The Cognitive Neuroscience of Memory

Encoding and Retrieval

Amanda Parker, Timothy J. Bussey, Edward L. Wilding

• Publisher: Psychology Press
• Year: 2005
• Print ISBN: 9781841692463
• eBook ISBN: 9781135430726
• Topic: Psicologia
• Subtopic: Psicologia cognitiva e cognizione

Recent advances in techniques available to memory researchers have led to a rapid expansion in the field of cognitive neuroscience of memory. This book provides accessible coverage of four key areas of recent advance, including research on functional imaging, electrophysiological and lesion studies, and developments from the computational modelling approach. The first section reviews functional imaging studies in humans, with particular emphasis on how imaging methods have clarified the cortical areas involved in memory formation and retrieval. The second section describes electrophysiological and lesion research in monkeys, where lesion and disconnection studies are rapidly adding to our knowledge of both information processing and modulatory aspects of memory formation. In the third section, electrophysiological and lesion studies in rats are reviewed allowing for a detailed study of the role of novelty and exploration in memory formation. The final section reviews current research in computational modelling which has allowed the development of new theoretical and experimental approaches to the study of memory encoding and retrieval. This volume draws together the current developments in each field, allowing the synthesis of ideas and providing converging evidence from a range of sources. It will be a useful resource for both advanced undergraduate and postgraduate students of psychology, as well as researchers in the field and anyone with an interest in cognitive neuroscience.

PART ONE
Human imaging studies

CHAPTER ONE

Episodic memory retrieval: An (event-related) functional neuroimaging perspective

Michael D. Rugg and Richard N.A. Henson
University College London, UK

INTRODUCTION

In this chapter we focus on recent studies employing functional imaging methods to investigate human episodic memory retrieval. Episodic retrieval was one of the first aspects of memory to receive systematic study using neuroimaging methods, and has continued to be studied intensively. Many of these studies have been described in review articles published within the last few years (see Buckner & Koutstaal, 1998; Cabeza & Nyberg, 2000; Desgranges, Baron, & Eustache, 1998; Fletcher, Frith, & Rugg, 1997, for reviews of neuroimaging studies; and see Friedman & Johnson, 2000; Rugg, 1995; Rugg & Allan, 1999, for reviews of related electrophysiological work), as have some of the theoretical notions inspired by this research (Rugg & Wilding, 2000; Tulving, Kapur, Craik, Moscovitch, & Houle, 1994a; Wheeler, Stuss, & Tulving, 1997). It is not the goal of the present chapter to revisit the ground covered by these earlier reviews; instead, we concentrate on recent studies of retrieval that have employed “event-related” neuroimaging methods. We address three principal questions: (1) to what extent are the findings from event-related studies consistent with those obtained using older methodologies? (2) What do the findings tell us about the functional and neural bases of episodic retrieval? (3) What directions should be taken by future research employing these methods?

Episodic memory

For present purposes, episodic memory retrieval is defined as the cognitive operations necessary to support the explicit (conscious) retrieval of information about recently experienced events and the spatial and temporal contexts in which they occurred. The majority of neuroimaging studies of episodic memory retrieval have been conducted within the “verbal learning” tradition, wherein to-be-remembered items (“study” items) are lists of pre-experimentally familiar words. Most studies have employed memory tests that involve the presentation of cues that are in some way related to the studied items. One of the simplest and most popular tests is “yes/no” recognition memory, when entire items (“copy cues”) are presented, and the participants’ task is to judge whether each item was presented at study. Other tests employ less informative cues. For example, in word-stem cued recall the test items comprise the first three letters of a word (e.g. MOT—), and the task is to decide whether a word fitting the cue was presented at study.

Whatever the retrieval task that is employed, two important considerations arise. The first concerns the need to try to distinguish between “pre-” and “post-” retrieval processing (see Rugg & Wilding, 2000, for a detailed discussion of the different kinds of process that might be active during an episodic retrieval task; and see Burgess & Shallice, 1996, for a functional model of retrieval that embodies many of these processes). Preretrieval processing refers to those cognitive operations that support an attempt to use a cue to retrieve information from memory. Postretrieval processing, by contrast, involves cognitive operations that operate on the products of a retrieval attempt; these operations might include, for example, the maintenance in working memory of retrieved information and its evaluation with respect to current behavioural goals. Importantly, the notion of postretrieval processing is distinct from that of “retrieval success”. The latter term refers to the situation wherein a retrieval attempt leads to successful recovery of information about a relevant past episode. Whereas retrieval success can often be sufficient to engage postretrieval processes, it is unlikely to be necessary. Postretrieval processing will be engaged to some extent whenever the products of a retrieval attempt must be evaluated prior to a memory judgement, even if the judgement is ultimately negative (signalling a failure to retrieve). The distinction implied here between processes involved in monitoring the outcome of a retrieval attempt and those that operate on the products of successful retrieval appears to be important for the interpretation of some of the findings reviewed in the later section “Event-related studies of episodic retrieval”.

A second consideration when interpreting findings from functional imaging studies of memory retrieval arises from the argument that few, if any, retrieval tasks are “process pure” (Jacoby & Kelley, 1992). One well-known example of process “impurity” is the influence of explicit memory on indirect memory tests intended to assess implicit memory. But as pointed out by Jacoby and his associates (e.g. Jacoby & Kelley, 1992), performance on direct memory tests used to assess explicit memory might also be influenced by more than one kind of memory. For example, correct performance on word-stem cued recall can reflect both episodic retrieval and implicit memory (i.e. the same processes that support priming effects on word-stems; Jacoby, Toth, & Yonelinas, 1993). Clearly, if a retrieval task engages multiple kinds of memory, interpretation of the resulting imaging data will be far from straightforward. It is therefore unfortunate that the most common retrieval task in neuroimaging studies of episodic memory—recognition memory—is a task on which performance is almost certainly determined by the contribution of multiple processes (e.g. Yonelinas, 1994). It is possible to design recognition-like retrieval tasks that allow the contributions of episodic and non-episodic memory to be fractionated. For example, memory judgements based on episodic retrieval can be identified by requiring judgements of source rather than simple recognition, or alternatively by requiring recognition judgements to be accompanied by introspective report. Such tasks have been employed in several electrophysiological studies of retrieval (e.g. Rugg, Schloerscheidt, & Mark, 1998b), but they have only recently been used in studies employing functional neuroimaging methods.

Neuroimaging: Methods and measures

A description of currently available methods for the non-invasive measurement of human brain activity can be found in Rugg (1999). Irrespective of the method employed, an important distinction is that between transient changes in neural activity that follow a specific event such as the presentation of a stimulus (item-related activity), and more sustained modulations of activity that accompany engagement in a specific task and are unaffected by the presentation of specific items (state-related activity). This distinction is important because it is likely that the two kinds of activity reflect different kinds of cognitive operation (Rugg & Wilding, 2000), and also because it is central to the current debate about the functional significance of many neuroimaging findings regarding episodic retrieval.

Until relatively recently, studies employing functional neuroimaging methods based on the detection of blood flow and oxygenation—the socalled “haemodynamic” methods of positron emission tomography (PET) and functional magnetic resonance imaging (fMRI)—were designed in such a way that item- and state-related brain activity were always confounded. Because of the constraints imposed by PET methodology, PET images of regional cerebral blood flow are integrated over an acquisition period (and a corresponding block of experimental trials) lasting some 40–60s, making it impossible to distinguish between item- and state-related activity. The same problem is encountered in so-called “blocked” fMRI designs, when contrasts are performed on data from two or more blocks of trials, each representing an experimental condition. In both of these cases, any differences between experimental conditions in patterns of cerebral activity represent an unknown mixture of item- and state-related effects. Whereas it is possible in principle to design blocked studies to fractionate these effects, in practice it is difficult, if not impossible, to demonstrate that the fractionation was successful. For example, in an effort to investigate the item-related neural correlates of successful retrieval of episodic information, researchers have compared mean brain activity during recognition memory judgements made on a blocks of predominantly old versus predominantly new items (Kapur, Craik, Brown, Houle, & Tulving, 1995; Nyberg et al., 1995; Rugg et al., 1998a; Rugg, Fletcher, Frith, Frackowiak, & Dolan, 1996, 1997). Although attempts were made to disguise the manipulation of the ratio of old to new items, between-block differences in brain activity cannot be attributed unequivocally to item-related effects. Differences in state-related activity might still have occurred, for example, if participants adopted different task strategies following a few consecutive presentations of items belonging to the same class.

Historically, methods capable of distinguishing item- and state-related effects unequivocally have been based on electrophysiological rather than haemodynamic measures, notably, scalp-recorded electrical activity (the electroencephalogram or EEG). Electrophysiological methods can be employed in cognitive studies to measure time-locked modulations of the EEG elicited by a particular class of experimental items (e.g. “new” as opposed to “old” items in a recognition memory task). The resulting waveforms, known as event-related potentials (ERPs), provide a measure of that component of item-related neural activity that can be detected at the scalp (see also Chapter 2). Advances in fMRI methodology during the late 1990s led to the development of “event-related” methods, which permit functional images to be obtained in a manner analogous to that employed to record ERPs (Dale & Buckner, 1997; Josephs, Turner, & Friston, 1997; Zarahn, Aguirre, & D’Esposito, 1997). As is the case with ERPs, event-related fMRI allows item-related effects to be identified unequivocally. Whereas fMRI has by far the better spatial resolution, the sluggishness of the haemodynamic response means that the temporal resolution of event-related fMRI signals is of the order of hundreds of milliseconds. This compares unfavourably with the millisecond-level resolution that can be attained with electrophysiological measures. Thus, the two methods provide complementary perspectives on event-related brain activity.

Event-related methods have advantages that go beyond the capacity merely to detect item-related activity. First, the methods make it possible to employ randomised experimental designs, whereby trials belonging to different experimental conditions are intermixed in an unpredictable sequence. With such designs, effects on item-related measures resulting from the adoption of condition-specific “sets” are eliminated. Furthermore, by comparing the item-related activity elicited in randomised versus blocked designs, set effects can be identified and characterised. For example, using ERPs, Johnson et al. (1997b) compared the item-related activity elicited by “true” and “related lure” items in a “false memory” paradigm (see the later section “Retrieval success”) when the two classes of item were randomly intermixed and when they were presented in separate test blocks. Differences in the ERPs elicited by the two kinds of item were found only for the blocked conditions, indicating that such differences depended on the adoption of different task sets (and, perhaps, on different patterns of state-related activity, although Johnson et al., 1997b, did not address this issue).

A second benefit of event-related methods, of particular importance for memory studies, is that they permit experimental trials to be allocated to different experimental conditions post hoc, on the basis of behavioural performance. Thus, it is possible to compare brain activity elicited by, say, “old” items in a recognition memory test according to whether the items were detected correctly or misclassified as new. The comparison of the patterns of brain activity elicited by items attracting different responses has been a cornerstone of ERP studies of memory retrieval for a considerable time (Rugg, 1995) and, as will become apparent, has already proven to be important in the case of event-related fMRI.

Despite the advantages of event-related designs in studies of memory retrieval, there remain circumstances when such designs are difficult to employ, and blocked procedures are preferable. This will be the case for example when the retrieval task does not involve the presentation of discrete retrieval cues, such as in free recall. More generally, the advantages of event-related over blocked designs will decline as the time-locking between external events and the cognitive operations of interest becomes weaker, and the intertrial variance in the timing of item-related activity correspondingly greater.

Finally, it is important to note that the employment of event-related designs does not by itself resolve the issue of how to identify and characterise state-related changes in brain activity. It is possible however to design both electrophysiological (Duzel et al., 1999) and fMRI studies (Chawla, Rees, & Friston, 1999; Donaldson, Petersen, Ollinger, & Buckner, 2001) in such a way that item- and state-related activity can be assessed concurrently; as will become clear, there are good reasons why such designs are preferable to those focusing exclusively on event-related activity.

Interpretation of event-related fMRI data

An important issue in the interpretation of event-related data relates to the nature of the contrasts employed to identify brain regions that are active in different experimental conditions. In our view, the claim that a given brain region is selectively activated by items belonging to a given experimental condition is justified only when the event-related responses elicited by those items differs significantly from the responses elicited by items from another experimental condition. In other words, the finding that items from one condition elicit responses that differ reliably from the interstimulus baseline, whereas items from another condition do not, provides insufficient grounds for concluding that the responses elicited by the two conditions are significantly different (requiring, as it does, an acceptance of the null hypothesis).

More generally, it is arguable that “raw” event-related responses—item-related signal changes relative to an interstimulus baseline—are difficult, if not impossible, to interpret in the context of studies of higher cognitive processing. This is because the responses reflect a mixture of “low-level” processes common to all tasks, task-specific processes common to all item-classes, and processes specific to the item-class eliciting the response. Unlike, say, the simple case of visual cortex responses to brief visual stimulation against a static background, we cannot be certain what cognitive processes are engaged during the baseline periods between events in typical memory tasks. In the period prior to the presentation of a new item, for example, the participant might still be engaged in evaluating the episodic information retrieved in response to the previous old item. To separate these different kinds of item-related activity, it is necessary to contrast directly the responses elicited by the same types of item in different tasks, and by different item-types within the same task. Therefore it is important that event-related fMRI studies are designed so that differential item-related activity can be detected with adequate sensitivity. It turns out that for the kinds of randomised designs favoured in experimental psychology, sensitivity to differential activity is an inverse function of stimulus onset asynchrony (SOA) (Josephs & Henson, 1999). For this reason, the more recent event-related fMRI studies of episodic retrieval have employed relatively short SOAs (about 2–4s). With such short intervals it is not possible to obtain the “raw” response elicited by each type of item relative to the prestimulus baseline. None the less, the form of these responses can be important in constraining the interpretation of differential effects (e.g. whether the effects reflect differences in the amplitude or the latency of the responses elicited by different item classes; Henson, Price, Rugg, Turner, & Friston, 2002). It has therefore become common for event-related studies to include so-called “fixation” or “null” trials along with other trial types (Buckner et al., 1998a), effectively producing a stochastic distribution of SOAs (Friston, Zarahn, Josephs, Henson, & Dale, 1999), which allows item-related activity relative to baseline to be estimated.

BLOCKED STUDIES OF EPISODIC RETRIEVAL

In this section, we briefly review what we see as the more important of the findings to have emerged from PET and blocked fMRI studies of episodic retrieval. In these studies, several regions have been consistently reported to be active when participants engage in an episodic retrieval task relative to a non-episodic control task. Chief among these regions are dorsolateral and anterior prefrontal cortex, and medial and lateral parietal cortex. It is noteworthy that, on the basis of “classical” findings from human and animal neuropsychology, most of these regions would not be regarded as playing a central role in episodic memory.

Prefrontal cortex

Activation of prefrontal cortex has been reported in the majority of functional neuroimaging studies of episodic retrieval (see Desgranges et al., 1998; Fletcher & Henson, 2001; Nolde, Johnson, & Raye, 1998b, for detailed reviews of these findings). In light of reports from the neuropsychological literature of relatively subtle memory impairments following f...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• List of contributors
• Series preface
• Introduction
• Part One: Human imaging studies
• Part Two: Non-human primate studies
• Part Three: Rat studies
• Part Four: Computer models of memory encoding and retrieval