Integrating Theories of Working Memory

Multiple theories of working memory are described in the chapters of this book and often these theories are viewed as being mutually incompatible, yet each is associated with a supporting body of empirical evidence. This chapter argues that many of these differences reflect different research questions, different levels of explanation, and differences in how participants perform their assigned tasks in different laboratories, rather than fundamental theoretical adversity. It describes a version of a multiple component working memory in which a range of specialized cognitive functions (or mental tools) act in concert, giving the impression, at a different level of explanation, of a unified cognitive system. The chapter argues that more rapid and more substantial scientific progress on the understanding of the concept of working memory would be achieved through identifying the levels of explanation explored within each theoretical framework, and attempting to integrate theoretical frameworks rather than perpetuating debate with no clear resolution in sight.

Manifold Visual Working Memory

Visual working memory allows us to maintain information in mind for use in ongoing cognition. Research on visual working memory often characterizes it within the context of its interaction with long-term memory (LTM). These embedded-processes models describe memory representations as existing in three potential states: inactivated LTM, including all representations stored in LTM; activated LTM, latent representations that can quickly be brought into an active state due to contextual priming or recency; and the focus of attention, an active but sharply limited state in which only a small number of items can be represented simultaneously. This chapter extends the embedded-processes framework of working memory. It proposes that working memory should be defined operationally based on neural activity. By defining working memory in this way, the important theoretical distinction between working memory and LTM is maintained, while still acknowledging that they operate together. It is additionally proposed that active working memory should be further subdivided into at least two subcomponent processes that index item-based storage and currently prioritized spatial locations. This fractionation of working memory is based on recent research that has found that the maintenance of information distinctly relies on item-based representations as well as prioritization of spatial locations. It is hoped that this updated framework of the definition of working memory within the embedded-processes model provides further traction for understanding how we maintain information in mind.

Remembering Over the Short and Long Term

This chapter reviews evidence from behavioural and cognitive neuroscience research that supports a unitary view of memory whereby working memory and long-term memory phenomena arise from representations and processes that are largely shared when remembering over the short or long term. Using ‘false working memories’ as a case study, it highlights several paradoxes that cannot be explained by a multisystem view of memory in which working memory and long-term memory are structurally distinct. Instead, it is posited that behavioural memory effects over the short and long term relating to semantic processing, modality/domain-specificity, dual-task interference, strategic processing, and so on arise from the differences in activational states and availability of different representational features (e.g. sensory/perceptual, associative, action-based) that vary in their time courses and activity, attentional priority, and susceptibility to interference. Cognitive neuroscience evidence primarily from brain imaging methodologies that support this view is reviewed.

Multicomponent Working Memory System with Distributed Executive Control

The working memory model with distributed executive control accounts for the interactions between working memory and multi-tasking performance. The working memory system supports planned actions by relying on two capacity-limited domain-general and two time-limited domain-specific modules. Domain-general modules are the episodic buffer and the executive module. The episodic buffer stores multimodal representations and uses attentional refreshment to counteract information loss and to consolidate information in episodic long-term memory. The executive module maintains domain-general information relevant for the current task. The phonological buffer and the visuospatial module are domain specific; the former uses inner speech to maintain and to rehearse phonological information, whereas the latter holds visual and spatial representations active by means of image revival. For its operation, working memory interacts with declarative and procedural long-term memory, gets input from sensory registers, and uses the motor system for output.

The Time-Based Resource-Sharing Model of Working Memory

The time-based resource-sharing model considers working memory as the workspace in which mental representations are built, maintained, and transformed for completing goal-oriented tasks. Its main component is made of an episodic buffer and a procedural system that form an executive loop in which processing and storage share domain-general attentional resources on a temporal basis. Because working memory representations decay with time when attention is diverted, the cognitive load of a given activity is the proportion of time during which it occupies attention and prevents it from counteracting this decay through attentional refreshing. Consequently, recall in working memory tasks is an inverse function of the cognitive load of concurrent processing. Besides this system, an independent domain-specific maintenance system exists for verbal, but not visuospatial, information. Within this framework, working memory development mainly results from increasing processing speed that affects both the duration of the distraction of attention by concurrent tasks and refreshing efficiency.

Individual Differences in Attention Control

This chapter outlines the executive attention theory of higher-order cognition, which argues that individual differences in the ability to maintain information in working memory and disengage from irrelevant information is inextricably linked to variation in the ability to deploy domain-free attentional resources in a goal-directed fashion. It also summarizes recent addendums to the theory, particularly regarding the relationship between attention control, working memory capacity, and fluid intelligence. Specifically, the chapter argues that working memory capacity and fluid intelligence measures require different allocations of the same attentional resources, a fact which accounts for their strong correlation. At various points, it addresses theoretical alternatives to the executive attention theory of working memory capacity and empirical complications of the study of attention control, including difficulties deriving coherent attention control latent factors.

Towards a Theory of Working Memory

Working memory provides a medium for building and manipulating new representations that control our thoughts and actions. To fulfil this function, a working memory system needs to meet six requirements: (1) it must have a mechanism for rapidly forming temporary bindings to combine elements into new structures; (2) it needs a focus of attention for selectively accessing individual elements for processing; (3) it must hold both declarative representations of what is the case, and procedural representations of how to act on the current situation; (4) it needs a process for rapid updating, including rapid removal of outdated contents. Moreover, contents of working memory (5) need to be shielded from interference from long-term memory, while (6) working memory should be able to use information in long-term memory when it is useful. This chapter summarizes evidence in support of these mechanisms and processes. It presents three computational models that each implement some of these mechanisms, and explains different subsets of empirical findings about working memory: the SOB-CS model accounts for behaviour in tests of immediate serial recall, including complex-span tasks. The interference model explains data from a common test of visual working memory, the continuous-reproduction task. The set-selection model explains how people learn memory sets and task sets, how these sets are retrieved from long-term memory, and how these mechanisms enable switching between memory sets and task sets.

A Multicomponent Model of Working Memory

The multicomponent model aims to provide a broad theoretical framework enabling both more detailed fractionation and analysis of its components, and a capacity for it be used fruitfully beyond the laboratory. In its current form it comprises four interacting components. Two of these are modality-specific memory storage systems, one verbal-acoustic, the phonological loop, and one visuospatial, the sketchpad. Information in both these stores can be temporarily maintained via focused attention termed ‘refreshing’, while the phonological loop can also maintain familiar verbalizable material by subvocal or overt rehearsal. Both subsystems are controlled by a third component, the central executive, a supervisory system with limited resources. The central executive is principally concerned with internally directed attentional control processes but also has a role in the attentional selection of perceptual information. Information from these three components is coordinated with information from perception and long-term memory through the fourth component, a multidimensional, multimodal episodic buffer. This component is capable of holding up to around four episodic chunks, and is a valuable but essentially passive storage system, controlled by the central executive and accessible to conscious awareness. The multicomponent model has been systematically developed using a number of experimental tools. These include, principally, similarity effects to identify the type of coding involved, concurrent task methods to assess the contributions of the various subsystems to complex tasks, and neuropsychological evidence, in particular from the study of single cases with very specific deficits. The model continues to evolve and has proved successful both in accounting for a broad range of data on memory and related cognitive areas and in its application to the understanding of a wide range of cognitive activities and populations.

The State of the Science of Working Memory

The last five decades have seen a dramatic expansion of empirical and theoretical diversity in research on the concept of working memory. That diversity is reflected across the other 13 chapters in this book, written by prominent international scientists at the leading edge of that research. Authors were asked to respond to set of common questions regarding their research methods, theoretical assumptions, and how they address evidence that is not consistent with those assumptions. Each chapter starts with a summary of those responses. Chapters 2–6 describe contrasting theoretical perspectives and the evidence associated with each, including empirical, behavioural studies and computational modelling. Chapters 7–9 cover individual differences in working memory, including effects of brain damage and of expertise. Chapters 10–13 explore neural correlates and neurobiological models. Finally, Chapter 14 offers a possible means to integrate the seemingly diverse views in the other chapters by considering different levels of explanation and different participant strategies for performing working memory tasks.

Domain-Specific Working Memory

The domain-specific approach to working memory assumes specialized working memory systems dedicated to maintaining different types of information (e.g. orthographic, phonological, semantic, visuospatial) which serve to support processing in that domain. These storage systems are assumed to be separate from long-term memory representations in each domain and also from attentional and cognitive control processes. This chapter provides an overview of support for this approach drawn mainly from neuropsychological case study and case series approaches, though it also integrates findings from behavioural and imaging studies of healthy individuals that were motivated by the neuropsychological findings or provide confirmation of those findings. The neuropsychological findings not only demonstrate dissociations between working memory in different domains but also provide a rich source of evidence to address the nature of forgetting in working memory, the interactions between working memory and long-term memory, and the role of aspects of working memory in language comprehension and production.