Introduction

For humans, the most prized of all our mental faculties is likely our capability for directing our thoughts and actions in accordance with internal goals, and for flexibly readjusting these goals when necessary. The term “executive function” is typically used to describe these regulatory and goal-directed components of cognition. However, the term itself, with its implicit reference to a hidden homunculus, indicates the difficulty of understanding how such executive functions might arise in the brain. Thus, a fundamental challenge for cognitive neuroscience is to determine the underlying representations, computations, and neural specializations that enable cognition and action to appear coordinated, purposeful, and self-regulatory. Cognitive neuroscience investigations of executive functions have accelerated in an amazingly rapid fashion within the last decade, due in large part to the development of sophisticated neuroimaging methods for noninvasively monitoring human brain activity. Indeed, a striking example of this accelerating growth is the observation that studies of executive processes are currently one of the largest components of the cognitive neuroimaging literature, yet in the previous edition of this book (published in 2001) there was no chapter devoted to executive function as a distinct area.

Although neuroimaging contributions to executive control research have only been very recent, there has been a long tradition of study in this area within neuropsychology, behavioral neuroscience, and cognitive science. Neuropsychological research has focused attention on the critical role of the frontal lobes in self-regulatory behavior, starting from the earliest case studies of individuals such as Phineas Gage (e.g., Harlow, 1848; Macmillan, 2000), to the more focused and systematic reports of Luria (1966). Starting with the seminal work of Milner (1963), experimental neuropsychology investigations have been instrumental in developing task paradigms that provide sensitive and objective probes for assessing executive function and its relationship to underlying brain areas. For example, studies with the Wisconsin Card Sort Task (WCST) indicated that frontal patients were more likely than other brain-damaged groups to show “perseverative” type behavior, staying with an old card-sorting rule even after repeated feedback that the rule was no longer correct. In addition, the experimental neuropsychology literature has contributed a number of other tasks for studying executive functions developed in the context of work with brain-damaged populations. These include tests of verbal fluency, planning (the Tower of London), attention switching (Trail Making), inhibition (go-no go), and sustained/selective attention (continuous performance test).

In parallel with the human neuropsychological research, behavioral neuroscience studies have provided strong confirmation and elaboration of the importance of the prefrontal cortex (PFC) in executive control. Early studies demonstrated fairly convincingly that frontal lesions markedly increased distractability in delayed-response tasks (which require holding a reward-eliciting behavior in check for a short period of time following presentation of an environmental cue) by impairing the ability to maintain goal-relevant information or context in mind even when such information is no longer perceptually available (e.g., Jacobsen, 1935). Later electrophysiological studies in awake behaving monkeys showed that during such delayed-response tasks, prefrontal neurons showed stimulus-specific increases in activity that were sustained throughout the delay interval and correlated with correct performance (Fuster, 1989; Goldman-Rakic, 1987). These studies have been of critical importance for establishing the role of PFC in active maintenance functions, and in specifying the representational and dynamic properties of this brain region. More recent work has emphasized how these properties of PFC might subserve cognitive control functions by representing rule-like information, enabling the learning of temporal associations between stimuli and response, filtering perceptual information in accordance with task goals, and exerting a top-down bias on action and perceptual systems (E. K. Miller & Cohen, 2001).

These types of neuropsychological and neurophysiological findings influenced the development of general theoretical models of cognition, starting with early work on planning (G. A. Miller et al., 1960) and progressing to more recent computational cognitive architectures (Anderson, 1983; Meyer & Kieras, 1997; Newell, 1990). In these models, a central role was given to a working memory structure where goal-related representations could be temporarily activated to bias sequences of thought and action. Ideas regarding the top-down biasing effects of goal information on action selection were more firmly crystallized in the supervisory attentional system (SAS) theory of Norman and Shallice (1986), which specified a control mechanism selectively invoked when automatic stimulus-response sequences are inappropriate or inadequate for completing actions in a goal-compatible fashion.

The SAS theory explicitly suggested that this control structure was located in the frontal lobes, and further specified the types of tasks situations most dependent upon frontal integrity: (a) when the task requires that a habitual response be suppressed; (b) when the task is novel and unpracticed; (c) when the task is dangerous or technically difficult; (d) when planning is required; and (e) when errors need to be monitored or corrected. Consequently, the SAS model has been very influential in guiding modern cognitive neuroscience investigations regarding the decomposition of different executive functions, as described below. Similarly, the SAS account has influenced more neuroscience-based theoretical models of control. For example, the models of Cohen and colleagues show how active maintenance of goal-like representations can occur in PFC via local recurrent connectivity, and how simple feedback connections allow these PFC representations to serve as a top-down bias on local competitive interactions within direct task pathways, thus achieving a simple form of control in cognitive tasks such as the Stroop (Cohen et al., 1996; Cohen et al., 1990).

Notions of executive function have also strongly influenced more domain-specific cognitive theories, in particular theories of memory. The classic information-processing theory of Atkinson and Shiffrin (1971) suggested that control processes play a significant role in how information gets into and out of the short-term and long-term stores, via modulation of encoding, rehearsal, decision, and retrieval strategies. This basic idea has been elaborated in more recent theories postulating that PFC may serve as this control interface, “working with memory” by strategically filtering retrieved content (Moscovitch, 1992; Shimamura, 1995, 2000).

Baddeley, in his influential model of working memory (WM), reinterpreted the Atkinson and Shiffrin view of short-term memory by suggesting that (a) short-term storage occurs in qualitatively distinct buffers for verbal (the phonological loop) and visuospatial (the visuospatial scratch pad) information, and (b) that these buffers serve as a “mental blackboard” for complex cognition (Baddeley, 1986). Critically, the Baddeley model cast a specific role for a “central executive” that appropriately utilized the stored information in the service of task-related processing. Baddeley's structurally based account provided a strong impetus for cognitive neuroscience-based research programs aimed at dissociating storage and control functions within WM in terms of their underlying neural substrates (D’Esposito et al., 1998; Smith & Jonides, 1999).

Nevertheless, a limitation of the Baddeley model, repeatedly acknowledged by Baddeley himself, is that the theory did not clearly specify what constituted an executive function within WM. Thus, it has not been clear whether the central executive referred to a single, monolithic mechanism or to a collection of distinct subprocesses subserved by multiple separate mechanisms. This major gap in theorizing regarding executive control has begun to be remedied (Baddeley, 1996). However, there is still no consensus as to exactly what are the functions that should be termed “executive,” and how many distinct functions there are (Miyake et al., 2000).

The lack of conceptual clarity regarding the components of executive control is especially noticeable when surveying the large and ever-growing neuroimaging literature in this domain. Thus, there is some ambiguity as to how to structure a review of the literature in a chapter such as this one. In the following section, we have decided to organize the literature into categories according to what seem to be natural groupings of studies and tasks, rather than arguing for strict conceptual divisions.