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 general-purpose 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).

Introduction

Immediate memory capacity for lists is usually estimated at around 5–6 words, whereas for sentences, Miller and Self ridge (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.

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).

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 rehearsal through concurrent articulation.

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.

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).

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.

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).

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.

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.

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).

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.

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.

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.

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.

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.

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.

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.

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.