactic deficits (Ostrin & Schwartz, 1986) could also derive from an interaction of a syntactic deficit with a memory deficit. If it is not necessary to maintain information past a syntactic complexity in a phonological form, then either of two conclusions might be drawn with regard to normal sentence processing. First, it is possible that sentence processing occurs so quickly in normal subjects, even for unusual syntactic forms, that sentence processing never lags behind the input. Second, although attentional processing does on occasion lag behind the input, sufficient syntactic and semantic processing of the remainder of the sentence can be carried out at an unattended level that a phonological record is not necessary. There would appear to be no data pertinent to deciding between these possibilities. Before the downstream phonological buffer hypothesis is ruled out entirely because of RE's comprehension abilities, it would be important to have further information from normal subjects about the time course of processing relative to input, and the degree of processing that can be carried out on unattended sentence forms. The final role for phonological memory that was investigated here was as an aid in

The role of STM in sentence processing

423

sentence repetition. The sentence repetition data obtained from RE as well as from a patient studied by Saffran and Marin (1975) are consistent with an important role for the retention of a phonological code in sentence repetition. The data from our lab on EA suggested that verbatim repetition may be possible for sentences exceeding memory span if there are semantic constraints on the possible relations of the content words and if the syntactic structure is such that remembering a few words verbatim constrains the possible syntactic structure of the remainder. If these conditions are not met, repetition is likely to retain the gist of the sentence but deviate markedly from the surface form. These data imply that repetition for EA is a constructive process relying on the derived meaning of the sentence plus whatever surviving phonological information is available. The same could be said for repetition in normal subjects except that they have more surviving verbatim phonological information to guide the reconstructive process. In summary, the strongest relationship between memory span and sentence processing appears to be that verbatim recall of sentences relies on the same memory abilities tapped by memory span. Verbatim recall of sentences is obviously useful in the case of sentence repetition, but also appears useful in a comprehension task if the derived meaning of a sentence is difficult to retain, as in the case of the token sentences. For both repetition and comprehension, the memory abilities tapped by memory span are most useful when the sentences approach random lists, that is, when there are few strong or predictable relationships among the words. Other hypotheses about the role of short-term memory in sentence processing that might apply to sentences more like those one would encounter in everyday conversation (i.e., sentences with greater semantic coherence) were not supported. In particular, a phonological record does not appear necessary to support syntactic analysis, nor to recover from the misinterpretation of a lexically ambiguous word. The downstream phonological buffer hypothesis received some support, but contradictory evidence has been reported by other researchers (Butterworth et al, 1986). Even if the downstream phonological buffer hypothesis proves correct, it is evident that a great deal of sentence processing can be carried out despite very impaired articulatory and phonological memory capacities. Because syntactic and semantic processes are applied immediately and rapidly to the input, there is no need to maintain a large number of words in a phonological form prior to the initation of processing. Even in the case of structural ambiguity, it appears that one syntactic structure is chosen quickly. It remain possible that a phonological memory representation is needed in the case that this structure proves incorrect; however, it is clear that more than a phonological representation would be needed to prevent the same incorrect analysis from being repeated. Also, it appears that the output of syntactic analysis does not involve a phonological memory representation. Consequently, a complex syntactic structure may be maintained in working memory prior to semantic interpretation despite very limited phonological memory abilities.

424

R. Martin

Because of the degree of sentence processing that can be carried out despite very restricted memory span, a natural question then arises as to whether there is some other more substantial purpose for these articulatory and phonological memory systems involved in memory span. One plausible suggestion (Gardner, 1974; Caplan, personal communication) is that phonological memory plays an important role in language acquisition. As the child is acquiring language, it may be important for the child to be able to hear a word or series of words and be able to repeat them back. As Caplan (personal communication) has suggested, in the case of sentence comprehension, immediacy of processing may not be possible for the child because syntactic and semantic processing procedures have not yet become rapid and reliable. Thus, the child may have a need to hang on to a phonological representation of a sentence longer than an adult while trying to derive the underlying meaning. If so, then the phonological memory abilities of an adult may represent the residual of a system that was once vital to language processing but that only comes into play in exceptional situations in adult language.

Note 1. As discussed earlier, the only plausible relationship of a phonological memory system to the maintenance of syntactic structure would be as a means for maintaining lexical units in the parse tree.

References Baddeley, A. D. (1986). Working memory. Oxford: Clarendon Press. Baddeley, A. D v & 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. Baddeley, A. D., Lewis, V., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly Journal of Experimental Psychology, 36, 233—252.

Baddeley, A , Thomson, N., Buchanan, M. (1975). Word length and the structure of short-term memory. Journal of Verbal Learning and Verbal Behavior, 14, 575-589 Baddeley, A., & Wilson, B. (1985). Phonological coding and short-term memory in patients without speech. Journal of Memory and Language, 24, 490—502. Barnard, P. (1985) Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197-258). 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. Berwick, R., & Weinberg, A. (1984). The grammatical basis of linguistic performance. Cambridge, MA: MIT Press. Bishop, D. V. M. (1982). TROG Test for the Reception of Grammar. Psychology Department, University of Manchester, England. Butterworth, B., Campbell, R., & Howard, D. (1986). The uses of short-term memory: A case study. Quarterly Journal of Experimental Psychology, 38A, 705-737.

The role of STM in sentence processing

425

Campbell, R., & Butterworth, B. (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, 37, 435-475. Caplan, D., & Hildebrandt, N. (1988). Disorders of syntactic comprehension. Cambridge, MA: MIT Press. Caramazza, A., Basili, A. G., Koller, J., & Berndt, R. S. (1981). An investigation of repetition and language processing in a case of conduction aphasia. Brain and Language, 14, 235-271 Carpenter, P. A., & Daneman, M. (1981). Lexical retrieval and error recovery in reading: A model based on eye fixations. Journal of Verbal Learning and Verbal Behavior, 20, 137-160. Carpenter, P. A., & Just, M. A. (1981). Cognitive processes in reading: Models based on readers' eye fixations. In A. M. Lesgold & C. A. Perfetti (Eds.), Interactive processes in reading (pp. 177-213). Hillsdale, NJ: Erlbaum. Chase, W. (1976). Does memory scanning involve implicit speech? In S. Dornic (Ed.), Attention and performance, (Vol. 6, pp. 607-628). London: Erlbaum. Clark, H. H., & Clark, E. V. (1977). Psychology and language. New York: Harcourt Brace Jovanovich. Clifton, C, & Tash, J. (1973). Effects of syllabic word length on memory-search rate. Journal of Experimental Psychology, 99, 231-235. Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by the psychological syntax processor. In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds.), Natural language parsing (pp. 320-358). Cambridge: Cambridge University Press. Crowder, R. G. (1978). Memory for phonologically uniform lists. Journal of Verbal Learning and Verbal Behavior, 17, 73-89. DeRenzi, E., & Vignolo, L. A. (1962). The Token Test: A sensitive test to detect receptive disturbances in aphasics. Brain, 85, 665-678. De Villiers, J., Flusberg, H., Hakuta, K., & Cohen, M. (1979). Children's comprehension of relative clauses. Journal of Psycholinguistic Research, 8, 499-518. Eich, E. (1984). Memory for unattended events: Remembering with and without awareness. Memory and Cognition, 12, 105-111. Feher, E. (1987). An examination of short-term memory deficits in non-fluent aphasics. Unpublished doctoral dissertation, University of Houston, Houston, TX. Ford, M. (1983). A method for obtaining measures of local parsing complexity throughout sentences. Journal of Verbal Learning and Verbal Behavior, 22, 203-218 Forster, K. I., & Olbrei, I. (1973). Semantic heuristics and syntactic analysis. Cognition, 2,319-347. Forster, K. I., & Ryder, L. A. (1971). Perceiving the structure and meaning of sentences. Journal of Verbal Learning and Verbal Behavior, 10, 285-296. Frazier, L. (1985). Syntactic complexity. In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds.), Natural language parsing (pp. 129-189). Cambridge: Cambridge University Press. Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence comprehension: Eye movements in the analysis of structurally ambiguous sentences. Cognitive Psychology, 14, 178-210. Friedrich, F., Glenn, C, & Martin, O. S. M. (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. Gardner, H. (1974). The shattered mind. New York: Random House. Heilman, K. M., Scholes, R., & Watson, R. T. (1976). Defects of immediate memory in Broca's and conduction aphasia. Brain and Language, 3, 201-208. Hunt, E. B., Lunneborg, C, & Lewis, J. (1975). What does it mean to be high verbal? Cognitive Psychology, 2, 194-227. Jerger, S., Martin, R., & Jerger, J. (1987). Specific auditory perceptual deficit in a learning disabled child. Ear and Hearing, 8, 7S-8>6.

426

R. Martin

Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87, 329-354. Just, M. A., & Carpenter, P. A. (1987). The psychology of reading and language comprehension. Newton, MA: Allyn & Bacon. Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review, 85, 363-394. Klapp, S. T., Marshburn, E. A., & Lester, P. T. (1983). Short-term memory does not involve the "working memory" of information processing: The demise of a common assumption. Journal of Experimental Psychology: General, 112, 240-264. Klatzky, R. (1980). Human memory: Structures and processes. San Francisco: Freeman. Levy, B. A. (1971). Role of articulation in auditory and visual short-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 123-132. Linebarger, M , Schwartz, M., & Saffran, E. (1983). Sensitivity to grammatical structure in socalled agrammatic aphasics. Cognition, 13, 361-392. McCarthy, R., & Warrington, E. K. (1987a). The double dissociation of short-term memory for lists and sentences. Brain, 110, 1545-1563. McCarthy, R., & Warrington, E. K. (1987b). Understanding: A function of short-term memory? Brain, 110, 1565-1578. Marcus, M. (1980). A theory of syntactic recognition for natural languages. Cambridge, MA: MIT Press. Marslen-Wilson, W., & Welsh, A. (1978). Processing interactions and lexical access during word recognition in continuous speech. Cognitive Psychology, 10, 29-63. Martin, R. C. (1987). Articulatory and phonological deficits in short-term memory and their relation to syntactic processing. Brain and Language, 32, 159—192. Martin, R. C, & Caramazza, A. (1982). Short-term memory in the absence of phonological coding. Brain and Cognition, 1, 50-70. Martin, R. C, Feher, E. (in press). The consequences of reduced memory span for the comprehension of semantic versus syntactic information. Brain and Language. Martin, R. C, Jerger, S., & Breedin, S. (1987). Syntactic processing for auditory and visual sentences in a learning disabled child: Relation to short-term memory. Developmental Neuropsychology, 3, 129-152. Miller, G., & Selfridge, J. (1950). Verbal context and the recall of meaningful material. American Journal of Psychology, 63, 176-187. Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed capacities" view. In H. Bouma & D. G. Bowhuis (Eds.), Attention and performance X: Control of language processes (pp. 327-350). Hillsdale, NJ.: Erlbaum. Ostrin, R., & Schwartz, M. (1986). Reconstructing from a degraded trace: A study of sentence repetition in agrammatism. Brain and Language, 28, 328-345. Penney, C. G. (1975). Modality effects in short-term verbal memory. Psychological Bulletin, 82, 68-84. Penney, C. G. (1989). Modality effects and the structure of short-term verbal memory. Memory and Cognition, 17, 398-422. Pereira, F. (1985). A new theory of attachment preferences. In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds.), Natural language parsing (pp. 307-319). Cambridge: Cambridge University Press. Perfetti, C, & Goldman, S. (1976). Discourse memory and reading comprehension skill. Journal of Verbal Learning and Verbal Behavior, 15, 33-42. Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction of syntax and semantics during sentence processing: Eye movements in the analysis of semantically biased sentences. Journal of Verbal Learning and Verbal Behavior, 22, 358-374. Sachs, J. S. (1967). Recognition for syntactic and semantic aspects of connected discourse. Perception and Psychophysics, 2, 437-442.

The role of STM

in sentence processing

All

Saffran, E. M., & Marin, O. S. M. (1975). Immediate memory for word lists and sentences in a patient with auditory short-term memory. Brain and Language, 2, 420-433. Shallice, T. (1979). Neuropsychological research and the fractionation of memory systems. In L. J. Nillson (Ed.), Perspectives on memory research (pp. 257—277). Hillsdale, NJ: Erlbaum. Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech. Neuropsychologia, 15, 729-735. Shallice, T., & Warrington, E. (1970). Independent functioning of verbal memory stores: A neuropsychological study. Quarterly Journal of Experimental Psychology, 22, 261-273. Shallice, T., & Warrington, E. (1974). The dissociation between short-term retention of meaningful sounds and verbal material. Neuropsychologia, 12, 553—555. Tejirian, E. (1968). Syntactic and semantic structure in the recall of orders of approximation to English. Journal of Verbal Learning and Verbal Behavior, 7, 1010-1015. Thibadeau, R., Just, M. A., & Carpenter, P. A. (1982). A model of the time course and content of reading. Cognitive Science, 6, 157-203. Treisman, A. (1964). Monitoring and storage of irrelevant messages in selective attention. Journal of Verbal Learning and Verbal Behavior, 3, 449—459. Tyler, L. K., & Marslen-Wilson, W. D. (1977). The on-line effects of semantic context on processing. Journal of Verbal Learning and Verbal Behavior, 16, 683-692. 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, 121-142. 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, 412-427. Wanner, E., & Maratsos, M. (1978). An ATN approach to comprehension. In M. Halle, J. Bresnan, & G. Miller (Eds.), Linguistic theory and psychological reality (pp. 119-161). Cambridge, MA: MIT Press. Warrington, E. K., & Shallice, T. (1969). The selective impairment of auditory-verbal short-term memory. Brain, 92, 885-896. Watkins, M. (1977). The intricacy of memory span. Memory and Cognition, 5, 529-534. Wright, M. (1979). Duration differences between rare and common words and their implications for the interpretation of word frequency effects. Memory and Cognition, 7, 411-419.

16. Short-term memory impairment and sentence processing: a case study ELEANOR M. SAFFRAN AND NADINE MARTIN

16.1. Introduction As a task that requires the integration of temporally distributed input, sentence comprehension clearly involves short-term memory (STM) capacity. Little has been done to define the storage requirements for sentence-processing tasks, although the STM capacities underlying the retention of word list materials have been extensively studied. It is likely, however, since both tasks deal with strings of lexical items, that they have some mnestic requirements in common. This is implicit in the assumption that the phonological store, thought to be the primary vehicle for information storage in spantype tasks (see Baddeley, this volume, chapter 2), contributes to sentence processing as well (e.g., Clark & Clark, 1977). Studies that have demonstrated trade-offs between concurrent comprehension and list memory tasks (Savin & Perchonock, 1965; Wanner & Maratsos, 1978) provide prima facie support for this notion, as do indications that sentential input is held in phonological form prior to the identification of clausal units (Jarvella, 1971; but see Von Eckhardt & Potter, 1985). Additional evidence for storage capacities common to sentence processing and list retention has come from neuropsychological investigations. Brain-damaged patients with selective STM deficits, as defined by Shallice and Vallar in chapter 1 of this volume, have invariably demonstrated some degree of impairment on tests of sentence comprehension. The difficulties of these patients appear to lie, moreover in structural aspects of sentence processing, which is where access to an information store that represents a linear array of lexical items is likely to be most useful. The evidence for syntactic impairment comes from clinical assessment, where STM patients tend to perform poorly on the section of the Token Test regarded as the most difficult syntactically (e.g., Warrington, Logue, & Pratt, 1971; Basso, Spinnler, Vallar, & Zanobio, 1982), and from experimental investigations, where these patients have difficulty understanding sentences in which thematic role assignment critically depends This study was supported by Grant NS 18429 from the National Institutes of Health. A version of this paper was presented by E. Saffran at the Second Venice Conference on Cognitive neuropsychology in March 1985.

428

STM impairment and sentence processing

429

on structural information (Caramazza, Basili, Koller, & Berndt, 1981; Friedrich, Martin, & Kemper, 1985). The evidence that STM patients have difficulty with structural aspects of sentence processing has been taken as support for the notion that syntactic operations are dependent on span-type storage capacities. Thus, Vallar and Baddeley (1984) suggest that their patient PVs inability to detect structurally encoded semantic anomalies (e.g., The world divides the equator into two hemispheres) reflects breakdown of the comprehension process "when the abnormally reduced capacity of the short-term store is exceeded by material which has to be preserved in its superficial structure" (p. 138). Caramazza et al. (1981) provide a more explicit account of the comprehension deficit of their STM patient, MC. They suggest that the patient has limited ability to recover structural information because he is only able to apply his "unimpaired syntactic knowledge to a small segment of the presented material at a time" (p. 269). This account is compatible with models of sentence parsing that assume that initial structure-building operations are performed over a "window" of five or six successive items (Frazier & Fodor, 1978). The effect of the STM deficit, according to Caramazza et al., is to narrow the "window," and hence to restrict the scope of parsing operations. On this hypothesis, then, the comprehension deficit is the direct consequence of a parsing limitation. The difficulty is not detected under conditions in which lexical, contextual, or pragmatic constraints are likely to yield correct responses, but emerges in tasks where structural information is critical to sentence interpretation: in comprehension tasks that involve semantically reversible sentences, and in the detection of structurally encoded semantic anomalies, such as those used by Vallar and Baddeley (1984), where semantic and pragmatic biases dispose toward nonanomalous readings. The "narrow window" hypothesis carries the further implication that length will be an additional factor in patients' performance. Although the severity of the deficit varies across patients (see Saffran, in press, for review), the available data are generally consistent with this formulation.1 In this study of an STM patient (TI), we put the syntactic breakdown hypothesis to direct test by examining TI's ability to detect grammatical violations. This task, which does not require semantic interpretation of the input string, taps structural operations more directly than the comprehension tests heretofore used to examine sentence processing in STM patients. We show that despite TI's limited span, he remains remarkably sensitive to syntactic violations, even when the critical items are widely separated in the input string. Furthermore, an examination of TI's comprehension and repetition performance across a range of sentence types reveals a pattern that is not directly interpretable in terms of loss of a phonological record. These observations lead us to a critical examination of the role of STM processes in sentence comprehension.

430

Saffran and N. Martin

16.2. Case description TI is a right-handed 70-year-old male high school graduate who held a managerial position in a manufacturing concern until his retirement in 1970. He had a history of coronary artery disease and suffered two cerebral infarctions, in October and December of 1983. A CT scan performed in February of 1985 indicated a left hemisphere infarction involving the posterior parietal and posterior temporal regions, as well as an infarction of the inferior frontal gyms on the right. A neurological examination performed at this time revealed minimal impairment, the most significant finding being a mild bilateral ideomotor apraxia with frequent use of body part as object. Motor examination indicated a mild clumsiness of the left hand manifested by impairment in rapid, precise movements. Deep tendon reflexes were brisk bilaterally, with slight left-sided preponderance. Cranial nerve and sensory examinations were normal. TI was living independently, and, throughout the period of this study, was able to drive himself to the hospital for testing. TI had a mild hearing loss in both ears (20-40 db, up to 60-70 db at 4000 Hz), for which he wore a hearing aid in his left ear; speech discrimination as measured in the audiological evaluation was, however, unaffected (90% correct in the right ear and 100% in the left). TFs spontaneous speech was fluent and grammatical, although punctuated with hesitancies and restarts that resembled a stuttering pattern. He produced occasional literal paraphasias that almost always involved multisyllabic words. On formal language testing (Boston Diagnostic Aphasia Examination, BDAE, administered in June, 1984), TI performed above the aphasic mean except on Phrase Repetition and Complex Commands. He was not impaired on the reading subtests of the BDAE, but his writing and spelling were grossly deficient. TI achieved a score of 71 on the 85-item version of the Boston Naming Test, which is at about the 70th percentile for his age group. On the Token Test, he performed errorlessly on Parts I—III and made only a single error on Part IV; however, he scored only 50% correct on Part V, where the response hinges on the interpretation of word order in locative sentences (e.g., Put the red circle on top of the white rectangle). TI's performance on other comprehension tests provided additional evidence of difficulty in interpreting syntactically encoded material (see section 16.4). He was also found to be markedly impaired on span tasks (section 16.3). In view of the reported association between STM deficits and impairments in phonological processing (Allport, 1984; Friedrich, Glenn, & Marin, 1984), a set of phonological tasks was administered to TI. On the Goldman-Fristoe-Woodcock Test of Auditory Discrimination, TI was correct on 28/30 items on the quiet subtest, which placed him at the 50th percentile for his age group; on the noise subtest, he was correct on only 5/20, placing him in the 2nd percentile. On an experimental test of phoneme discrimination, which involved probe recognition with minimal phonemic contrasts, he

STM impairment and sentence processing

431

was correct on 19 trials out of 20. He did, however, have some difficulty on an auditory lexical decision task, where he correctly identified all of the 29 words but incorrectly judged 14 of the 31 nonwords (d' = 2.46), which differed from words in a single phoneme. He performed quite well (91% correct) in discriminating pairs of rhyming words in the auditory modality but poorly (61%) on the same task administered with written stimuli. TI therefore appears to have some impairment in phonological processing, although the nature of his difficulty is not entirely clear. In most respects, then, TI appears similar to other patients who have been identified as having selective STM impairment (see Table 1.1 in Shallice and Vallar, chapter 1), although he differs from them in having right frontal damage in addition to the left posterior lesion typical of such cases. In the studies reported here, data from patients with right hemisphere lesion are provided where available. 16.3. STM performance TI was tested on a set of tasks routinely used in our laboratory to assess immediate memory span for list materials. These included digit and word repetition tests, pointing span, and same—different matching of letter strings. Digit and word span were tested, using both auditory and visual presentation. The stimuli in all of the auditory tasks, with the exception of letter string matching, were tape-recorded. In the visual tasks, items were presented serially on a computer screen. Presentation rate was one per second throughout. The results of the STM tests are summarized in Table 16.1. TI is clearly impaired across all tasks, irrespective of material type, modality of presentation, and mode of response. In the digit span task, he shows the loss of recency that other STM patients have demonstrated in serial recall tasks (e.g., Saffran & Marin, 1975; Friedrich et al, 1984). He fails, however, to show the improved performance with visual presentation that is characteristic of most STM patients (Shallice and Vallar, chapter 1). This difference could conceivably reflect the presence of right hemisphere damage in TI. 16.4. Sentence comprehension 16.4.1. Lexical versus syntactic contrasts As a basic measure of TI's comprehension abilities, we examined his performance on a sentence-picture matching test. This test, part of the Philadelphia Comprehension Battery (Saffran, Schwartz, Linebarger, Martin, & Bochetto, 1988), employs two types of distractor trials. On half the trials (lexical condition), a lexical distractor is employed (e.g., for The dog chases the cat, a picture of a dog chasing a rabbit); on the remainder (reversed role condition), the distractor involves a reversal of thematic roles (e.g., for The dog chases the cat, a picture of a cat chasing a dog).

432

Saffran and N. Martin

Table 16.1. STM span for digits and words with varied presentation conditions and response modes Proportion correct at each serial position String length

1

2

3

4

5

Proportion strings correct

Proportion items correct

A. Auditory presentation Oral response Digits 3(n = 20) 4(« = 20) 5(n = 10) Words l(n = 10) l(n = 20) 3(n = 20) 4(w = 20)

1.00 .90 .90

1.00 1.00 1.00

1.00 .95 .70

1.00 .95 .&5 .80

1.00 .60 .40

.75 .40

1.00 1.00 .95 1.00

1.00 .90 .85 .65

.95 .85 .65

1.00 .50 .10

1.00 .88 .72

.65

1.00 .95 .40 0

1.00 .98 .73 .56

.80 .50

1.00 .85 .60 .10

1.00 .95 .89 .68

.70 .60

.40

Pointing response

Digits 2(n = 20) 3(n = 20) 4(w = 20) 5(n = 10)

.60

B. Visual presentation Oral Response Digits l{n = 20) 3(n = 20) 4{n = 20) 5(n = 20) Words 2(w = 20) 3(« = 20) 4(w = 20) 5(« = 20)

.95 1.00 1.00 1.00

1.00 1.00 1.00 .85

.95 .65 .70

1.00 .90 1.00 .85

1.00 .55 .50 .75

.45 .50 .45

.20 .55

.35 .30

.15

.95 .90 .15 0

.98 .97 .71 .65

.65

1.00 .30 .05 0

1.00 .63 .59 .60

String condition

C. Letter string matching Proportion correct

Same {n = 35)

.86

(e.g., HJFG-HJFG)

Different in 1 letter (n == 20)

.40

(e.g., BTNJ-BFNJ)

Different in order (n = l 5) (e.g., GMPH-GPMH)

.13

STM

impairment and sentence processing

433

Table 16.2. Sentence comprehension: lexical vs. reverse- role dish'actors (proportion of sentences identified correctly) Syntactic structure

Reversible

Lexical

Active (n = 10) Active + terminal phrase (n = 10) Passive (n = 10) Locative {n = 10) Object relative {n = 6) Subject relative (n = 6)

.80

1.00

60 40 60 83 67

.90 .90

Total (n = 52)

.63

1.00 1.00 .S3

.94

Methods The materials for this task include six different sentence types: simple active declaratives, actives with a terminal adverbial phrase that is irrelevant to the picture choice, passives, locatives, object relatives, and subject relatives. There are 10 lexical and 10 reversed role items for each structural type except for the subject and object relatives, where there are six trials per condition. The sentences were read to TI, who responded by pointing to one of the two picture choices. Each sentence was presented only once.

Results and discussion

The results are summarized in Table 16.2. Although TI's overall performance on the reversed role condition was above chance (binomial test z = 2.26, p = 0.01), his performance on these syntactic contrast was quite poor and significantly worse than on the lexical condition (McNemar Test: %2[1] = 16.4, p < .001), which posed little difficulty for him. TI's difficulty with sentences clearly involves syntactic information. We sought to characterize this deficit further in the studies that follow.

16.4.2. Semantic anomaly TI's performance pattern on sentence—picture matching tests resembles that of agrammatic aphasics, who also have difficulty on structural contrasts but not on contrasts involving lexical information (e.g., Caramazza & Zurif, 1976). There are a number of possible explanations for this "asyntactic" comprehension pattern (cf. Schwartz, Linebarger, & Saffran, 1985). As suggested earlier, one account assumes impairment in

434

Saffran and N. Martin

the recovery of syntactic information, that is, a deficit at the level of parsing the input string. Alternatively, the difficulty could arise subsequent to parsing, in utilizing the structural information supplied by the parser to make thematic role assignments, that is, in "mapping" from syntactic structure to semantic representations (Schwartz et al., 1985). In a study directed primarily at the syntactic comprehension deficits of agrammatic aphasics, Schwartz, Linebarger, Saffran, and Pate (1987) sought evidence that would differentiate between these alternatives. Using the detection of semantic anomalies involving semantic role assignments (e.g., The worm swallowed the bird) as their comprehension measure, Schwartz et al. examined the effects of two forms of syntactic complexity that, they reasoned, would have different implications for parsing and mapping. The first complexity manipulation involved "padding" the sentence with phrasal and clausal material that is irrelevant to the semantic anomaly; if parsing the sentence is the problem, padding should make the task even more difficult. The other manipulation ("moved arguments") involved the use of structures (passives and clefts) in which one or more of the relevant NP arguments has been moved out of its canonical surface structure position. This presumably complicates the mapping process by decreasing the "transparency" between surface structure and thematic roles; thus, in the passive, the surface structure subject does not correspond to the "deep structure" subject that, in most cases, fills the agent role. Where the source of difficulty in sentence comprehension lies in the mapping process, this loss of "transparency" should result in further performance decrements. TI was one of a group of fluent aphasics who were contrasted with the agrammatic subjects in the Schwartz et al. (1987) study. His data, along with data for right hemisphere controls, are provided here.

Methods Two types of anomalies were contrasted: In one set (structure based), anomalies arose out of the thematic role reversals encoded syntactically (e.g., #The worm swallowed the bird); in the other (lexically based), the sentences were implausible under any assignment of thematic roles (e.g., #The cat divorced the milk). The basic set of lexical and structure-based anomalies comprised simple active declarative sentences, as in the preceding examples, and locatives (e.g., #The freezer was in the ice cream). The basic sentences were elaborated in two different ways. The moved-arguments elaboration involved a structural change such that the grammatical subject was no longer the agent; the padding condition involved the addition of lexical content, with a corresponding increase in structural complexity, but no change in the relationship between grammatical role and thematic role. Examples are given in Table 16.3. Each of 25 basic

STM impairment and sentence processing

435

Table 16.3. Semantic anomaly task conditions Condition Structure-based Basic Moved-arguments Padding

Lexical Basic Moved-arguments

Examples The bird swallowed the worm. #The worm swallowed the bird. It was the worm that the bird swallowed. #It was the bird that the worm swallowed. As the sun rose, the bird in the cool wet grass swallowed the worm quickly and went away. #As the sun rose, the worm in the cool wet grass swallowed the bird quickly and went away. The teacher spoke to the student. #The letter spoke to the package. It was the letter that his parents sent. #It was the package that the letter spoke to.

Padding

After the class ended, the history teacher spoke to the student who had recently transferred into the school from another state. #The letter from the child's aunt spoke angrily to the large and carefully wrapped package by the window.

sentences in the structure-based set (e.g., #The bird swallowed the worm) was subjected to both the moved-arguments and padding manipulations, and a plausible and implausible version of each was constructed. Each of 13 implausible (e.g., #The cat divorced the milk) and 12 plausible sentences (e.g., The man saw the woman) was similarly subjected to both manipulations. The various conditions are outlined in Table 16.3. There were 225 sentences in all, 50 in each of the structure-based conditions and 25 in each of the lexical conditions. The sentences were tape-recorded for presentation. After hearing each sentence twice, the subject had to indicate whether it was "silly" or "OK."

Results and discussion

TI's performance on this task, along with that of a group of right hemisphere lesioned controls (N= 10), is summarized in Figure 16.1.2 Although his performance is comparable to that of controls on the lexical condition, he is clearly impaired relative to the right hemisphere patients on the structure-based condition. As in the

1.0

f

0.9

Proportion correct

1.0 0.9

08

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0.0

0.0 Base

Movedarguments Lexical violations

Padded

Baa

Movedarguments Structure-based violatic

Padded

Sentence Condition

Figure 16.1. Performance of TI and right hemisphere-damaged controls on the semantic anomaly study.

H

TI

•

RH Controls

STM impairment and sentence processing

437

sentence—picture matching task in the preceding section, he had some difficulty even with the relatively simple sentences in the structure-based basic condition. Of most interest is the different effect of the two forms of elaboration. While padding did not produce a significant decrement in performance relative to the basic condition (McNemar Test: / 2 [1] = .1), there was a significant effect of the complexity introduced by the moved-arguments manipulation (McNemar Test: / 2 [1] = 4.1, p < .05). These data substantiate the result of the previous study (section 16.4.1), pointing again to a specific impairment involving structural information. TI's performance pattern on the Semantic Anomaly test allows us to characterize this deficit more precisely. If his problem lay in parsing the input strings, padding should have resulted in decreased performance relative to the basic condition. In (1), for example, failure to recover constituent structure would increase the possibilities for misinterpretation; in the absence of structural constraints, for example, the grass might be taken to be the agent of swallow. 1. As the sun rose, the bird in the cool wet grass swallowed the worm quickly and went away. Since TI was not significantly affected by the padding manipulation, his problem is unlikely to lie in the recovery of structural information. His poor performance on the moved-arguments condition suggests, rather, that his interpretive errors arise in mapping from syntactic structures to lexical-semantic representations. In this respect, his comprehension deficit appears similar to that of the agrammatic aphasics who were the primary focus of the Schwartz et al. (1987) study.

16.4.3. Comprehension versus repetition The data from the semantic anomaly study suggest that TI's problem arises in the use of structural information to assign thematic roles. Little is known about this aspect of sentence processing, and it is not clear how an STM deficit would impinge on these operations. It can be argued, nevertheless, that if the STM limitation is the source of the mapping problem, sentences that do not exceed the patient's limited span should be adequately comprehended. Accordingly, in this study we examine repetition as well as comprehension of sentences systematically varied for length as well as transparency, as defined earlier.

Methods The materials utilized in this study were those of a sentence-picture matching task (STM Comprehension Test) that is part of our laboratory's comprehension battery. In this task, selection of the matching picture (out of four alternatives) requires correct assignment of a modifier as well as thematic relations. The modifier relation is expressed

438

Saffran and N. Martin

Table 16.4. Examples of STM comprehension sentences Sentence type

Example

Simple active declarative (SAD) SAD + initial phrase

The squirrel watched the brown bird. In the forest, the squirrel watched the brown bird. The squirrel watched the brown bird in the forest. The brown bird was watched. The brown bird was watched by the squirrel The bird that the squirrel watched was brown.

SAD + terminal phrase Truncated passive Full passive Object relative

either in the form of a prenominal adjective (the brown squirrel) or as a predicate adjective in the matrix clause of an object relative construction (The squirrel that the bird watched was brown). Sentence length is manipulated by adding an adverbial phrase, which is irrelevant to the picture match, either to the beginning or end of an active declarative sentence. In addition to the three sets of active sentences, the materials include full and truncated passives and object relatives, which are "nontransparent" in the sense defined earlier. Examples are given in Table 16.4. There were 12 sentences per condition and four picture choices, representing all possible modifier and thematic role combinations, for each sentence. In the comprehension task, the sentences were read to TI twice by one of the experimenters; in the repetition task, administered in a later session, he heard the sentence only once. Comprehension results for TI and for a group of right hemisphere lesioned subjects (JV = 6) are presented in Figure 16.2. In general, TI performed rather poorly in contrast to right hemisphere patients, who averaged one error or less on all but the object relatives. The data indicate that TI had more difficulty with active sentences in the terminal phrase condition, which suggests that processing of the critical thematic-rolebearing information may be disrupted by the need to encode additional material. Interpretation of these data in terms of the STM deficit is complicated, however, by the results of the repetition task, summarized in Table 16.5. It is evident from these data that there is often a marked discrepancy between TI's ability to repeat these sentences and his ability to comprehend them. Although thematic role information was preserved in the vast majority of his repetition responses, thematic role assignment in the comprehension task was essentially at chance. The difference was particularly dramatic in the case of the truncated passives. These were the shortest sentences in the set and he repeated 10/11 verbatim; yet he made correct thematic role assignments on only .33% of the comprehension trials. Similarly, although he retained the gist of all of the full passives on the repetition test, he was, again, only

O

cc

.§>

Q

Q

61 3!

O -n

I

U S

2 S

^

u J2

l! •c <

Si — .2 D

440

Saffran and N. Martin

Table 16.5. Repetition of sentences from the STM

Syntactic structure Simple active declarative (SAD) SAD + initial phrase SAD + terminal phrase Truncated passive Full passive Object relative Total

Comprehension Test

Proportion of open class words repeated in serial order

Proportion of sentences in which thematic roles retained

.92(w = 12)

.94(w = 48)

1.00

.16(n = 12)

.65(n = 60)

.75

.50(n = 12) .91(n=ll)a .58>{n = 12) 0(w=I2) .51

.55{n = 60) .97(n = 33) .94(w = 48) A0(n = 48)

.75 1.00 1.00 .67

Proportion of sentences repeated verbatim

.71

.86

"One sentence was inadvertently omitted.

33% correct on comprehension trials in this condition. Collapsing across sentence types, there was a significant difference between TFs retention of thematic role information in the repetition test and his ability to assign thematic roles in sentence-picture matching (McNemar Test: x2 Ul = 12.9, p < .005). We also looked at the extent to which the surface structure of the target sentence was retained in TI's repetition responses. Although there were various word substitutions and omissions, the basic structure of the original sentence was preserved in all conditions except the object relative, where the surface structure was fully retained in only 3/12 cases. In three instances, the embedded clause was transformed into a truncated passive, as in (2): 2. Target: The robber that the policeman shot was fat. Response: The robber that was shot was fat.

In three instances, the structure was changed to a subject relative, and in two others the response was an anomalous structure that retained the relative pronoun. There was only a single response in which no trace of a relative clause structure was evident. However, as the first several words were generally repeated verbatim, there were few structural options that would allow the sentence to be completed grammatically. These data are notable for the lack of relationship between TI's ability to repeat these relatively short sentences and his ability to interpret them correctly. Any hypothesis that links the comprehension problem to an inability to maintain sentential input in veridical form would have to predict more parallelism in these two sets of data. There are, of course, ways to account for the differences. It could plausibly be argued, for example, that sentence repetition performance does not provide a true estimate of

STM impairment and sentence processing

441

retention capacity under the more taxing conditions of the comprehension task, where the subject has to scrutinize the picture choices as well as process the sentence (although it should be noted that in this study, the sentences were presented twice in the comprehension trials and only once in the repetition test). It is also the case that the ability to repeat a sentence verbatim cannot be equated with retention of a veridical representation of the input string. The repetition response may be generated on the basis of information represented at several different levels, of which a phonological record - not necessarily complete - is only one. An account of the comprehension deficit in terms of loss of veridical information is, nevertheless, challenged by evidence that TI can repeat sentences that he consistently fails to understand. Further difficulties for this hypothesis are raised by the data from the studies of grammaticality judgment performance that follow. 16.5. Grammaticality judgment The most explicit account of the relationship between the STM deficit and the comprehension impairment thus far provided is the "narrow window" hypothesis of Caramazza et al. (1981) - that the STM limitation restricts the informational "window" over which syntactic structures are computed. TI's insensitivity to the padding manipulation in the semantic anomaly study provides little encouragement for this view. A more direct test of the parsing failure hypothesis was carried out by means of the grammaticality judgment paradigm previously used to demonstrate preserved syntactic abilities in agrammatic aphasics (Linebarger, Schwartz, and Saffran, 1983). To conduct a stringent test of the requirement for span-type capacities in parsing, we examined the effect of interpolating additional lexical material between the elements that carry the grammatical violation.

16.5.1. Methods TI was first given the battery of grammaticality judgment tests described in Linebarger et al. (1983). This task, which is composed of 451 sentences ( of which 221 are illformed), samples 10 different types of structural violations. These sentences are fairly short, typically seven or eight words in length; the distance between the elements that carry the violation seldom exceeds three items. The ill-formed string in (3) is a typical item from this test. 3. *How many do they have friends in Philadelphia? To examine the effect of distance between the critical elements (in italics), a second set of sentences (memory-stressed grammaticality judgments) was constructed by adding material to a subset of sentences from the Linebarger et al. set. For example, (4) was derived from (3).3

442

Safifran and N. Martin

Table 16.6. Performance on the memory-stressed grammaticality judgment task (proportion of correct judgments as a function of sentence condition and violation type)

Condition

Subject-aux Inversion

Strict subcategorization

d' Intervening material Across-theboard padding Original materials from Linebarger et al. (1983)

Incomplete extractions

d'

d!

.90

2.56

.80

2.58

.85

2.85

1.00

4.66

.95

3.60

.90

3.17

.95

3.60

.93

2.41

.95

3.28

4. *How many do you think that she and her husband actually have friends in Philadelphia? The memory-stressed sentences included 3 of the 10 violation types from the Linebarger et al. (1983) materials: incomplete extraction, as in (4), above; strict subcategorization, as in (5); and subject-aux inversion, as in (6): 5. *My brother George let his best friend to drive the new car all the way home. 6. *Has the little girl with brown curly hair has eaten her breakfast yet? Two types of lexical padding were contrasted in the memory-stressed test. In one condition, intervening material was inserted between the critical elements, as in (4) and (6); in the other, lexical material was added across the board, as in (5). There were 120 sentences in all, half of them ill-formed. Presentation conditions were identical to those employed in the original grammaticality judgment test of Linebarger et al. (1983).

16.5.2. Results and discussion The results of this study are summarized in Table 16.6. TI was 0.93 correct, overall, on the original grammaticality judgment task (N = 451), and 0.94 correct on the subset of sentences used to construct the padded materials (N = 120). His performance on the memory-stressed sentences (12 errors in all) was within the range of right hemisphere controls ( N = 4 ; mean number of errors = 13.8, range = 3—20) whose digit spans exceeded his by 2—5 items. Across-the-board padding had no effect on his performance, relative to the original sentences; the effect of intervening material was small and nonsignificant (McNemar Test: #2[1] = 1.78, p >.10). These results have two important implications. TI's good performance on the grammaticality judgment task indicates, first, that his comprehension deficit is unlikely to reflect a breakdown of parsing operations, and, second, that a phonological record is not required to compute long-distance structural dependencies. If short-term storage in

STM impairment and sentence processing

443

the form of a phonological record is, indeed, necessary in processing sentence materials, it must be in some process other than parsing. We will return to this point in the General Discussion.

16.6. General discussion To summarize, although TI shows the type of impairment in sentence comprehension that is typical of patients with STM deficits, his span limitation does not offer any clear explanation of his performance pattern over various types of sentence materials and sentence-processing tasks. Thus, length per se is not an adequate predictor of his comprehension performance, nor is verbatim retention, as indicated by his performance on a sentence repetition task. TI's ability to detect grammatical violations, even where the distance between the element critical to the well-formedness judgment considerably exceeded his span, indicates that the comprehension deficit is not due to failure in parsing; a further implication of his good performance on this task is that parsing is not crucially dependent on the phonological capacities implicated in normal span performance. These data argue against the "narrow window" hypothesis of Caramazza et al. (1981), which explicitly links the STM deficit to the comprehension impairment. The data from TI force us to consider more carefully how — or even whether — shortterm storage capacity as assessed by span-type tasks relates to the process of sentence comprehension. That sentence processing entails temporary information storage is not in question. What is at issue is whether there is some commonality between the storage capacities essential to processing sentence materials and those underlying performance on standard STM tasks that employ list materials. The span task, which calls for verbatim retention of a string of unrelated items, appears to rely heavily on capacities for maintaining information in phonological form. Although it has been suggested that sentential input is held in phonological form until structural constituents are computed (e.g., Jarvella, 1971; Clark & Clark, 1977), the function of the phonological buffer in this process has not been made explicit. The assumption of most parsing models that structure-building operations have access to a linear array (or "window," "lookahead") of lexical items does seem to imply the need for spanlike capacity, and it is indeed tempting to identify this storage function with the phonological buffer that supports list recall. It must be remembered, however, that the parser does not operate on phonological information per se; what it requires, rather, is knowledge of the form class (verb, noun, etc.) of the elements in the input string and their subcategorization features (e.g., whether a verb is transitive or intransitive), and the order in which these elements occur. Assuming the need for "lookahead," it is necessary to have some means of representing lexical-syntactic information in an array that corresponds to input order. One way to maintain the input

444

Saffran and N. Martin

sequence would be to index the syntactic descriptions to their corresponding phonological representations, in the manner that color and other separately extracted features of visual input are thought to be indexed to retinal location in vision (e.g., Marr, 1982). Indexation to a phonological record that directly represents the order of the items in the input string could stabilize the input to the parser, protecting it against perturbations in serial order that might arise from variations in the time course of lexical retrieval operations. Assuming this characterization of the role of a phonological buffer to be correct, how would reduction or elimination of the buffer store affect sentence processing? With respect to the recovery of structural information, this depends on the size of the "window" required for parsing. As noted earlier, Frazier and Fodor's (1978) parser has access to an array of five or six items; other parsing models, such as those of Marcus (1980) and Berwick and Weinberg (1984), assume a "lookahead" limited to two or three items that operates in conjunction with a "pushdown" stack of partially assembled structures. On such a model, the limited phonological capacities of the STM patient might well suffice for parsing, except perhaps in circumstances where identification of a lexical item is delayed or amended. TI's ability to perform grammaticality judgments is therefore compatible with parsing models that have limited lookahead requirements. It may be, however, that other levels of sentence processing require a more extensive phonological record. The data indicate that the difficulties of the STM patient arise in the mapping between structural information and thematic roles. Is a verbatim record required in the mapping operation? Thematic role assignment entails the integration of lexical representations with syntactic structures. On the basis of structural assignments made by the parser, noun phrases must be brought into alignment with verb argument structures; this configuration serves, in turn, as the basis for semantic interpretation (e.g., who/what is V-ed by whom/what). Very little is known about the nature of representations that are constructed at these "deep" levels of sentence processing. Among the uncertainties is the manner in which both syntactic and lexical information are articulated: Are lexical items represented within syntactic structures, as in the tree structures drawn by linguists, where the terminal nodes of the tree are labeled with specific lexical items? Or are structural and lexical information represented separately, as, for example, in tree structures where the terminal nodes only specify categorical information (e.g., NP, V)? (The latter possibility seems quite a reasonable one, if linguistic representations are as modular and distributed as visual representations seem to be.) If this were the case, there would need to be some way to link lexical items to constituent structures. This could be done along the lines suggested earlier, by indexing syntactic information (category nodes) to a phonological representation of the input string. According to this notion, the phonological record serves as the "glue" that binds together the various representations constructed in processing sentence materials.

STM impairment and sentence processing

445

Assuming this notion to be correct, what would be the consequences of an ability to maintain a phonological record? The impact on sentence comprehension should depend on the ease with which successive lexical items can be integrated into a semantic representation. Difficulty should not be encountered in cases where lexical input can immediately be assimilated into a semantic structure. Problems would be expected to arise, however, in instances where correct interpretation hinges on information that comes later in the sentence, and when semantic compiling operations are overloaded, as might perhaps occur with multiply conjoined NPs or concatenation of modifiers. Repetition data reported by Butterworth, Shallice, and Watson (this volume, chapter 8) suggest, for example, that multiply conjoined NPs (e.g., The removal firm took a bed, a cabinet, a wardrobe, and a chair) may tax even the normal listener's capacity to bind lexical input into a cohesive semantic representation. This raises the possibility that the key to the STM patient's performance in sentence comprehension tasks lies in what we might call "intepretive exigency." At risk of losing sentential input altogether unless it is rapidly encoded into semantic form, the patient may opt for "quick-and-dirty" semantic encoding. This would involve reliance on shortcuts such as heuristics for thematic role assignment - taking the first plausible candidate to be the agent, for example. Habitual use of such strategies could result in failure to utilize structural information effectively when it is available, which might explain TI's poor performance on sentences he was able to repeat correctly. We suggest, in other words, that TI is in full possession of the information needed to interpret structures like the truncated passives he was able to repeat verbatim but not to understand. His failures reflect overreliance on heuristics, applied indiscriminately in circumstances where he should be able to carry out a structurally based interpretation. According to this account, the relationship between STM impairment and comprehension performance is causal but indirect: The lack of a phonological record leads to adoption of a set of strategies that allow the patient to arrive quickly at a semantic interpretation, but the comprehension pattern that ensues does not reflect the storage limitation directly. Thus, parameters that bear a direct relationship to storage capacity, such as length factors, do not necessarily impinge on patients' comprehension performance. This account therefore accommodates some problematic aspects of the patient data. By invoking strategy differences across patients, it may also be possible to account for variations in performance patterns among the STM patients studied thus far (see Saffran, in press, for a summary of these data), as well as the apparent lack of comprehension disorder in a developmental STM case investigated by Butterworth, Campbell, and Howard (1986). The objection to arguments of this nature is that they are not easily disproved. This account does, however, carry the testable prediction that the comprehension deficits of STM patients should be amenable to strategic modification. As explanations go, this account lacks the virtue of a tight linkage between the

446

Saffran and N. Martin

storage deficit and the syntactic processing limitation, as in the "narrow window" hypothesis of Caramazza et al. (1981). The data do not, however, offer support for a direct relationship of this kind. As things now stand, it would seem that the only alternative to the looser account we have proposed here is to relinquish the notion that the STM deficit and comprehension problem are causally related. This suggestion has, in fact, been made (Caplan, Vanier, & Baker, 1986). Our view is that the possibility of such a relationship should not yet be abandoned. Even if this notion ultimately proves to be incorrect, pursuing it further will have the benefit of forcing us to make our conceptions of sentence processing and of short-term memory capacity more explicit.

Notes 1. In particular, PV, the patient studied by Vallar and Baddeley (1984), was substantially less impaired than the other cases reported. She performed well on semantically reversible sentences, and had difficulty only with long sentences presented for anomaly detection. 2. Data for TI and right hemisphere controls were previously reported in Schwartz et al. (1987). 3. We thank Marcia Linebarger for providing these materials.

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. 313—325). Hillsdale, NJ: Erlbaum: Basso, A., Spinnler, H., Vallar, G., & Zanobio, M. (1982). Left hemisphere damage and selective impairment of auditory—verbal short-term memory. Neuropsychologia, 20, 263—274. Berwick, R., & Weinberg, A. (1984). The grammatical basis of linguistic performance: Language use and acquisition. Cambridge, MA: MIT Press. 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, N., & 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., & Zurif, E. (1976). Dissociations of algorithmic and heuristic processes in language comprehension: Evidence from aphasia. Brain and Language, 3, 572-582. Clark, H. H., & Clark, E. V. (1977). Psychology and language. New York: Harcourt Brace Jovanovich. Frazier, L, & Fodor, J. D. (1978). The sausage machine: A new two-stage parsing model. Cognition, 6, 291-325. Friedrich, F., Glenn, C, & Marin, O. (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 a phonological coding deficit on sentence processing. Cognitive Neuropsychology, 2, 385-412. Jarvella, R. J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and Verbal Behavior, 10, 409-416. Linebarger, M. C , Schwartz, M. F., & Saffran, E. M. (1983). Sensitivity to grammatical structure in so-called "agrammatic" aphasics. Cognition, 13, 361-392.

STM

impairment and sentence processing

447

Marcus, M. (1980). A theory of syntactic recognition for written language. Cambridge, MA: MIT Press. Marr, D. (1982). Vision. San Francisco: Freeman. 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. Saffran, E. M., & Martin, 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. Saffran, E. M., Schwartz, M. F., Linebarger, M., Martin, N., & Bochetto, P. The Philadelphia Comprehension Battery. Unpublished test. Savin, H., & Perchonock, E. (1965). Grammatical structure and immediate recall of sentences. Journal of Verbal Learning and Verbal Behavior, 9, 348-353. Schwartz, M. F., Linebarger, M. C, & Saffran, E. M. (1985). The status of the syntactic theory of agrammatism. In M. L. Kean (Ed.) Agrammatism (pp. 83—124). New York: Academic Press. Schwartz, M. F., Linebarger, M. C, Saffran, E. M., & Pate, D. S. (1987). Syntactic transparancy and sentence interpretation in aphasia. Language and Cognitive Processes, 2, 85—113. Vallar, G., & Baddeley, A. D. (1984). Phonological short-term store. Phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121-141. Von Eckhardt, B., & Potter, M. C. (1985). Clauses and the semantic representation of words. Memory and Cognition, 13, 371—376. Wanner, E., & Maratsos, M. (1978). An ATN approach to comprehension. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality, (pp. 119-161). Cambridge, MA: MIT Press. 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.

17. Phonological processing and sentence comprehension: a neuropsychological case study GIUSEPPE VALLAR, ANNA BASSO, AND GABRIELLA BOTTINI

17.1. Introduction Patients suffering from an acquired deficit of verbal short-term memory (phonological auditory-verbal short-term input store, PSTS: see Warrington & Shallice, 1969; Vallar & Baddeley, 1984a; Shallice & Vallar, this volume, chapter 1) typically show a cooccurring impairment of sentence comprehension (see Vallar & Baddeley, 1984b). This association had led a number of investigators to suggest that the PSTS contributes to certain aspects of sentence comprehension (see, e.g., Saffran & Marin, 1975; Shallice, 1979; Caramazza, Basili, Koller, & Berndt, 1981; Vallar & Baddeley, 1984b, for further details). If one takes a closer look at the empirical data base from which these conclusions are drawn, it is, however, apparent that the observed patterns of comprehension impairment are, at least prima facie, rather heterogeneous. Most reported cases show a defective performance on the Token Test (e.g., Warrington, Logue, & Pratt, 1971; Caramazza et al., 1981; Basso, Spinnler, Vallar, & Zanobio, 1982; McCarthy & Warrington, 1987a). When more specific aspects of this sentence comprehension deficit, putatively produced by the phonological memory disorder, are considered, a comparative evaluation is more difficult, since variable methods and materials have been used in different patients. In a sentence repetition task in which the ability to paraphrase is regarded as an index of comprehension, case IL is, in most instances, able to offer an "acceptable" paraphrase. However, in the presence of an introductory clause passive constructions are frequently transformed into active, and when the sentence is semantically reversible, the meaning is also reversed (Saffran & Marin, 1975). Case MC (Caramazza et al., 1981) shows defective comprehension of semantically reversible sentences in a sentence—picture matching task with both auditory and visual input, while his performance is comparatively preserved in the case of nonreversible items. In

We wish to thank David Howard and Tim Shallice and, most of all, Eleanor Saffran, for their suggestions and comments on an earlier version of this paper. Luigi Pizzamiglio kindly provided the syntactic comprehension test.

448

Phonological processing and sentence comprehension

449

PV's case, comprehension of semantically reversible sentences, as assessed by both two- and four-choice sentence-picture matching tasks, is preserved. With both auditory and visual presentation, however, she shows impaired performance in a verification task in which fairly long and complex sentences are made semantically anomalous by the reversal of two relevant items. Conversely, her performance is within the normal range in the case of sentences made anomalous by a semantic mismatch between two lexical items (Vallar & Baddeley, 1984b, 1987). Case EA's comprehension of both auditory and visual sentences is selectively impaired in a sentence-picture matching task when reversed role distractors are used, whereas performance with lexical distractors is errorless (Friedrich, Martin, & Kemper, 1985). EA's immediate repetition parallels her comprehension impairment, since she is comparatively more defective in the case of semantically reversible sentences compared with irreversible items. TI (see Saffran, 1985; Saffran & Martin, this volume, chapter 16; Schwartz, Linebarger, Saffran, & Pate, 1987) shows defective comprehension of reversible sentences, as assessed by a sentence-picture matching task. In an auditory sentence verification task TFs comprehension deteriorates when the semantic anomaly is conveyed by syntactic structure (i.e., the reversal of two items), but not by lexical items alone. Finally, cases RAN and NHA (McCarthy & Warrington, 1987a) are both grossly impaired on the Token Test, a task that includes items where word order is relevant, but their comprehension of reversible sentences, as assessed by a sentence-picture matching task is, respectively, marginally defective and preserved (McCarthy & Warrington, 1987b, this volume, chapter 7). Taken together, the performance of the patients in verification and matching tasks gives rise to an "asyntactic" pattern of comprehension, where performance deteriorates when the syntactic structure of the sentence conveys crucial information, and preserved lexical—semantic processing of the major lexical items does not guarantee adequate understanding. It should be noted, however, that dissociations between tasks have been found. For instance, MC and TI, but not PV and NHA, show a defective performance in sentence-picture matching. Probably due at least in part to the heterogeneity of the empirical data (see, for instance, Caplan, Vanier & Baker, 1986, and Butterworth, Campbell, & Howard, 1986), the precise relationship between syntactic processing and PSTS has been a matter of considerable controversy and disagreement. The view, which may be traced back to Clark and Clark (1977), that there is an intimate link is suggested by the idea that the PSTS is the working space of the syntactic parser (Caramazza & Berndt, 1985). Alternatively, the PSTS may not be an intrinsic subcomponent of the parsing device, but a comparatively independent short-term retention system, bearing a more or less distant relationship with the parser and which becomes, in specific instances, involved in the comprehension process. One of the first suggestions was that the PSTS may be a backup store, which allows repeated non-real-time attempts — by syntactic and semantic processes - to understand "complex" sentences (Saffran &

450

Vallar, Basso, and Bottini

Marin, 1975; Shallice, 1979). Vallar and Baddeley (1984b) suggested a relevant role of the PSTS in a specific condition, when word order, typically preserved by a phonological record, determines the meaning of the sentence. Caplan and co-workers (Caplan, Vanier, & Baker, 1986; Caplan & Waters, this volume, chapter 14) locate the possible contribution of the PSTS-rehearsal system in a postinterpretive stage, where it would allow adjudication between conflicting interpretations of a given sentence. Finally, according to an extreme view expressed recently by Butterworth et al. (1986), the PSTS does not play any role in comprehension of sentential material (but see Vallar & Baddeley, 1987). In the present set of experiments we have investigated immediate memory for lists of unrelated items and sentences, and sentence comprehension. The subject under study was a left brain-damaged patient, ER, who had a selective deficit of auditory-verbal memory span. In the first section of the study we have attempted to specify the functional locus (or loci) of impairment, using a model of phonological shortterm memory (Vallar & Cappa, 1987; see a flow chart in Shallice & Vallar, this volume, chapter 1) that distinguishes (a) phonological analysis (b) phonological short-term storage, (c) articulatory rehearsal and (d) phonological recoding subcomponents. In this model auditorily presented verbal items, after phonological analysis, have a direct access to the PSTS, whereas in the case of visual input two successive stages are needed: phonological recoding, which provides phonological conversion of visual items, and articulatory rehearsal, which feeds the output of phonological recoding to the PSTS and recirculates information held in this latter system. In the second part of the study we have assessed sentence processing by both comprehension and repetition tasks, using material that has been shown to pose problems to patients with a putative defect of the PSTS. In both components of the study the contribution of nonphonological factors has also been assessed.

17.2. Case study 17.2.1. Case ER ER is a right-handed woman (born 1950), who has suffered from an aortic valvular stenosis since infancy. On October 6, 1984, she was admitted to the hospital after the sudden onset of aphasia and loss of strength in her right limb. A CT scan (January 1, 1985) revealed an ischaemic temporoparietal lesion; the insular region was also involved (see Figure 17.1). Given a standard language examination (Basso, Capitani, & Vignolo, 1979) ER exhibited a fluent speech with phonemic paraphasias. Oral comprehension of individual words was assessed by a word-picture matching task, including two subtests in which ER was requested to point to the word pronounced by the examiner on a 20-item

Phonological processing and sentence comprehension

451

Figure 17.1. CT scan of patient ER.

multiple-choice display. The first subtest comprised 20 unrelated living and nonliving items (60% high-to-middle frequency, 40% low frequency, i.e., lower than 5 per 500,000; see Bortolini, Tagliavini, & Zampolli,1972); in the second, the 20 items (25% high-to-middle frequency, 75% low frequency) belonged to five semantic categories (animals, flowers, containers, cutting tools, vehicles). ER scored 95% and 100% correct answers, respectively, in the first and in the second subtest. Comprehension of auditorily presented individual words was further assessed by a two-choice word-picture matching task, where the distractor was phonemically similar, semantically related or unrelated to the target item. The test included 90 items (30 for each distractor condition: 50% high-to-middle frequency, 50% low frequency) presented in a random order to ER, who scored 89/90 correct answers, making one error in the phonemic distractor condition. In the case of sentence comprehension, ER showed a defective performance on the Token Test (22/36, cutoff 29; DeRenzi & Faglioni, 1978). However, in a verification task in which anomalous sentences were produced by a semantic mismatch between two lexical items (e.g., A cat has a tail vs. A car is a flower), she had an errorless performance (40/40) with both auditory and visual presentation. Repetition of letters, syllables, words, nonwords, and sentences was disproportion-

452

Vallar, Basso, and Bottini

ately defective. On the sentence repetition test of the Italian version of the Multilingual Aphasia Examination (Benton & Hamsher, 1978), she was unable to repeat items longer than four words and had a repetition score of 3 (average score of 49 normal subjects with 3-5 years of schooling = 17.72, SD = 2.34; Previdi, 1975). Reading aloud and comprehension of written material were preserved. On the basis of this pattern of impairment she was classified as a conduction aphasic. Speech production was assessed in greater detail at the level of individual words and sentences by word fluency, sentence rearrangement, and production tasks. In the phonemic fluency task, ER was required to pronounce as many words as she could, beginning with the letters F, P, and L The semantic fluency task differed from the previous one only in that three categories were used (makes of cars, fruits, animals). ER's adjusted scores were 27 and 29, above the cutoffs of 17 and 25, respectively, for the phonemic and semantic task (Novelli et al, 1986a). ER was also assigned a sentence anagram task. Each word of a given sentence was written on an individual card. ER was instructed to arrange the cards, presented at random in front of her, into a sentence. The test comprised active and passive direct object sentences; active indirect object; relative, imperative, and interrogative sentences; and active sentences with locative prepositions. The patient had an errorless performance on this task, giving 28 out of 28 correct answers. In an Italian version of the Story Completion Test (Goodglass, Gleason, Bernholtz, & Hyde, 1972), ER scored 24 out of 28 correct answers (11 control subjects: 22.18, range = 19-26; Vallar, Papagno, & Cappa, 1988). The patient was not apraxic. Her score on a general intelligence test such as the Raven P.M. 47 was 24 (cutoff 18). Visuospatial memory (DeRenzi, 1977) was unimpaired. ER had a span of four on Corsi's Block Tapping Test and took 12 trials to learn a supraspan sequence (controls: 11.94, SD = 7.75). She took 7 trials and made 18 errors to learn the route of a visual maze (controls: trials, 13.41; errors, 37.5). Verbal learning of meaningful material (Novelli et al., 1986b) was preserved. ER scored 13 and 17, respectively (cutoff 8 in both tasks), in committing a short story and three lists of words to memory.

17.2.2. Phonological processing ER was given a consonant discrimination task posing a minimal short-term memory load. She was asked to judge whether two stop consonant-vowel (CV) syllables (/pa/, Ibal, Ital, leal, /da/, and Igal), presented one immediately after the other, were identical (e.g., ha-ha) or different (e.g., ha-pa) in sound. The test includes discrimination of phonemes contrasting in voicing only, in place only, and in both distinctive features (see Vallar & Cappa, 1987, for details). The CV pairs were recorded on tape by one of the authors (G.B.) and transmitted to ER through headphones. The patient scored 49/60, and five control subjects had an errorless performance. Pisoni (1973) has suggested a distinction between auditory and phonetic codes

Phonological processing and sentence comprehension

453

involved in speech perception, where auditory components are comparatively more available for vowel than for consonant processing. ER was then tested by a vowel discrimination task, which differed from the previous one in that the initial consonant Ibl was held constant, while the vowel varied (/a/, lei, HI, lol, lul). The task comprised 20 identical (e.g., ba-ba) and 20 different (e.g., ba-be) pairs recorded on tape in a random order. ER had a remarkably good performance in this task, scoring 39/40 correct answers. These findings, similar to the recent observations of Friedrich, Glenn, and Marin (1984) in case EA, provide some indication that ER's deficit may be confined to phonological analysis and does not extend to auditory processing of speech input. ER's substantial preservation of comprehension of auditorily presented individual words is in line with data from group studies showing frequent instances of phonological processing difficulties associated with a preserved comprehension at the single-word level (Miceli, Gainotti, Caltagirone, & Masullo, 1980) and poor correlations between phonological deficits and auditory comprehension (Blumstein, Baker, & Goodglass, 1977). Similarly, EA's comprehension of individual words is reported to be largely preserved (Martin, 1987).

17.2.3. Phonological short-term memory Auditory and visual digit span Lists of auditory and visual digits were presented at the rate of one item per second. Twenty strings for each list length were given, and no digit was repeated within a string. ER was instructed to recall the list serially at a nonverbal signal by the examiner (a tap on the table), given immediately after presentation. As shown in Table 17.1, ER's span is clearly defective, with a better level of performance in the case of visual input (for three-item lists: #2[1] = 16.42, p < .001). For three-digit auditory lists, error analysis revealed omissions {78%), digit substitutions (17%), and order errors (5%) (i.e., a correct item not recalled in the appropriate serial position).

Phonological similarity effect

Strings of phonologically similar (D, T, G, P, B, V, C) and dissimilar (F, K, W, X, R, Z, Q) letters were used. The procedure described for digit span was followed. As shown in Table 17.2, ER showed the standard effect of phonological similarity with auditory (two- and three-item lists: %2[1] = 9.74, p < .01) but not visual (three- and four-item lists: / 2 [1] = 1.03, n.s.) input. Again, her performance level was superior in the case of visual presentation. For three-item auditory phonologically similar lists, error analysis revealed letter substitutions (50%), order (32.5%), and omission (17.5%) errors; for

454

Vallar, Basso, and Bottini

Table 17.1. Auditory and visual digit spans: percentage of items correct in the correct serial position; sequences correct in parentheses Auditory

Visual

100% 100% 60% (45%)

100% 100% 92% (85%)

String length

1 2 3

Table 17.2. Serial recall of phonologically similar (PS) and dissimilar (PD) auditorily and visually presented letters: percentage of items correct in the correct serial position; sequences correct in parentheses

PS String length 1 2 3 4

Auditory PD

65% 60% (30%) 32% (0%) —

100% 8>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

Phonological processing and sentence comprehension

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

Phonological processing and sentence comprehension

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

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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

Vallar, Basso, and Bottini

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

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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,

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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,

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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.

Phonological processing and sentence comprehension

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

Vallar, Basso, and Bottini

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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.

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

(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.

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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

Working memory and reading disorder

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

Crain, Shankweiler, Macaruso, and Bar-Shalom

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, /?<.O3), as shown in Figure 18.3, reveals this discrepancy. As

496

Crain, Shankweiler, Macaruso, and Bar-Shalom Good Readers Poor Readers 40 -

30 -

10 -

-D

Complex NPs

Simple NPs

Figure 18.2. Percentage of incorrect responses to simple and complex NP sentences (Experiment 2).

50-i

Good Readers Poor Readers

30-

c o

20

10"

D-

After

-D

Before

Figure 18.3. Percentage of incorrect responses when conceptual order conflicts with order of mention in the no felicity condition (Experiment 2).

predicted, the worst case for poor readers was the after sentences that pose a conflict between order of mention and conceptual order and no contextual support in the form of satisfied presuppositions. Good readers, on the other hand, were not handicapped by the memory demands this situation places on the subject. A glance at the type of errors made by poor readers in the most difficult situation reveals almost as many responses in which they identified the wrong object (19%) as responses in which they acted out the clauses in the wrong order (25%). In a few instances both error types occurred in processing the same sentence. However, in the

Working memory and reading disorder

497

corresponding situation in Experiment 1 (i.e., in response to after sentences when order of mention conflicts with conceptual order), poor readers made relatively few wrong object responses (7%) compared to wrong order responses (14%). Most likely, this difference in the relative frequency of error types is due to the manipulation of NP complexity across the experiments. Poor readers were more likely to choose a wrong object in sentences containing complex NPs in Experiment 2, since choosing an object denoted by a complex NP requires overcoming the additional problems of compiling discontinuity. Taken together, the findings of Experiments 1 and 2 indicate that as processing demands are increased, poor readers' performance on an object manipulation task involving temporal terms sentences is eroded much more than good readers' performance. Decreasing the processing demands, either by satisfying the felicity conditions or by using less complex NPs, elevates performance by poor readers, such that group differences diminish and all subjects perform at a high level of success. Results showing improved performance by poor readers when processing demands are lowered cannot be explained on the hypothesis that poor readers are lacking the relevant syntactic structures. Only an account in terms of processing limitations can explain these findings. We hypothesize, based on the results of Experiments 1 and 2, that if we conducted an additional experiment that presented only simple NPs (as in Experiment 1), but with the felicity conditions satisfied, we would find near-perfect performance by both groups.

18.4.4. Garden path effects (Experiments 3 and 4) In Experiment 3 we examine the responses of good and poor readers to the kind of garden path sentences discussed in section 18.2. As we noted, an incorrect response to these sentences can be explained in two ways, corresponding to the two hypotheses we have been considering about the source of poor readers' problems in sentence comprehension. First, errors could reflect the absence in a subject's grammar of a structural constraint on extraction — Subjacency. Alternatively, they could be the result of an inability to overcome the effect of the parsing strategy Right Association, presumably due to limited working memory resources. Three types of garden path sentences were presented, in order to vary the processing load placed on working memory. As indicated earlier, varying sentence types puts us in a position to examine the error pattern across sentence types to help distinguish between the PLH and the SLH. The three types of syntactic constructions are illustrated in (14)-(16). There were four sentences of each type (adapted from Crain & Fodor, 1985). 14. Relative clause: Who is Bill pushing the cat that is singing tol 15. Prepositional phrase (deep): What is Jennifer drawing a picture of a boy with! 16. Prepositional phrase (distant): Who is Susan handing over the big heart-shaped card tol

498

Crain, Shankweiler, Macaruso, and Bar-Shalom

Sentence (15) is labeled "deep" because the origin of the "extracted" Wh- phrase is a prepositional phrase that is embedded in an NP that is itself embedded in an NP. This contrasts with the "distant" case (16), in which there is only one level of embedding. Although the sentences are matched for length, we anticipated that distant PP sentences would be easier to process than either the deep PP sentences or the relative clause sentences. The depth of syntactic embedding in both relative clause and deep PP sentences means that they deviate more than the distant PP sentences from the simple look-up-and-concatenation translation process presented in the preceding section. A set of 44 good readers and 46 poor readers (which includes all subjects of Experiments 1 and 2) participated in this study. The mean combined word and nonword reading scores on the DST for the good and poor readers were 96.3 and 37.5, respectively. On each trial subjects were asked to listen carefully to a tape-recorded set of sentences that described a scene depicted in a large cartoon drawing placed in front of them. Immediately following the description, they were asked to respond to a question about some aspect of the drawing. As an example, the context sentences in (17) preceded the test question (14). 17. Bill's father is waiting for Bill to bring him the cat. The cat loves to sing and has made up a song for his toy mouse.

The grammatically correct response to this question is his father. The response the mouse is incorrect, since it represents an apparent violation of Subjacency. The PLH would argue that an examination of the pattern of errors across sentence types for each group may provide evidence that these errors are not in fact violations of Subjacency at all, but are the result of the processing burdens these sentences impose on working memory. It is difficult to say exactly what the SLH would predict about the pattern of responses by good and poor readers for any of the sentence types presented in this experiment. It seems reasonable to suppose, however, that the SLH might anticipate that the reader groups would display different profiles, since these sentences are exceedingly complex. To reiterate, the PLH predicts that both groups will manifest a similar pattern of errors across sentences of varying syntactic types, with poor readers penalized to a greater degree than good readers on sentences that are costly of working memory resources (e.g., the PP deep and relative clause sentences.) This is exactly what was found. Analysis of the percentage of incorrect responses made by both groups reveals main effects of sentence type (F[2, 176] = 21.53, p < .001) and reader group (F[l, SS] = 8.95, p < .004), but no interaction of sentence type and reader group. Figure 18.4 shows that the effect of sentence type is due mainly to the higher percentage of errors for relative clause (30.5% errors) and PP deep (32.4% errors) sentences than for PP distant (15.8% errors) sentences, as anticipated by the model of working memory we presented. The reader group difference reflects the fact that good readers (21.4% errors) performed significantly better than poor readers (31.0% errors).

Working memory and reading disorder 50 i

499

Good Readers Poor Readers

40 -

20 -

a. 10 -

Relative

Clause

PP Deep

PP Distant

Figure 18.4. Percentage of incorrect responses to garden path sentences (Experiment 3).

The absence of interaction means that poor readers show a general decrement in performance, but responded in the same way as good readers to the three different types of garden path sentences. As may be seen in Figure 18.4, there are no constructions that provide disproportionately greater difficulty for poor readers. This invites the inference that errors for both reader groups should receive the same interpretation. Since good readers exhibit a high proportion of correct responses on these sentences, it is reasonable to conclude that their errors are ones of performance and do not reflect an underlying deficiency in syntactic knowledge. The significant reader group difference across sentence types might, at first glance, suggest that poor readers are lacking this knowledge, were it not for the absence of a group by sentence type interaction.13 To support our contention that poor readers do not suffer from a lack of grammatical competence, we conducted a further experiment to assess competence with the structure that mostly clearly differentiated the two reader groups, the relative clause (good readers: 23.9% errors; poor readers: 37.0% errors). If the PLH is correct, and poor readers are not missing some relative clause structures, then the differences between reader groups should evaporate when competence is assessed in a way that reduces the processing burdens of working memory. This would reinforce the inference that the reader group differences in evidence here should be attributed to a processing limitation on the part of the poor reader group and not a structural deficit.

18.4.5. Experiment 4 Following the proposal presented in the previous discussion, Experiment 4 uses the relative clause as a further testing ground between the hypotheses. To this end, we

500

Crain, Shankweiler, Macaruso, and Bar-Shalom

sought to establish that the reader group differences in response to relative clause sentences in Experiment 3 do not reflect lack of grammatical competence on the part of poor readers, by showing that poor readers are able successfully to comprehend relative clauses in contexts that ease memory load. Ideally, we would like to provide evidence that the apparent violations of Subjacency by poor readers disappear when the demands on memory are reduced. However, reducing processing load for sentences that could induce Subjacency violations is nearly impossible. Therefore, we chose an alternative approach: to assess the ability of poor readers to comprehend the most complex substructure contained in the sentences that evoked the greatest number of errors (i.e., the ones containing relative clauses). Thus, this experiment was designed to assess the competence of poor readers in interpreting relative clauses with memory demands held to a minimum. Positive results in this case would lend further support to the joint claim that poor readers possess the universal constraint of Subjacency but failed to perform as well as the good readers in Experiment 3 because of their deficient processing capabilities.14 The relative clause has been a focus of research in normal language acquisition as well as in the literature on reading disability. Relative clauses are found to evoke difficulties in interpretation for preschool children (Tavakolian, 1981), and for older children who are poor readers (Byrne, 1981; Stein et al., 1984). Early research in both areas led some researchers to the conclusion that children's poor performance was due to a lack of syntactic knowledge. However, Mann et al. (1984) tested good and poor readers' comprehension of relative clauses using an act-out task and found that, although good readers performed significantly better than poor readers overall, both groups were affected in the same way by the difficulty of the type of relative clause. This familiar error pattern of good and poor readers is further evidence that they differ in processing capabilities, rather than in structural competence, as we have seen. Further support for the view that poor readers' difficulties with relative clauses reflect performance factors comes from an additional study by Smith, Macaruso, Shankweiler, and Crain (in press) of good and poor readers' comprehension of relative clauses. Adapting several experimental innovations from the literature on language acquisition, Smith et al. found that poor readers made few errors when the pragmatic presuppositions on the use of relative clauses were satisfied (Hamburger & Crain, 1982). Smith et al. compared their findings with those of the Mann et al. study, in which the same subject selection criteria were used, but in which the presuppositions of relative clauses were unsatisfied. Taken together, the data from these studies revealed that the changes in methodology had the effect of eliminating reader group differences, with the result that both good and poor readers performed at a high level of success. The present study employed an act-out task that incorporated some of the methodological innovations used by Smith et al. and by Hamburger and Crain, in order to assess the grammatical competence of a subset of the good and poor readers who

Working memory and reading disorder 50 -i

501

Good Readers Poor Readers

40 -

30-

Q.

10"

SO

OO

OS

Figure 18.5. Percentage of incorrect responses to relative clause sentences (Experiment 4).

participated in Experiment 3. The same set of 14 good and 14 poor readers from Experiment 1 participated in the present experiment. Three types of relative clause constructions were used.15 Examples are provided in (18)—(20). 18. SO: The lion that the bear bit jumped over the fence. 19. OO: The boy touched the girl who the ice cream fell on. 20. OS: The lady hugged the man who picked up the suitcase.

In order to reduce the processing burden imposed by sentences such as (18)-(20), we satisfied one of the presuppositions of relative clauses. Specifically, we incorporated two objects in the experimental workspace corresponding to the head noun of the relative clause. For example, in (19), there were two figurines from which the subject could choose to act out the sentence. By including the extra figurine, we satisfied the requirement of a restrictive relative clause to restrict, in the example, the set of girls to the one on which the ice cream fell. As in the Smith et al. study, we found no significant group differences and a low error rate for all subjects (good readers: 8.7% errors; poor readers: 11.1% errors). In addition, the pattern of errors across sentence types was virtually identical for both groups. This is shown in Figure 18.5. The main finding was that poor readers acted out relative clause sentences with an 89% success rate when the appropriate presuppositions were met. In fact, for two of the sentence types, poor readers made only 7% errors. This provides support for the contention that a syntactic deficit was not the cause of the inferior performance of poor readers in Experiment 3. The results of Experiment 4 strongly suggest that poor readers are able to comprehend various relative clause constructions. These results invite the inference that the poor readers' higher error rate in the context

502

Crain, Shankweiler, Macaruso, and Bar-Shalom

provided in the earlier study by Mann et al. was a consequence of their abnormal limitations in working memory.

18.5. Conclusion The purpose of the present research was to develop an approach to language comprehension problems that is sufficiently detailed to allow specific predictions and powerful enough to identify causes. The first half of the chapter lays the groundwork. It presents the major assumptions about the language apparatus that underlie our research. Working within the framework of a modular view of language, we gave an account of verbal working memory that emphasizes its control functions more than its storage functions. We proposed that the task of the control component of working memory, in extracting meaning from linguistic forms, is to regulate the flow of linguistic material through the system of parsers, from lower to higher levels of representation. The inherent limitation in buffer capacity entails that higher-level processing must be executed within very short stretches of text or discourse. To deal with this problem, the control component of working memory must rapidly transfer partial results of linguistic analyses between levels of structural representation, that is, the phonology, the lexicon, the syntax, and the semantics. A further goal was to show how this model could be successfully applied to get to the root of difficulties that children who are poor readers often display in processing spoken sentences. Alternative hypotheses about the causes of poor readers' sentence comprehension problems were posed, and the conceptual machinery for testing between these hypotheses was introduced. The hypotheses make different predictions because they locate the source of reading difficulties in different components of the language apparatus. Roughly, the views turn on the distinction between structure and process. On the SLH, poor readers suffer from a structural deficit, (i.e., a deficit in the stored mental representation of principles of syntax), in addition to their (unrelated) deficiencies in phonological processing, verbal working memory, and so on. On the PLH, each of the deficits of the poor readers is a reflection of their limitations in processing phonological information. To test between these hypotheses, we reviewed the results of four interlocking experiments. A pattern of findings emerged that indicates that the necessary syntactic structures were in place, and that the poor readers' difficulties in comprehension of spoken sentences stemmed from inefficiencies in on-line processing of sentences that for one reason or another stressed working memory. Thus, inefficiency of verbal working memory and not failure to acquire critical language structures was identified as the factor responsible for the comprehension difficulties. We argued that, ultimately, the problems for poor readers originate at the phonological level, and that difficulties that

Working memory and reading disorder

503

might appear to reflect a syntactic deficiency are, in reality, manifestations of a special limitation in accessing and processing phonological structures.

Notes 1.

2.

3.

4.

5.

6.

For instance, poor readers have been found to be less accurate than their age-matched controls in understanding sentences with infinitival complements of object-control adjectives like easy (Byrne, 1981), which Chomsky (1969) found problematic for children as old as 9. Relative clauses also pose difficulties for poor readers in some contexts (Goldsmith, 1980; Byrne, 1981; Mann, Shankweiler, & Smith, 1984; Stein et al., 1984; but also see Smith, Macaruso, Shankweiler, & Crain, in press); the relative clause is a grammatical structure many researches believe to develop late (Brown, 1973; Sheldon, 1974; de Villiers, Tager-Flusberg, Hakuta & Cohen, 1979; Tavakolian 1981; but see Hamburger & Crain, 1982). Poor readers have also been shown to encounter difficulty with some dative constructions (Fletcher et al., 1981) that have also been found to evoke comprehension errors in young children (Scholes, 1978). Further discussion of the implications of this hypothesis concerning the causes of reading disorder is presented in Shankweiler & Crain (1986), Crain and Shankweiler (1988), Shankweiler et al. (in press). This distinguishes our conception of verbal working memory from proposals by Baddeley (1986) and Carpenter and Just (1988). These researchers see working memory as a generalpurpose device that plays a central role not only in language but also in reasoning, problem solving, and in other forms of complex thinking. As we conceive of it, a plan is a mental representation used to guide action. The formation of a cognitive plan is an important aspect of any comprehension task involving the manipulation of objects, since the complexity of a plan is a potential source of difficulty in sentence comprehension. Therefore, when children perform poorly in language comprehension tasks that involve planning, it is important to consider the possibility that formulating the relevant plan is the locus of the problem, rather than other purely linguistic aspects of the task, such as their imperfect knowledge of linguistic rules. Where our objective is the assessment of the limits of children's linguistic knowledge (i.e., competence), we must develop our understanding of the nature and relative complexity of plans (whereby that competence is demonstrated), and we must also devise experimental paradigms in which the impact of plan complexity on linguistic processing is minimized (see section 18.4). In addition to plan complexity, the compiling analogy allows us to entertain several other types of difficulty that a sentence can present to a subject whose task it is to plan and execute an appropriate response. Although the issue is too thorny to take up here, it may be useful to mention just one of many apparent counterexamples to the isomorphism between syntax and semantics. The example involves the phrase three consecutive rainy days. The preferred, but noncompositional, interpretation of this phrase means three consecutive days on which it rained. The compositional interpretation required a composite set of ordered rainy days, perhaps taken from a set of weeks, each containing a rainy day or two. So if it had rained on three consecutive Thursdays, Donald might have forgotten his umbrella on three consecutive rainy days. Since the compositional interpretation requires a more complex set of presuppositions, the default is to interpret this phrase in a way that violates compositionality. As we noted, it is reasonable to suppose that this could add to processing difficulty. There are other circumstances in which noncontiguous elements must be combined, in violation of Condition (b). This problem arises whenever there are discontinuous

504

7.

8.

9.

10.

11.

Crain, Shankweiler, Macaruso, and Bar-Shalom dependencies in the source language. In English, for example, this problem (of inverse compiling discontinuity) occurs in verb-particle constructions such as look up, and in cases of verb subcategorization (e.g., the verb put takes both an NP and a PP complement). Clearly, these properties of the linguistic input are seen to impose demands on memory, by postponing the closure of syntactic categories. It is worth noting that changes in word order could circumvent the sequencing problem, by permitting a more nearly optimal translation between source and target language. Consider a language in which normal modifiers appear after the noun, as in bear brown second. In this language, fragments of the target plan would be composed in the order in which they were mentioned in the phrase. However, it should be pointed out that in many situations, the construction of plans that satisfy Condition (c) may still require a commitment of memory resources. Specifically, the memory system must retain a data structure corresponding to the semantic value of each element of the phrase. For example, interpreting the noun bear as part of the phrase bear brown second requires the formation of a data structure delimiting the set of bears in the discourse setting. This data structure must somehow be stored, awaiting the formation of the semantic value for the remainder of the phrase. This burden on memory could be eliminated in certain contexts, such as situations in which the objects denoted by the noun are aligned in a way that allows one to restrict attention to that set alone. This would occur, for instance, if all of the bears were in the same field of*view. A simple shift of attention would allow the parser to restrict attention to the remainder of the phrase. At present, we cannot say whether unmet presuppositions thwart cognitive compiling and thereby result in premature execution or whether, instead, the satisfaction of presuppositions facilitates compile-mode behavior. In any event, it is clear that children's mistaken responses in situations that flout presuppositions exonerate the syntactic component and implicate limitations in working memory capacity. An example may be helpful in illustrating this point. It is proposed in Hamburger and Crain (1987) that certain principles of compositional semantics restrict the set of word order possibilities in natural language. In determining semantic constituency relations, the following principle is proposed: (P) An operator must be composed with appropriate operands. To apply this principle, let us consider two categories of modifiers, selectors and restrictors. Modifiers like striped are restrictors, since they perform the semantic operation of acting on a set to produce a (not necessarily proper) subset as output. A selector also acts on a set, but it returns an element as output. For example, in the phrase second striped ball the value of applying second to the set designated by striped ball is an individual ball. Principle (P) allows a restrictor to apply before a selector, but not after one, since the output produced by a selector is not appropriate input to a restrictor. This explains the illicit character of striped second ball, even though this alternative phrasal order might be preferred from the standpoint of the working memory system. It is important to note that even the test sentences used in Experiment 1 are not the least complex case of the relevant construction, since the noun phrases in both clauses contained an inessential prenominal modifier. We chose to introduce this slight additional complexity in order to avoid inducing ceiling performance by subjects so as to make reader group differences more pronounced. In evaluating the results of this experiment, two points should be kept in mind. First, in the temporal terms experiments, there were two sources of errors: ordering and object selection errors. For example, in (11), an ordering error would consist of pushing the smallest horse first, whereas an object selection error might result in picking up the middle-sized horse. Unless otherwise indicated, errors of both types were combined in the analyses. Second, all of the analyses of variance we report treated subjects, and not items, as a random factor. Therefore, the findings cannot be generalized to items outside of these experiments.

Working memory and reading disorder

505

12. It should be noted that the obvious control sentences - those sentences that express agreement between order of mention and conceptual order — are not adequate for our purposes. Correct responses to these sentence might not be indicative of a subjects successful comprehension; rather, correct responses could reflect a simple strategy to act out complex sentences in an order of mention fashion. That is, they might evoke the right response for the wrong reason. 13. It should be noted that a processing limitation account would be compatible with one kind of interaction, in which reader group differences increase to a greater extent on sentences that are particularly costly of working memory resources, such as relative clauses and deep PPs. There is a hint of this type of interaction in Figure 18.4, and this interaction has been obtained with other populations who exhibit more severe limitations in memory capacity (e.g., Lukatela, 1989). 14. It is worth noting that even normal adults respond incorrectly to garden path sentences in apparent violation of Subjacency when demands on working memory are raised substantially (Crain & Fodor, 1985). This supports our expectation that the problems poor readers have with these same sentences are also due to processing factors. 15. The two-letter code indicates how these sentence types differ. The first letter represents the noun phrase in the matrix sentence (subject or object) that bears the relative clause; the second letter represents the empty noun phrase position in the relative clause. For example, an OS relative like (20) is one in which the object of the main clause is modified by a relative clause with a superficially null subject.

References Amidon, A., & Carey, P. (1972). Why five-year-olds cannot understand before and after. Journal of Verbal Learning and Verbal Behaviour, 11, 417-423. Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press. Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47-89). New York: Academic Press. Baddeley, A. D., Vallar, G., & Wilson, B. (1987). Comprehension and the articulatory loop: Some neuropsychological evidence. In M. Coltheart (Ed.), Attention and performance XII (pp. 509-529). London: Erlbaum. Bradley, L, & Bryant, P. (1983). Difficulties in auditory organization as a possible cause of reading backwardness. Nature, 271, 746-747. Brown, R. (1973). A first language. Cambridge, MA: Harvard University Press. Byrne, B. (1981). Deficient syntactic control in poor readers: Is a weak phonetic memory code responsible? Applied Psycholinguistics, 2, 201—212. Carpenter, P. A., & Just, M. A. (1988). The role of working memory in language comprehension. In D. Klahr & K. Kotovsky (Eds.), Complex information processing: The impact of Herbert A. Simon. Hillsdale, NJ: Erlbaum. Chomsky, C. (1969). The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT Press. Clark, E. V. (1970). How young children describe events in time. In G. B. Flores d'Arcais & W. J. M. Levelt (Eds.), Advances in psycholinguistics. Amsterdam: North Holland. Conrad, R. (1964). Acoustic confusions in immediate memory. British Journal of Psychology, 3, 75-84. Conrad, R. (1972). Speech and reading. In J. F. Kavanagh & I. G. Mattingly (Eds.), Language by ear and by eye: The relationships between speech and reading (pp. 205-240). Cambridge, MA: MIT Press. Cossu, G., Shankweiler, D., Liberman, I. Y., Tola, G., & Katz, L. (1988). Awareness of phonological segments and reading ability in Italian children. Applied Psycholinguistics, 9, 1-16.

506

Crain, Shankweiler, Macaruso, and Bar-Shalom

Crain, S. (1982). Temporal terms: Mastery by age five. In Papers and Reports on Child Development, 21, 33-38 (Stanford University). Crain, S., & Fodor, J. D. (1985). On the innateness of Subjacency. In the Proceedings of the Eastern States Conference on Linguistics (Vol. 1). Columbus, OH: Ohio State University. Crain, S., & McKee, C. (1985). Acquisition of structural restrictions on anaphora. In the Proceedings of the Northeast Linguistics Society, 16, Amherst, MA: University of Massachusetts. Crain, S., & Nakayama, M. (1987). Structure dependence in grammar formation. Language, 63, 522-543. Crain, S., & Shankweiler, D. (1988). Syntactic complexity and reading acquisition. In A. Davison & G. Green (Eds.), Linguistic complexity and text comprehension: Readability issues reconsidered. Hillsdale, NJ: Erlbaum. Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by the psychological syntax processor. In D. R. Dowty, L. Karttunen, & A. Zwicky (Eds.), Natural language parsing: Psychological, computational, and theoretical perspectives (pp. 320—358). Cambridge: Cambridge University Press. Crain, S., Thornton, R., & Murasugi, K. (1987). Capturing the evasive passive. Paper presented at the Boston University Conference on Language Development, Boston, MA. Crain, W. (1986). Restrictions on the comprehension of syntax by mentally retarded adults. Unpublished doctoral dissertation, Claremont Graduate School, Claremont, CA. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behaviour, 19, 450-466. Denckla, M. B., & Rudel, R. G. (1976). Naming of object-drawings by dyslexic and other learning disabled children. Brain and Language, 3, 1—15. de Villiers, J. G., Tager-Flusberg, H. B. T., Hakuta, K., & Cohen, M. (1979). Children's comprehension of relative clauses. Journal of Pycholinguistic Research, 8, 499-518. Dunn, L, & Dunn, L. (1981). Peabody Picture Vocabulary Test - Revised. Circle Pines, MN; American Guidance Service. Fletcher, J. M., Satz, P., & Scholes, R. (1981). Developmental changes in the linguistic performance correlates of reading achievement. Brain and Language, 13, 78—90. Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press. Frazier, L. (1978). On comprehending sentences: Syntactic parsing strategies. Unpublished doctoral dissertation, University of Connecticut, Storrs. Goldsmith, S. (1980). The psycholinguistic bases of reading disability: A study in sentence comprehension. Unpublished doctoral dissertation, The City University of New York. Gorrell, P., Crain, S., & Fodor, J. D. (1986). Contextual information and temporal terms. Paper presented at the Boston University Conference on Language Development, Boston, MA. Hamburger, H., & Crain, S. (1982). Relative acquisition. In S. Kuczaj II (Ed.), Language development: Vol. 1 Syntax and semantics. Hillsdale, NJ: Erlbaum. Hamburger, H., & Crain, S. (1984). Acquisition of cognitive compiling. Cognition, 17, 85-136. Hamburger, H., & Crain, S. (1987). Plans and semantics in human processing of language. Cognitive Science, 11, 101—136. Jansky, J., & de Hirsch, K. (1972). Preventing reading failure: Prediction, diagnosis, intervention. New York: Harper & Row. Johnson, M. L. (1975). The meaning of before and after for pre-school children. Journal of Experimental Child Psychology, 19, 88-99. Katz, R. B. (1985). Phonological deficiencies in children with reading disability: Evidence from an object-naming task. Cognition, 22, 225-257. Kimball, J. P. (1973). Seven principles of surface structure parsing. Cognition, 2, 15—47. Lewis, D. (1979). Scorekeeping in a language game. Journal of Philosophical Logic, 8, 339-359. Liberman, A. M., Mattingly, I. G., & Turvey, M. T. (1972). Language codes and memory codes. In A. W. Melton & E. Martin (Eds.), Coding processes and human memory. Washington, DC: Winston.

Working memory and reading disorder

507

Liberman, I. Y., & Shankweiler, D. (1985). Phonology and the problem of learning to read and write. Remedial and Special Education, 6, 8—17.

Liberman, I. Y., Shankweiler, D., Liberman, A. M., Fowler, G, & Fischer, F. W. (1977). Phonetic segmentation and recoding in the beginning reader. In A. S. Reber & D. L. Scarborough (Eds.), Toward a psychology of reading: The Proceedings of the CUNY Conferences. Hillsdale, NJ:

Erlbaum. Linebarger, M. C, Schwartz, M. F., & Saffran, E. M. (1983). Sensitivity to grammatical structure in so-called agrammatic aphasics. Cognition, 13, 361—392. Lukatela, K. (1989). Sentence processing in fluent and non-fluent aphasia. Unpublished doctoral

dissertation, University of Connecticut, Storrs. Lundberg, I., Oloffson, A., & Wall, S. (1980). Reading and spelling skills in the first school years predicted from phonemic awareness skills in kindergarten. Scandinavian Journal of Psychology, 21, 159-173. Mann, V. A., Liberman, I. Y., & Shankweiler, D. (1980). Children's memory for sentences and word strings in relation to reading ability. Memory and Cognition, 8, 329-335. Mann, V. A., Shankweiler, D., & Smith, S. T. (1984). The association between comprehension of spoken sentences and early reading ability: The role of phonetic representation. Journal of Child Language, 11, 6 2 7 - 6 4 3 .

Marin, O. S. M., Saffran, E. M., and Schwartz, M. F. (1976). Dissociations of language in aphasia: Implications for normal function. Annals of the New York Academy of Science, 2S0, S6S-8S4.

Marlsen-Wilson, W. D., & Tyler, L. K. (1980). The temporal structure of spoken language understanding: The perception of sentences and words in sentences. Cognition, 8, 1-7A. Martin, R. C (1985). The relationship between short-term memory and sentence comprehension deficits

in agrammatic and conduction aphasics. Paper presented at the Annual Meeting of the Academy of Aphasia, Pittsburgh, PA. Matthei, E. M. (1981). Children's interpretation of sentences containing reciprocals. In S. L. Tavakolian (Ed.), Language acquisition and linguistic theory. Cambridge, MA: MIT Press.

Mattingly, I. G., & Liberman, A. M. (1988). Specialized perceiving systems for speech and other biologically significant sounds. In G. M. Edelman, W. E. Gall, & W. M. Cowan (Eds.), Auditory function: The neurobiological bases of hearing. N e w York: Wiley.

Miller, G. A. (1956). The magical number seven, plus or minus two: Some limitations of our capacity for processing information. Psychological Review, 60, 81-97. Morais, J., Cluytens, M., & Alegria, J. (1984). Segmentation abilities of dyslexics and normal readers. Perceptual and Motor Skills, 58, 221—222.

Olson, R. K., Davidson, B. J., Kliegl, R., & Davies, S. E. (1984). Development of phonetic memory in disabled and normal readers. Journal of Experimental Child Psychology, 37, 187—206. Otsu, Y. (1981). Universal grammar and syntactic development in children: Toward a theory of syntactic

development. Unpublished doctoral dissertation, Massachusetts Institute of Technology, Cambridge, MA. Perfetti, G A. (1985). Reading ability. New York: Oxford University Press. Perfetti, C A., & Lesgold, A. M. (1977). Discourse comprehension and sources of individual differences. In M. A. Just & P. A. Carpenter (Eds.), Cognitive processes in comprehension. Hillsdale, NJ: Erlbaum. Richardson, E., & DiBenedetto, B. (1986). Decoding Skills Test. Parkton, MD: York Press. Roeper, T. W. (1972). Approaches to a theory of language acquisition with examples from German

children. Unpublished doctoral dissertation, Harvard University, Cambridge, MA. Saffran, E. M. (1985). Short-term memory and sentence processing: Evidence from a case study. Paper

presented at the Annual Meeting of the Academy of Aphasia, Pittsburgh, PA. Scholes, R. J. (1978). Syntactic and lexical components of sentence comprehension. In A. Caramazza & E. B. Zurif (Eds.), Language acquisition and language breakdown: Parallels and

divergences (pp. 163-193). Baltimore, MD: Johns Hopkins University Press.

508

Crain, Shankweiler, Macaruso, and Bar-Shalom

Shankweiler, D., & Crain, S. (1986). Language mechanisms and reading disorder: A modular approach. Cognition, 24, 139-168. Shankweiler, D., Crain S., Brady, S., & Macaruso, P. (in press). Identifying the causes of reading disability. In P. B. Gough (Ed.), Reading acquisition. Hillsdale, NJ: Erlbaum. Shankweiler, D., Crain, S., Gorrel, P., & Tuller, B. (1989). Reception of language in Broca's aphasia. Language and Cognitive Processes, 4, 1-33. Shankweiler, D., & Liberman, I. Y. (1972). Misreading: A search for causes. In J. F. Kavanagh & I. G. Mattingly (Eds.), Language by ear and by eye: The relationships between speech and reading. Cambridge, MA: MIT Press. Shankweiler, D., Liberman, I. Y., Mark, L. S., Fowler, C. A., & Fischer, F. W. (1979). The speech code and learning to read. Journal of Experimental Psychology: Human Learning and Memory, 5, 531-545. Sheldon, A. (1974). The role of parallel function in the acquisition of relative clauses in English. Journal of Verbal Learning and Verbal Behavior, 13, 272-281. Smith, S. T., Macaruso, P., Shankweiler, D., & Crain, S. (in press). Syntactic comprehension in young poor readers. Applied Psycholinguistics. Stanovich, K. E. (1982). Individual differences in the cognitive processes of reading: 1. Word decoding. Journal of Learning Disabilities, 15, 449-512. Stein, C. L., Cairns, H. S., & Zurif, E. B. (1984). Sentence comprehension limitations related to syntactic deficits in reading-disabled children. Applied Psycholinguistics, 5, 305-322. Tavakolian, S. L. (1981). Language acquisition and linguistic theory. Cambridge, MA: MIT Press. Vellutino, F. R. (1979). Dyslexia: Theory and research. Cambridge MA: MIT Press. Wagner, R. K., & Torgesen, J. K. (1987). The nature of phonological processing and its causal role in the acquisition of reading skills. Psychological Bulletin, 101, 192-212. Whitaker, H. (1976). A case of isolation of the language function. In H. Whitaker and H. A. Whitaker (Eds.), Studies in neurolinguistics (Vol. 2). New York: Academic Press. Wingfield, A., & Butterworth, B. L. (1984). Running memory for sentences and parts of sentences: syntactic parsing as a control function in working memory. In H. Bouma and D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 351-363). Hillsdale, NJ: Erlbaum. Wolf, M. (1981). The word-retrieval process and reading in children and aphasics. In K. Nelson (Ed.), Children's language (Vol. 3). New York: Gardner Press.

Name index

Adam, R, 107 Alajouanine, T., 40 Alberoni, M, 61 Albert, M L, 16n, 20, 13, 358 Alexander, G. E., 325 Allegria, J., 65, 479 Allik, J. P., 227 Allport, D. A., 4, 24, 76, 81, 81, 108, 116, 161, 191, 210n, 288, 289, 310, 313, 345, 349, 353, 354, 430, 457, 464, 465 Amidon, A., 488 Anderson, C. B. M, 37 Anderson, J. A., 4 Anderson, P. A., 248, 262 Angelergues, R., 13 Arbuckle, T. Y., 88, 290, 385n Aten, J., 124 Atkinson, R. G, 2, 3, 7, 11, 17, 56, 117, 111 Avons, S., 297 Ayres, T. J., 276 Bachanan, M., 289 Baddeley, A. D., 2, 3, 8, 11, 16-19, 21, 22, 24, 15, 17-32, 34-6, 39, 46, 55-61, 64-7, 69, 70, 74, 76-8, 80, 81, 83, 87, 89, 90, 116, 123, 125, 126, 154, 168, 176, 188, 208, 209, 210n, 216, 218, 221, 224, 225f, 116, 118, 131, 241, 248-50, 252, 257, 261-5, 288-90, 307, 308, 311, 313, 324, 327, 334, 337, 338, 340, 342, 345, 348, 351, 358, 364, 365, 367, 368, 370, 371, 375, 381, 384n, 390, 400, 403, 404, 412, 419, 421, 429, 446n, 448-50, 457-9, 464, 465, 467, 469-71, 479, 481, 483, 503n Badecker, W., 43 Baker, C, 208, 338, 340, 348, 374, 446 509

Baker, E., 76, 90, 357 Baker, S. L, 137 Baldwin, J. M., 221 Barbizet, J., 320 Barley, M, 269 Barnard, P., 25, 141, 161, 181, 224, 278, 394, 401 Barnard, T., 381 Barnes, M, 343 Barresi, J., 190 Barrett, G., 95, 101, 107, 108, 346 Basili, A. G., 2, 16n, 13, 39n, 40, 68, 79, 116, 146, 295, 339, 349, 355, 358, 359, 390, 403, 405, 419, 429, 441, 443, 446, 448, 458, 464, 471 Basso, A., 16n, 18, 35, 37, 38, 39n, 241, 358, 364, 403, 428, 448, 450 Beattie, G., 168 Beirman, K., 321 Benson, D. F., 41 Benton, A. L, 123, 452 Berndt, R. J., 2, 16n, 24, 39n, 40, 68, 79, 116, 138, 146, 310, 333, 338, 339, 349, 354, 355, 358, 359, 377, 390, 403, 405, 419, 429, 441, 443, 446, 448, 449, 458, 464, 466, 471, 472 Bernholtz, N. A., 452 Berti, A., 42 Berwick, R. C, 344, 391, 392, 394, 418, 444 Besner, D., 34, 66, 74, 75, 76, 289, 290, 311, 312, 314, 315n, 384 Bever, T. G., 201, 250, 347, 470, 471 Bishop, D. V. M, 63, 194, 195f, 311, 368, 411 Bisiach, E., 42 Bjork, R. A., 22, 35 Black, S., 340

510

Name index

Block, R. R, 326 Bloom, P. A., 171 Blumstein, S., 76, 90, 142, 357, 453 Boakes, R. A., 189 Bochetto, P., 431 Bock, J. K., 153, 161 Bolton, L, 145 Borkowski, J. G., 162 Bortolini, U., 451, 455 Bouchard, R., 41 Bower, P. M., 64 Bradley, L, 64, 479 Brady, S., 477, 491 Bramwell, B., 274 Breedin, S., 358, 369, 370, 390, 403, 419 Brener, R., 145, 162, 164 Bressi, S., 70, 262, 263 Broadbent, G, 61 Broadbent, D. E., 2, 11, 19, 37, 54, 222, 224 Broadbent, M, 19, 37 Brown, G. D. A., 289, 290 Brown, J., 54, 191 Brown, R. M, 237, 503n Bryant, P. E., 64, 479 Bub, A., 293, 340 Bub, D., 346 Bub, J., 346 Buchanan, ML, 28, 29, 30, 31, 46, 59, 337, 403 Bushke, H., 55, 188 Butterworth, B., 3, 16n, 17, 30, 41, 66, 68, 74, 76, 116, 187, 190-2, 198, 205, 208, 209, 211, 310, 345, 349, 353, 358, 371, 377, 380, 390, 403, 404, 420, 421, 423, 445, 449, 450, 471, 482 Byng, S. C, 295 Byrd, M, 249 Byrne, B., 478, 500, 5O3n

Caramazza, A., 1, 2, 12, 16n, 19, 23, 39n, 40, 43, 45, 68, 79, 116, 146, 168, 295, 310, 333, 338, 339, 347, 349, 355, 358, 359, 390, 402, 403, 405, 419, 433, 448, 449, 458, 464, 466, 471 Carey, P., 488 Carlson, G. N., 346 Carlson, M, 392, 396 Carpenter, P. A., 248, 249, 252, 261, 390-2, 397, 398, 419, 481, 503n Case, R. M., 240 Cavallotti, S. F., 41 Cavanaugh, ]., 82 Ceralla, J., 248 Cermack, L. S., 218 Chapman, R., 346 Chase, W. G., 253, 405 Chomsky, C, 503 Clark, E. V., 188, 333, 344, 390, 428, 443, 449, 466, 487 Clark, K H., 188, 253, 333, 344, 390, 428, 443, 449, 466 Clifton, C, 346, 405 Cluytens, M, 479 Cohen, G., 248 Cohen, M, 408, 503n Cohen, N. J., 323 Cole, M., 223 Cole, R. A., 313 Colle, H. A., 59, 60, 76, 89 Collins, G. I., 107 Coltheart, M, 1, 2, 43, 79, 81, 117, 140, 222, 242, 307 Coltheart, V., 297 Conrad, R., 17, 18, 19, 27, 46, 55, 59, 74, 188, 229, 337, 479 Content, A., 65 Cook, V. J., 348 Corin, W., 79 Cairns, H. S., 478, 500, 503n Cornoldi, C, 455 Correl, P., 480 Caltagirone, C, 453 Campbell, R., 3, 16n, 17, 31, 66, 68, 78, 79, Coslett, H. B., 146, 148 188, 269-73, 276-8, 281, 310, 345, 371, Cossu, G., 479 Coughlan, A. K., 170 377, 380, 390, 404, 420, 421, 423, 445 Craik, F. I. Mv 7, 11, 17, 22, 24, 25, 35, 37, Cancelliere, A., 293 57, 188, 190, 247-9, 258, 260-2, 264, Capitani, E., 450, 452 265, 310, 326, 466 Caplan, D., 208, 338, 340, 346-8, 358, 366, Crain, S., 391, 477, 480-2, 484-6, 488, 374, 376, 390, 395, 418, 419, 446, 449, 490, 491, 494, 500, 503n, 504n 450, 470, 471 Crick, F., 11 Cappa, S. F., 27, 29, 33, 34, 40, 41, 62, 74, Crowder, R. G., 11, 18, 19, 35, 37, 38, 74, 76, 117, 133, 187, 190, 198, 208, 209, 76-9, 116, 128, 187, 189, 196, 276, 277, 405, 450, 452, 457, 464

511

Name index 281, 288, 304, 310, 311, 324, 337, 401 Cubelli, R., 62 Cunitz, A. R., 21, 55 Damasio, A. R., 33, 41, 42 Damasio, H , 33, 41, 42 Daneman, M , 248, 249, 253, 261, 392, 397, 398, 419, 481 Daniels, S., 74, 76, 289, 290, 312, 315n Darwin, G, 78, 80, 337 Davelaar, E., 74, 75, 76, 289, 305, 311, 312, 314 Davidson, B.J., 479 Davies, J., 74, 76, 289, 290, 312, 315n Davis, R., 63 Dehaut, F., 348 Dell, G. S., 161, 163 Delia Sala, S., 62, 70, 262, 263 Dempster, F. N., 223, 240 Denckla, M B., 479 De Renzi, E., 119, 169, 407, 420, 451, 452 Deutsch, J. A., 324 De Villiers, J. G., 408, 503n DiBenedetto, B., 491 Dodd, B., 78, 79, 117, 133, 188, 269, 277, 278 Donchin, E., 94, 97 Dowty, D. R., 348 Drewnowski, A., 117, 129, 154 Dubois, J., 23 Dunn, L. M , 354, 492 Dworetzky, B., 142

Fodor, J. A., 201, 222, 250, 480 Fodor, J. D., 168, 328, 345, 346, 383, 444, 488, 490 Ford, J. M , 107, 108 Ford, M , 396, 408, 412 Forster, K. I., 161, 397, 412 Francis, W. N., 121, 131, 151, 164n, 250 Frankel, F., 223 Frankenhauser, M , 321 Franklin, S., 273, 274, 342, 374 Frauenfelder, A. R , 383 Frazier, L, 168, 346, 391, 392, 395, 396, 398, 429, 444, 489 Friedrich, F. J., 16n, 20, 37, 38, 39n, 40, 42, 76, 80, 82, 83, 88, 90, 116, 349, 354, 390, 403, 412, 415, 420, 429-31, 449, 453, 462, 464, 465, 471 Fuster, J. M , 325 Futter, C, 347 Gaes, G., 321 Gaillard, A. W. K., 107 Gainotti, G., 321 Gardiner, J. M , 35, 37 Gardner, R , 23, 38, 39, 358 Garnham, A., 183 Garnsey, S., 346 Garret, M. F., 201, 250 Garwood, }., 273, 277 Gathercole, S. E., 37, 65, 70, 117, 276 Gazdar, G., 209 Geschwind, H , 23, 41

Gick, M. L, 249 Eich, E., 397 Eisenberg, K., 177, 184 Eisner, H. C, 162 Eldridge, M , 46, 66 Ellis, A. W., 1, 12, 27, 30, 43, 46, 108, 168 Ellis, N. C, 65, 69 Elman, J. L, 44, 75, 84, 117, 161, 190, 217, 270, 276, 280 Elton, C, 297 Estes, W. K., 337 Faglioni, P., 169, 451 Farah, M , 59 Feher, E., 405, 407, 408f, 409f, 411, 413 Feldman, J. A., 12 Ferreira, F., 346 Fischler, I., 171 Fletcher, J. M , 478, 503n Flusberg, R , 408

Gilewski, M. J., 248 Glanzer, M., 3, 11, 21, 35, 55, 222 Gleason, J. B., 452 Glenberg, A. M , 36 Glenn, C. G., 16n, 20, 37, 38, 39n, 40, 42, 76, 80, 82, 83, 88, 90, 116, 349, 354, 403, 412, 415, 420, 430, 431, 453, 464, 465 Goldberg, J., 240 Goldfield, A., 270 Goldman, S., 400 Goldsmith, S., 503n Gomer, F. E., 107 Goodglass, R , 23, 76, 90, 118, 357, 452 Gordon, B., 151 Gormican, S., 163 Gorrell, P., 488 Grant, W., 59 Green, D. W., 27

512

Name index

Green, E., 41 Greene, R. L, 276, 324, 337 Greenstein, J.J., 69 Gregg, V. H., 35, 37 Grier, J. B., 367 Grodzinsky, Y., 367 Gross, A. E., 190 Hacker, M. J., 327 Hagan, J. W., 230 Hakuta, K., 408, 503n Hall, J. W., 64 Halliday, M. S., 29, 104, 224, 226-9, 231-3, 235, 237, 238, 240, 243 Hamburger, H., 480, 484, 485, 486, 488, 500, 503n, 504n Hamsher, K. D., 123, 452 Hanley, J. R., 61 Hasher, L, 264 Hayes, D. S., 237 Hebb, D. O., 2, 12, 54, 159 Hecaen, H., 23 Heilman, K. M., 168, 403 Henik, A., 88, 277 Herman, S. J., 190 Heron, A., 24 Hicks, R. E., 321, 326, 328-9 Hildebrandt, N., 346-8, 358, 374, 376, 390, 395, 418, 419 Hillyard, S. A., 94 Hink, R. F., 94 Hinton, G. E., 4, 11, 44, 69 Hirsch, K., 479 Hitch, G. J., 8, 11, 22, 25, 27, 29, 35, 36, 57, 58, 168, 224, 226, 227, 228f, 229, 231-3, 235, 237, 240, 243, 248-50, 252, 261, 400, 472, 481 Hopkins, W. F., 107, 108 Howard, D., 3, 17, 66, 68, 187, 190, 208, 209, 221, 224, 225f, 273, 274, 342, 345, 371, 374, 377, 380, 390, 420, 423, 445, 449, 450, 471 Howell, J., 340 Howes, D. H., 41 Huey, E. B., 66 Hull, A. J., 55 Hulme, C., 226, 227, 229 Humphreys, M. S., 64 Hunt, E. B., 400 Huttenlocher, J., 177, 184, 371 Hyde, M. R., 452

Jacoby, L. L, 164n James, C T , 151 James, W., 218, 319, 320, 327 Jansky, J., 479 Jarvella, R. J., 3, 25, 189, 190, 366, 428, 443 Jeffers, J., 269 Jerger, J., 412 Jerger, S., 358, 369, 370, 390, 403, 412, 419 Johnson, M. L, 487 Johnson, S. T., 28, 337 Johnson-Laird, P. N., 183, 188, 208, 368 Johnston, R., 64 Jordan, L. S., 39, 47 Jordan, M. I., 44 Just, M. A., 390-2, 398, 481, 503n Kahneman, D., 277 Kail, R. V., 223, 230 Kaplan, E., 23, 118 Kartunen, L, 348 Katz, L, 479 Kawamoto, A. H , 12 Kemper, S., 80, 83, 349, 354, 390, 412, 429, 449, 462, 471 Kertesz, A., 293, 340 Kimball, J. P., 489 Kinsbourne, M , 16n, 22, 38, 45n, 108, 320, 321, 327-9, 358 Kintsch, W., 55, 188, 390, 397 Klapp, S. T., 400 Klatt, D. R , 117 Klatzky, R., 400 Kleiman, G. M , 66, 338 Kleist, K., 46 Kliegl, R., 479 Koller, J. J., 2, 16n, 23, 39n, 40, 68, 116, 146, 295, 339, 349, 355, 358, 359, 390, 403, 405, 419, 429, 441, 443, 446, 448, 458, 464, 471 Komoda, M. K., 88, 290, 385n Koppell, B. S., 107, 108 Koppenaal, L, 35 Kroll, J. F., 130, 134 Kucera, H., 121, 131, 151, 164n, 250 Kurland, M. D., 240 Kutas, M , 94 Laiacona, M , 452 Landis, T., 273 Large, B., 65 Lawson, E. A., 107 Laxon, V., 297

Name index Ledoux, D., 40 Leiman, J. L, 343 Lesgold, A. M., 481 Lesser, R., 169, 170 Levy, B. A., 28, 76, 190, 196, 337, 403 Lewis, D., 494 Lewis, V. J., 28, 29, 31, 46, 60, 61, 66, 230, 231, 290, 308, 337, 338, 400 Lezak, M. D., 120 Lhermitte, F., 40 Liberman, A. M, 479 Liberman, I. Y., 64, 478-80 Lieberman, K., 59 Lifschultz, A. J., 277 Light, L. L, 248, 262 Linebarger, M. C, 198, 199, 367, 405, 431, 433, 437, 441, 442, 446n, 449, 460, 471, 480 Littler, J. E., 227 Lockhart, R. S., 8, 57, 316, 466 Lodwick, B., 189 Logie, R. R, 62, 262, 263, 277 Logue, V., 16n, 21, 35, 38, 39n, 42, 70, 191, 205, 240, 262, 263, 349, 353, 358, 428, 448 Loveless, N. E., 94 Luce, P. A., 117 Lukatela, K., 505n Lundberg, I., 479 Lunneberg, C, 400 Luria, A. R., 25, 41, 320 Macaruso, P., 477, 491, 500, 503n McCallum, W. C, 97 McCarthy, R. A., 16n, 18, 25, 26f, 38, 39n, 45n, 75, 83, 168, 169, 171, 172, 175, 176, 178, 179, 183, 187, 189, 190, 208, 338, 345, 358, 370, 382, 390, 418, 448, 449, 467, 472 McClelland, J. L, 4, 12, 44, 75, 84, 85, 117, 161, 190, 217, 270, 276, 280 McCutchen, D., 377 McDonald, I., 104 MacDonald, J., 270 McDowd, J. M., 249, 262 McGrath, M., 270 McGurk, H., 270 McKee, G, 481 McLean, J., 270 McLeod, P. D., 29, 63, 87, 88, 115, 134, 241, 264

513

Madigan, S., 151, 164n Mann, V. A., 64, 479, 500-3 Maratsos, M., 209, 395, 409, 428 Marcie, P., 13 Marcus, M. P., 168, 344, 394, 444 Marcus, S. M., 277 Margrain, S. A., 19 Marin, O. S. M., 2, 16n, 20, 24, 29, 30, 37-40, 42, 68, 76, 80, 81, 83, 88, 90, 116, 168, 176, 199, 200f, 208, 209, 272, 339, 347, 349, 353, 354, 357, 358, 390, 403, 412, 415, 419, 420, 423, 430, 431, 448, 450, 453, 462, 464, 465, 480 Marr, D., 444 Marsh, G. R., 107 Marshall, J. G, 1, 2, 43 Marslen-Wilson, W. Dv 383, 391, 393, 467, 481 Martin, N., 157 Martin, R. G, 80, 83, 349, 355, 358, 369, 370, 390, 401, 403, 405, 406f, 411-13, 419, 429, 431, 449, 453, 462, 465, 471, 482 Masani, P., 187 Massaro, D. W., 270 Massullo, C, 453 Matthei, E. M., 485 Mattingly, I. G., 479, 480 Maufras Du Chatelier, A., 23 Mellish, C, 209 Melton, A. W., 54 Mervis, J. S., 130, 134 Meudell, P. R., 63, 241 Miceli, G., 272, 453 Michalewski, H. J., 95, 108 Michon, J. A., 324 Milberg, W., 142 Miles, T. R., 69 Miller, G. A., 2, 187, 399, 489 Miller, G. W., 321, 326 Milner, B., 56 Mohs, R. C, 107, 108 Monsell, S., 25, 85, 111, 140, 164n, 181, 224, 381, 384n Moore, M., 248 Morais, J., 65, 479 Moray, N., 63 Morris, R. G., 249, 264 Morton, J., 3, 12, 25, 27, 38, 77, 116, 128, 222, 176, 177, 281, 310, 337 Mueller, C W., 207

514

Name index

Murasugi, K., 481 Murdock, B. B., 17, 21, 117, 129, 154, 321 Murray, D.J., 28, 337 Mushin, J., 104 Muter, P., 327 Naatanen, R., 94 Nairne, J. S., 79 Nakayama, M v 481 Neisser, U., 17, 44 Nelson, R., 35 Newell, A., 11, 44 Nichelli, P., 61 Nicolson, R., 226 Norman, D. A., 2, 3, 7, 11, 13, 17, 21, 44, 190, 247, 260, 263 Novelli, G., 452

Poon, L, 248 Posner, M. L, 29, 63, 87, 88, 264 Postman, L, 58, 132, 326 Potter, M. C, 428 Poynter, W. D., 326 Pratt, H., 95, 108 Pratt, R. T. C, 16n, 21, 35, 38, 39n, 42, 191, 205, 241, 349, 353, 358, 428, 448 Previdi, P., 452 Price, D., 41 Prill, K., 249, 263 Prisko, L, 324 Prussin, H. A., 77, 277 Pullum, G., 209 Pulman, S. G., 467, 468

Rabin, P., 29, 47 Rabinowitz, J. C, 247 O'Donnell, R. D., 107 Rashotte, C A., 69 Ojemann, G. A., 41 Rayner, K., 392, 396, 398 Olbrei, I., 412 Renaud, D., 40 Oldfield, R. G, 151 Richardson, E., 491 Oloffson, A., 479 . Richardson, J. T. E., 145 Olson, R. K., 479 Rickard, M., 297 Ornstein, R. E., 326 Risse, G. L, 29, 47 Ostergaard, A. L, 63 Ritter, W., 97 Ostrin, R. K., 403, 405, 422, 462, 463 Roeper, T. W., 485 Otsu, Y., 490 Rogan, J. D., 249, 263 Ottley, P., 277 Rosen, S., 277 Roth, W. T., 107, 108 Paivio, A., 151, 162, 164n Papagno, C, 31, 32, 36, 38>, 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
521

defined, 457; pre- and postcategorical distinction, 78-9; pre- and postlexical, 81-4; relationship with auditory memory, 116; and sentence comprehension, 357 phonological dyslexia, 83 phonological processing (see also phonological coding; phonological storage); deficits and preserved comprehension, 452-3; and lexical access, 357; mapping hypothesis of, 471; and ordered recall, 479; patients with deficits in, 349—62, 377—82; and phonological processing, 465-6; and sentence comprehension, 357, 448-73; and single word comprehension, 357 phonological similarity: and articulatory loop, 59; and articulatory suppression, 76, 289-90; auditory vs. visual input, 89, 368; coding effects, 55; in conduction aphasia, 453-4; effects in children, 228-30; effects in poor readers, 64; explanation of, 230; in input-output stores, 282; and matching span, 300-1 phonological storage (see also auditory-verbal STM; phonological buffer; phonological processing; recall, immediate; short-term memory); and auditory word recognition, 117; as backup, 449-50; and comprehension deficits, role in, 483; contents of the input store, 188-91; deficit patterns in patients, 334, 342, 380; defined, 482; defining the capacity of, 145; distinguished from other STM components, 340; and immediate memory, 280; and language processing, 338; modality effects in, 121-3; patients with disturbances of, 360-74; and phonological processing, 465-6; and phonological similarity, 125-6; and precategorical acoustic storage, 278—80; and recency effects, 126; recovering from misinterpretations, 414-15; and registration, 117; and rehearsal, 281-2; relationship with other stores, 190; relationship with other systems, 111; relationship with verbal memory, 282; and repetition, 117; retention of meaning in, 188; and sentence comprehension, 332-5, 466-72, 469; and sentence processing, 411-18; and sentence repetition, 415-17; and sub vocal rehearsal, 403; and syntactic processing,

522

Subject index

phonological storage {cont.) 402, 418; and working memory, 288 phonological transpositions, 182 potentials, event-related, see event-related potentials precategorical acoustic store, 77 {see also auditory-verbal STM; phonological buffer; phonological processing; phonological storage; recall, immediate; short-term memory); defined, 276; and phonological memory, 278-80; and recency effects, 280-1 primacy: and proactive interference, 234; and trace decay, 234 priming tasks, 345; repetition, 159-60; semantic, 158 privileged loop, and working memory, 234-5 proactive interference, and primacy, 234 probe identification tasks, 94-103, 154-5; nonverbal sounds, 127-8 processing limitation hypothesis, 478; contrasted with structural lag hypothesis, 502 prospective memory, 325-6 pseudohomophones: articulatory suppression and recall of, 306; detection of, 289-90, 345; and input-output phonology, 288; visual list recall, 305-6 Rapid Serial Visual Presentation (RSVP), 343 Raven's Progressive Matrices, 452 reaction time: electrophysiological measures, 97-107; words-nonwords, 138 reading, 482-4 (see also reading comprehension; dyslexia); garden path effects in poor readers, 497-502; grammatical competence in poor readers, 499-502; memory limitations of poor readers, 493-7; a structural explanation in children, 488; syntactic knowledge in poor readers, 498-9 reading comprehension, 398; in children, 477-503; in poor readers, 491; process limitation hypothesis, 478; and spoken sentence comprehension, 477-80; structural lag hypothesis, 478; and syntactic processing in children, 478 reasoning, verbal, STM demands of, 472 recall (see also recall, immediate; word lists); of complex sentences, 200-2, 210; failure to recall last items first, 134; of gist and

comprehension, 206; lexical factors in recall, 140; of lipread lists, 278; of lists vs. sentences, 206-7; ordered, and phonological processing, 479; probability of recall of lists, 195-7; role of auditory nonphonological code, 140; of sentences and grammatical structure, 202-3; serial, and intrusion errors, 116; serial, and recency effects, 126; serial, role of STM in, 142; of six-item lists, 194; word list and presentation rate, 230-5 recall, immediate: and articulatory suppression, 60-1; and nonphonological systems, 112; and recency effect, 55; role of phonological store and semantic system in, 472-3; Type I and II recall, 140-1 recency effect, 134-5; in amnesics, 56; delayed, 57; digit list repetition, 129; electrophysiological measures of, 100-1; in free recall, 58; and immediate recall, 55; and lipreading, 141, 276-7; modality specific, 280-1; and phonological store, 464; and suffix words, 141 regularity effects, 343 rehearsal: articulatory and phonological store, 465; memory without rehearsal, 309; and phonological similarity effect, 464; vocal rehearsal of visual list, 304-5; and word length effect, 464 relative clauses: children's performance on, 486-7, 489-9; and reading disability, 500-2 remote memory, 119; see also amnesia; ecphory; episodic recollection; long-term memory repetition: known vs. unknown vocabulary, 174; and lexical functions, 146; lexical-semantic processing, role of, 172-4; of lists and phonological store, 472-3; of lists vs. sentences, 171, 180-1; modality effects in, 120-1; model of verbal repetition, 279; of nonwords, 157-8, 217-18; of sentences, 112, 294; of sentences, complete vs. incomplete, 171-2; of sentences and comprehension, 399-400; sentence, in a conduction aphasic, 461-4; sentence, and nonphonological factors, 462-4; of sentences and phonological store, 472-3; serial effects, 174; and syntactic content, 155-6; without comprehension, 175; of

523

Subject index word lists, 112, 130 reversible sentences, 468-9 Rey complex figure, 120 rhyme judgments, 289-90, 345; and articulatory suppression, 306; require phonological representations, 383 semantic anomalies, 468-9 semantic coding, and unattended speech, 60 semantic interpretation of sentences, 397-8 semantic similarity, 157-9 semantically reversible sentences, 429 sentence advantage effect, 171 sentence complexity, age effects, 250-2 sentence comprehension, see comprehension of sentences sentence processing, 400-2; see also comprehension of sentences; sentence recall sentence recall, 202-3, 210 (see also recallsentence span); effect of structure upon, 203-4; of lists vs. sentences, 206-7; of meaningless sentences, 198-9; recall of, 192-5; retention of, 195-7 sentence span, 67-8, 204 sentence verification, 459—61 serial position effects, 181 serial recall, see recall short-term memory (STM) (see also auditory-verbal STM; patients; phonological buffer, phonological processing; phonological storage; precategorical acoustic store; recall, immediate); and articulatory loop, 61-2; as back-up store, 180-1, 183-4; and comparative judgments, 180; differing groups of impairment, 339-41; dissociable STM systems, 181-2; and "extended present," 319-20; and language comprehension, 66-8, 337-84; and lexicality, 145-69; localization of impairments, 431; and mental arithmetic, 472; a model of components of, 309; multicomponent view of, 8; neuropsychological evidence, 56, 382-4; possible explanation of deficits in patients, 108-9; relation with LTM, 54-6; and sentence processing, 400-2, 428-46; sentence vs. list utilization, 208; as single functional entity, 1; STM patients, summaries of, 14-16, 349-62, 372-82; and verbal reasoning, 472

similarity effects (see also phonological similarity; semantic similarity); sensory similarity, 277; visual similarity and recall,

236-7 single-case design (see also dissociations; methodology; neuropsychological evidence); and generalization across patients, 1, 12 span tasks, 1-2 (see also sentence span; supraspan); age effects in reading span tasks, 252—5; in amnesia, 56; auditory vs. visual, 94—5; in a conduction aphasic, 453; digit span and articulatory suppression, 233-4; digit span, presentation rate effects in, 233—4; high and low frequency words, 82; and language comprehension, 180-1; modality effects, 148; nonword span, 456-7; and parsing, 443; pointing span, 298-9; and sentence comprehension, 411, 419-24; sentences, see sentences span; span for pseudohomophones, 314; STM demands of, 400-2; visual material, 218; word span, lexical-semantic and grammatical effects, 454 speech, internal, 62; and articulatory loop, 404—5; as explanation of memory deficits, 403; and rehearsal, 404-5 speech, spontaneous 14-16; see also articulatory loop speech, unattended, see unattended speech speech comprehension, role of working memory, 216 speech errors, 192 speech perception: and comprehension, 422; and lipreading, 269-71, 272(i; preserved and STM deficits, 9; and verbal STM, 80 speech production (see also articulatory loop); and STM, 2 speech rate, see articulation rate speech recognition, 276-7; see also models, TRACE spoonerisms, creation of, 345 stroop tasks, 345 structural lag hypothesis, 478; contrasted with processing limitation hypothesis, 502 subjacency, 490-1, 498 subvocal rehearsal, 403 supervisory attentional system, and working memory, 263 supraspan, 456

524

Subject index

syndromes, dangres of classification, 12 synonym judgment, 135-6 syntactic structure: content words and comprehension, 407-11; complexity, semantic anomalies as measure of, 434 temporal register, 324 temporal terms; and children's comprehension, 487-9; in good and poor readers, 491-7 Test for the Reception of Grammar, 294-5, 368, 381 thematic role assignment, 438; mechanism explained, 444; reversals, 468-9 time estimation, 319—28; and "extended present," 327-8; and prospective time judgment, 321; prospective vs. retrospective, 325-6; and STM-LTM distinction, 326-7 Token Test, 119, 208, 331, 333, 353, 356, 381 [see also comprehension); and articulatory suppression, 68; description of, 353-4; processes tapped by, 184; STM patients' performance on, 168—71; and syntactic impairments, 428; what it measures in STM, 168 trace decay; and primacy, 234; theories of, 54 TRACE model, see models, TRACE transcortical motor aphasia, 146; see also aphasia; conduction aphasia; Wernicke's aphasia Type I and II records, 140—1 unattended speech: and articulatory loop, 59—60; and sound intensity, 60; and vocal characteristics, 60 visual letter analysis, and immediate memory, 280 visual span, 14-16 visual word analysis, and phonological processing, 280 visuospatial scratch pad, 224-5 WAIS (Wechsler Adult Intelligence Scale), 2;

patterns in patients, 101-2 Wepman Auditory Discrimination Test, 354, 356 Wernicke's aphasia, 291 word frequency effects, 130-1 word length effect: and articulation rate, 226; and articulatory loop, 59; and articulatory suppression, 289; in children, 226-30, 239-41; explanation of, 230; in matching span tasks, 301-3; and phonological capacity, 282; and presentation mode, 227-8; and repetition, 126; visual vs. anditory presentation, 228 word lists: advantage for speaking aloud in recall, 311; and articulatory suppression, 230-5, 289; and presentation rate, 230-5; serial position effects, 230-5 word order effects: in complex sentences, 199—200; and sentence comprehension, 176-7 working memory, 4, 8, 54-70 (see also articulatory loop; central executive); age differences in, 247—67; application to patients, 288; articulatory loop, see articulatory loop; articulatory rehearsal, 77 [see also rehearsal); central executive, see central executive; and children, applicability to, 226-7; childrens' visual memory, 235-9; complexity, interaction with age, 248ff; developmental fractionation of, 221-45; division of attention in, 260—262; explanation of model, 224-6; and garden path effects, see garden path sentences; and language processing, role in, 481-2; and lexical influences, 160-4; mode effects in children, 229—30; and phonemic similarity, 228-30, 234; and phonological coding, 77; phonological storage, 77, 81, 288; and sentence comprehension in children, 477-503; and speech comprehension, role in, 216; and supervisory attentional system, 263; visuospatial sketch (scratch) pad, 58-9, 224, 235, 238-9; and word length effect in children, 228-30, 234