An attention-rich scenario

Suppose you are at your desk composing an email, but at the moment you are off in some imaginary mental space, enjoying a pleasant daydream. A colleague pokes her head in your door and gently knocks on the doorframe. She says, “Hey—do you have a second?”

How, if at all, will you heed that question? Will you even notice it? Though the request seems reasonable under these circumstances, is it always good advice to “pay attention”? Are there times when you are better off not paying attention—whatever that means? What if you have trouble “paying attention”? Can anything be done? And once you are “paying attention,” what does attention actually do, and how does that happen?

You say, “Sure—come in.” Your colleague sits down and begins to talk. You listen, but you also try to read what you have written in the email, which you are again thinking about. You start to type the next sentence in the message. Can you succeed at this multitasking?

Your colleague asks again, “Really—can you just pay attention?” You say, “Oh—OK. What’s this about?” Your colleague says, “Well, I know you don’t like math, but you’re the only person around and I need some help with this calculation. Here—take a look.”

What do we know about attention and how it works during this scenario?

Attention is separate

Attention is a set of mental functions separate from the rest of information processing, executed by neural systems separate from the regions of the brain that process the information to which attention is being allocated or denied. That is, specialized attentional systems exist separately from sensory, memory, language, and motor information-processing systems. The attentional systems modulate the operation of the information-processing systems.

What sort of evidence might support such a claim? Kastner, Pinsk, De Weerd, et al. (1999) as well as Corbetta and colleagues (e.g., Corbetta, Kincade, Ollinger, et al., 2000) used functional magnetic resonance imaging (fMRI) to show that different brain regions were active in response to a spatial cue telling the participant where to look for an upcoming stimulus than were subsequently active when the stimulus actually appeared and was being perceived. The latter perceptual regions were activated more strongly if the cue had correctly signaled where the stimulus would appear and hence the upcoming location had become the focus of spatial attention. Activation of the regions that processed the cue—the “attentional regions”—predicted both the activation of the perceptual regions and the speed and accuracy of the behavioral responses in the task. Thus, different brain regions appeared to be generating attentional information based on the cue, sending out control signals that then modulated the operation of the perceptual regions actually responsible for processing the stimulus. An excellent example of this type of research comes from Hopfinger (2000), whose results are shown in Figure 4-1.

Attention is complex

Second, Posner and Petersen argued that the brain regions devoted to implementing the functions of attention are themselves hierarchically organized systems distributed across cortical and subcortical structures. Considerable evidence supports this claim, some of which we will describe.

Attentional functions

What are the functions of attention? There has been much discussion of this question. For present purposes we follow the lead of Posner and Petersen (1990), distinguishing three major functions and attributing each to its own system of neural structures. System 1 is responsible for arousal, alerting, and vigilance. System 2 is responsible for orienting toward and selecting sources of information, both from the external perceivable world and from memory. System 3 is responsible for executive control and supervision of intentional, goal-directed task performances, and includes working memory.

The three manipulations

Box 4-1 summarizes the manipulations used in this study. We begin with the manipulation relevant to System 3—Executive Control and Supervision, because it is fundamental to understanding the logic of the ANT. Participants needed to respond to the direction of a target arrow, but the target arrow did not appear by itself. It occurred in the middle of a row of five arrows, two on either side, that were designated as irrelevant—the decision about direction was to be made on the basis of the middle arrow only. Thus, this task was a version of Eriksen and Hoffman’s (1972) “Flanker Task.” It is well established that if flanking stimuli are close enough in space to the target, they are difficult to ignore and they will get processed to some degree if attention is not focused on them very carefully. This is more likely to happen if the target stimulus is simple and easy to process and interpret (Lavie, 1995; 2006; see also Huang-Pollack, Nigg, & Carr, 2002), which of course an arrow certainly is, and of course the flanking arrows are also simple and easy to process.

These conditions create a need for executive control. When the flanking arrows point in the same direction as the target arrow (the “congruent” condition), then the participant need not succeed in focusing attention completely on the middle arrow. The directional response will be correct regardless. But if the flankiing arrows point in the opposite direction (the “incongruent” condition), then the participant must work hard enough and long enough to ultimately focus completely on the middle arrow before making his or her decision about direction.

Fan and colleagues obtained a measure of the burden that the flanking arrows placed on Executive Control by subtracting response times and accuracies in the easier congruent condition from response times and accuracies in the conflicted and therefore more demanding incongruent condition. A map of the brain regions involved in dealing with this burden was obtained by comparing blood oxygen levels in the congruent condition to those in the incongruent condition.

System 1—Arousal, Alerting, and Vigilance was manipulated by providing or withholding a warning signal, consisting of an asterisk that alerted the participant to prepare for the target display. The no-warning condition served as a baseline.

The manipulation of System 2—Orienting and Selecting added spatial information to alerting. On warning-present trials, the warning asterisk appeared in one of two locations: either in the middle of the screen, superimposed over the fixation cross that was always present (providing a warning that alerted the participant to get ready for an arrow, but no additional spatial information about where the target might occur), or in the location at which the target arrow was about to appear (providing a valid spatial cue to aid in orienting attention toward the correct location, in addition to providing an alerting cue). Thus, the manipulation of System 2—Orienting and Selecting was whether the cue that provided an alerting signal also provided spatial information that could be used to direct attention in advance of the target’s actual appearance.

As with Executive Control, substractions of the appropriate reaction times and accuracies provided measures of the impact of Alerting and of Orienting and Selecting on behavioral task performance, and comparisons between the appropriate blood flow images provided maps of the brain regions whose activity was associated with the behavioral effects.

Reaction time differences for the effects of alerting, selective attention (“orienting” in the table), and executive control (“conflict” in the table), as well as the correlations among these differences, are shown in Table 4-1. Each manipulation produced an impact. Performance was 60 ms faster (about 7.5%) when alerted, 31 ms faster (about 3.8%) when given orienting information, and 102 ms slower (about 12.5%) when dealing with the conflict created by the incongruent flankers. None of the correlations among these difference-scores was significant, meaning that statistically, the effects were independent of one another. This is consistent with the hypothesis that they reflect the operation of independent processes, as suggested by Posner and Petersen’s (1990) view of the organization of the attentional systems.

Brain activation maps were consistent with this view. The alerting effect was associated with activation of the thalamus and anterior and posterior cortical regions. The orienting effect was associated with activation in parietal regions and the frontal eye fields. The executive control effect was associated with activation of the anterior cingulate, regions in left and right prefrontal cortex, and regions in the left and right fusiform gyrus in ventral posterior cortex. Conjunction analyses seeking regions active in two or all three of the maps found very little overlap. Thus, the fMRI results indicate that the attention-relevant manipulations of the Attention Network Task activated three separate neural networks, each associated with a different attentional system.

Neurotransmitter-defined pathways of arousal, alerting, and vigilance

Neurochemical studies have focused on pathways between locus coeruleus and cortex that rely on noradrenalin as the primary neurotransmitter, as well as pathways connecting basal forebrain, cortex, and thalamus that rely on acetylcholine and dopamine (Cools & Robbins, 2004; Parasuraman, Warm, & See, 1998; Ron & Robbins, 2003). It appears that pathways involving these neurotransmitter systems in anterior cingulate cortex, right prefrontal cortex, and thalamus are particularly important in modulating arousal and the ability to mobilize effort.

Arousal and the rhythms of performance

The neural substrates of alertness, arousal, and vigilance remain objects of intense investigation. Quite a bit is known at the cognitive/behavioral level. It is clear that readiness as measured through task performance rises and falls in predictable patterns, and this happens on at least two different time scales.

Circadian rhythm

First, readiness varies with the circadian sleep-wakefulness cycle. Work relating this cycle to the speed and accuracy of task performance has been summarized by Hasher, Zacks, and May (1999). Some people are “morning people” and others are “evening people” or even “night owls.” That is, the time of day at which people feel most energetic and prefer to engage in demanding mental or physical activity varies systematically from person to person. Such preferences are persistent and reliable enough to be measured psychometrically—for example, using Horne and Ostberg’s Morningness-Eveningness Questionnaire. Hasher and colleagues reported that people tested at their preferred or optimal time of day perform faster and more accurately in a wide range of tasks than if they are tested at earlier or later times, when they say they are likely to feel tired or would prefer less strenuous activity.

Time-of-day preferences vary strikingly with age (Table 4-2). Young adults tend toward eveningness, whereas older adults tend rather strongly toward morningness. Testing younger and older adults in the evening exaggerates differences due to cognitive aging that generally favor young adults—young adults are more likely to be at or near their optimum times whereas older adults are wearing down. Testing in the morning reverses relative energy and readiness levels, with younger adults being at a down time and older adults being at or near their optimum time of day. The result is a reduction or even elimination of the age difference favoring younger adults, with the amount of reduction depending on the task.

In general, a person’s task performance improves as his or her optimal time of day approaches and falls after that time is passed. This performance cycle is thought to happen because arousal and the ability to prepare for task performance—to mobilize effort—rises and falls in accord with this same circadian pattern.

Alerting

A very similar pattern can be seen on a much smaller time scale in manipulations of alerting such as used by Fan and colleagues (2005). For example, Posner and Boies (1971) conducted a same-different matching task in which a pair of letters was presented sequentially. The participant’s job was to indicate as rapidly and accurately as possible whether the two letters were the same or different. Posner and Boies found that providing a warning signal in advance of the first letter speeded performance of the judgment.

The amount of benefit, however, depended on how much in advance the warning signal occurred. The benefit increased over the first half a second or so, reaching a maximum with an interval of 400 to 600 ms between the onset of the warning signal and the onset of the first letter. This is typical of reaction-time tasks requiring relatively simple judgments. Beyond this optimum warning interval of about half a second, benefit declines over the next few seconds, suggesting that it is difficult to maintain maximum readiness for very long (Kornblum & Requin, 1984; Thomas, 1974).

Vigilance

Extending readiness for longer periods—minutes or hours—proves to be a major problem. Failure under such demands for “vigilance” causes much human error in real-world task performance, especially when events that actually call for action are rare, or when the performer is at a nonoptimal part of the sleep-wakefulness cycle (Parasuraman et al., 1998; Thomas, Sing, Belenky, et al., 2000). Under sleep deprivation, performance becomes highly variable. Some stimuli elicit a normal response, some a very slow response, and some are missed altogether (Doran, Van Dongen, & Dinges, 2001).

We have seen so far that low levels of arousal and readiness produce poor performance and that this pattern occurs on two different time scales. What happens if the level of arousal rises above the optimal level, rather than falling below it? This can happen under conditions of extreme stimulation, whether externally induced (for example, by consuming caffeine or amphetamines or performing in a very noisy or otherwise annoying environment) or generated internally (as in anxiety, fear, or hypermotivation from an extreme desire to perform well). Under these conditions of high rather than low arousal, performance declines from the best levels achieved at the optimal level of arousal.

Yerkes-dodson law

The general shape of the resulting function is reflected in what has come to be called the Yerkes-Dodson Law, which relates level of arousal to the quality of performance. As shown in Figure 4-3, A, there is in general an optimum level of arousal and resultant readiness at which performance of any given task is maximized. Arousal levels that are lower or higher than this result in poorer performance. As shown in Figure 4-3, B, the optimum level of arousal varies by complexity of the task. In the simplest tasks, high levels of arousal are helpful. As task complexity increases, high levels of arousal become harmful. Progressively lower arousal levels are required for best performance as the number of component processes increases, or larger amounts of information must be attended to or maintained in working memory.

But for any task, arousal levels substantially lower than the optimum depress performance, as do levels substantially higher than the optimum, though perhaps for different reasons. Low arousal results in sluggishness or even drowsiness, leading to poor preparation and readiness to perform. Very high arousal appears to restrict attention and interfere with executive control and decision making (for discussions of the Yerkes-Dodson function, see Anderson, 1994; Kahneman, 1973; Yerkes & Dodson, 1908).

The costs and benefits of selective attention

Typing and walking are complicated activities extended in time. Much less complex perceptual and sensorimotor tasks show the impact of attention. One of the simplest and most widely used methods of isolating and measuring selective attention was applied in the Attentional Network Task of Fan and colleagues (2005). Stimuli could appear at two different locations in the visual environment—and even this limited variety taxed the perceiver’s capacity to some extent, as was evidenced by the benefit that arose when a cue directed attention to the upcoming stimulus location. Giving the perceiver the ability to select and monitor the relevant location from which information would be arriving proved to be helpful. Fan’s experiment did not measure what would have happened if the cue had been misleading—signaling and hence calling attention to a location that was not where the stimulus would ultimately appear. Other studies give the answer. Performance is hurt, with reaction time and/or accuracy falling below what would have been achieved with no cue at all. Both effects—the facilitation or benefit from a valid cue and the inhibition or cost from an invalid cue—are illustrated in Figure 4-4.

The functional architecture and functional anatomy of selective attention

Posner and Petersen (1990), using visual attention as an example, suggested that orienting and selecting a source of input is built around three component operations implemented in sequence. This is the “functional architecture” of selective attention, consisting of “disengage,” “move,” and “engage.” Each operation is accomplished by a different region of neural tissue (the “functional anatomy” of selective attention).

Posner and Petersen hypothesized that the “disengage” operation is supported by left and right posterior parietal cortex (and, as we will see, by neighboring posterior regions of superior temporal cortex). The disengage operation implements a decision to curtail the priority being given to input from the location on which attention is currently focused. The decision to give up a source of input as a priority in order to shift to another source might be made intentionally by executive control processes in working memory and transmitted as a top-down goal-directed instruction to parietal cortex, or it might arise from bottom-up signals driven by the sudden onset of a new, high-intensity, or potentially important stimulus in the environment. This difference between top-down or endogenous control and bottom-up and stimulus-driven exogenous control was not explicitly built into Posner and Petersen’s anatomical model, but it is necessary to the story. Exogenous control is also called automatic or reflexive attention capture and has been the topic of considerable research (Folk, Remington, & Johnston, 1992; Folk, Remington, & Wright, 1994; Schriej, Owens, & Theeuwes, 2008; Yantis, 1995).

In either case, implementing a shift of attention from its current focus to a new one begins with disengaging from the current focus. But after disengagement, where is attention going and how does it get there? The second operation is “move,” implemented via interactions between posterior parietal cortex and superior colliculus, both of which contain topographic maps of environmental space. Single shifts of attention appear to be the responsibility of more inferior regions of parietal cortex and neighboring temporal cortex interacting with superior colliculus, whereas a planned sequence of attentional foci, moving systematically from one location to another, appears to recruit additional parietal regions that are more superior (Corbetta, Shulman, Miezen, & Petersen, 1993).

Once the focus of attention arrives at its new location, it must “engage” with the new source and give priority to input coming from that location. “Engaging” is critically dependent on portions of the thalamus, in particular the pulvinar nucleus.

Since 1990, the functional anatomy of the “posterior attention system,” as Posner and Petersen called this network of regions responsible for selective attention, has grown more elaborate. Figure 4-5, borrowed from a widely used textbook in cognitive neuroscience (Gazzaniga, Ivry, & Mangun, 2009), assigns a larger array of attention-relevant functions to posterior anatomical regions in accord with this elaboration.

Auditory selective attention

Effects much like those observed in vision are obtained in studies of audition. A cue to the location from which sounds or speech will come facilitates detection and identification. A misleading cue that causes perceivers to orient attention to the wrong location harms detection and identification (Mondor & Zatorre, 1995).

There is a tight relationship between visual and auditory attention. Studies by Driver and Spence (e.g., 1994, 1998) show that visual cues attract auditory attention and vice versa, though these cross-modality cues may not be as powerful as within-modality cues. Cross-modal costs and benefits suggest a partially overlapping hierarchical organization, such that each sensory modality has to some degree its own dedicated piece of the attentional system, but these dedicated components in turn feed into and are modulated by a more general system that serves multiple modalities.

Cross-modal attention and speech

Cross-modal interactions are particularly important in speech perception. Visual information from lip configurations impacts the identification of spoken syllables and words—congruent information in which the lip configuration corresponds to the sound that is heard facilitates identication, whereas incongruent lip configuration and auditory informaton impedes identification. This is called the “McGurk Effect” (McGurk & McDonald, 1976; see also, e.g., Jones & Callan, 2003; Rosenblum, Yakel, & Green, 2000). In a study that required selective repetition of one of two auditory messages—one presented from the left side of space and the other from the right—Driver and Spence (1994) found that a video of the to-be-repeated speaker helped more if it was presented on the same side as the message was coming from. Thus, being able to align spatial attention in the visual and auditory modalities facilitates integration of the information.

How does selection work? biasing activation among representations

The mechanics of how attention intervenes to modulate information processing have been debated for a long time. In 1995 Desimone and Duncan (see also Duncan, 2004) proposed that attention operates directly on pathways of information flow and representations of information, adding activation to a pathway or representation in order to give it priority over others. According to this hypothesis, pathways that are irrelevant to the goal being served are ignored by attention and therefore left to their own devices. They may or may not become activated, but if they do, they are unlikely to become more active than the ones biased by the boost supplied from attention. Some might, however, which creates a mechanism for exogenous control of attention—the unintended reflexive shifts of attention discussed earlier. These “hyperactivating” stimuli include newly appearing stimuli with sudden onsets (Klein, 2004; Yantis, 1995), familiar stimuli with a long history of importance (Cherry, 1953; Treisman, 1969; Carr & Bacharach, 1976), as well as stimuli with properties that make them very easily recognizable as relevant to a current task goal, such as a particular color or shape (Folk, Remington, & Johnson, 1992; Folk, Remington, & Wright, 1994; Schreij, Owens, & Theeuwes, 2008).

Deploying and directing attention during language processing and communicative interactions

One of the most important communicative uses of language is to enable one person to engage the attention of another, so that they can exchange information via conversation, share a perceptual experience, or plan and collaborate in a joint action. A number of features of language facilitate such interactions. It is almost as if language were designed to capture attention and entrain the cognitive processes of conversational partners.

Signals from the speaker call the listener’s attention to language

Of course speech is an auditory signal that can attract attention. However, once speech begins, internal features further moderate the listener’s attention. Hesitations during otherwise fluent speech serve to orient attention to the subsequent word. In a study involving ERPs, Collard, Corley, MacGregor, and Donaldson (2008) manipulated spoken sentences that ended with either a predictable or unpredictable word. In half of the cases, the sentence-final word was preceded by a hesitation, “er.” The ERP results suggested that the hesitations helped to orient attention to the preceding word, thus decreasing the novelty (P300) response when the sentence-final word was unpredictable. This example showcases the important role of orienting and selective attention in understanding spoken language.

The development of shared attention via looking and pointing

A foundation is laid early in infancy for communicative interaction to direct attention. From birth, mothers and fathers look at the faces of infants during many of their interactions (though by no means all). In these face-to-face interactions, adults will follow the gaze of the infant and respond to it vocally, asking simple (but possibly quite developmentally profound) questions like “Oooh—what do you see?” Later on, adults will say things like “Oh, look at that” and accompany the directive command with a shift of gaze and a point. By 12 to 18 months of age, infants will follow these gazes and points, orienting head, eyes, and presumably attention in their direction. Only slightly later, infants will themselves look, point, and vocalize in apparently intentional ways during social interactions, trying to direct the attention of their interactional partner. These changes during the first year and half of life have been researched as the development of shared attention and joint regard. Bruner (1975) and Bates (1976) argued that these developments are important precursors both to the attentional powers of language and to one of the fundamental uses of communication (for a review, see Evans & Carr, 1984).

Eye movements give a real-time view of language comprehension

Where the head is turned and where the eyes are looking continue to be important indicators of attention throughout life, and have been used increasingly as real-time indicators of language comprehension (Henderson & Ferreira, 2004; Tanenhaus, 2007). A great boon to this research was the invention of small and lightweight eye-movement monitoring equipment that can be worn like a hat while carrying out a task that might involve reaching for objects, picking them up and moving them around or using them in response to instructions, or walking from one place to another (Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995).

Imagine a task in which the participant, wearing an eye-movement monitor, is confronted with an array of four or five objects laid out on the floor. A verbal instruction is played through earphones, such as “Put the apple that’s next to the knife on the plate” and the participant’s job is to carry out the instruction. Results from such studies indicate that gaze is often fixated on the target object, in this case the apple, within half a second or less of the onset of the object’s name in the instruction. This means that in many cases, attention is already moving toward the location of the target object before the target’s name has even been completely heard. Of course, this only happens when the target’s name is unambiguous in the task environment. Suppose that there are two apples in the array, so that the participant doesn’t know for sure which one is the target until hearing the word “knife.” Under circumstances like these, possible targets are being narrowed down. Gaze will move back and forth between the two apples until “knife” is heard, at which point the eyes will fix on the appropriate apple and reaching for it will be initiated. The close time-locking of the direction of attention as measured by eye movements provides a powerful and versatile onto the time course of language comprehension.

The impact of descriptions measured with memory

More generally, it is quite clear that descriptions of the visual environment serve to direct the listener’s attention. There are multiple consequences—establishment of joint regard and correct following of instructions among them. A lasting consequence involves the memories that the hearer of a description carries away from the experience. Bacharach, Carr, and Mehner (1976) showed fifth-grade children simple line drawings, each containing two objects interacting—a bee flying near a flower, a boy holding up his bicycle, and so forth. Before each picture, an even simpler description was provided: “This is a picture of a bee,” or “This is a picture of a bicycle.” After the list was completed, a memory test was administered, in which pictures of single objects were shown one at a time. The child’s job was to say for each object whether it had been in the objects seen in the pictures. A baseline condition with no descriptions established that all the objects were equally likely to be remembered. With a description before each picture, however, objects that had been named in the descriptions were remembered more often than baseline, whereas objects that had not been named were remembered less often.

The functional architecture and functional anatomy of working memory

The basics of the functional architecture of working memory can be seen in Figure 4-6. The pieces of mental equipment that comprise this multicomponent system are the executive controller plus three “buffers” or short-term-storage devices. One of these buffers is the “phonological loop,” which maintains a small amount of verbal information for rehearsal (for example, three to nine unrelated syllables, possibly more if the sequence consists of related words that can be chunked into groups). The exact capacity is debated (e.g., Carr, 1979; Cowan, 2000; Jonides, Lewis, Nee, et al., 2008; Miller, 1956), but there is wide agreement that rehearsal is needed to keep the information active and readily available. Once attention is turned to other activities, stopping rehearsal, then the information being rehearsed either fades or is replaced by new material to which attention has been turned.

A second buffer is the “visuospatial sketchpad,” which fulfills an analogous function for visual information—storing images rather than sounds, syllables, or words. An analogous debate has been waged over the capacity of the visual storage buffer, with perhaps slightly smaller estimates of capacity (Alvarez & Cavanagh, 2004; Awh, Barton, & Vogel, 2007; Jonides, Lewis, Nee, et al., 2008; Vogel & Machizawa, 2004; Xu & Chun, 2006).

The third storage subsystem, which is the main focus of the caption in Panel A, is the “episodic buffer.” The episodic buffer is a late addition to the theory of working memory (although it bears kinship to the concept of “long term working memory” described by Ericsson & Kintsch, 1995).

What is the functional anatomy or brain circuitry of working memory? Pioneering work on verbal versus spatial working memory by Smith and Jonides (1997) showed that verbal working memory was heavily left lateralized, as one might expect from the neuropsychological data on brain injury (Banich, 2004, Chapters 10 and 11). In this important study, a task requiring short-term memory for letters activated a large region of tissue in left prefrontal cortex and a smaller region in left parietal cortex. Subsequent neuroimaging research using fMRI rather than PET has been quite consistent with Smith and Jonides’ early findings. In contrast to verbal working memory, visuospatial working memory is heavily right-lateralized, also as might be expected from the neuropsychology of brain injury (Banich, 2004, Chapters 7 and 10). A task requiring short-term memory for the spatial locations of a pattern of dots activated a small region of tissue in right prefrontal cortex and a larger region of tissue in right parietal cortex. One might speculate from this comparison that for dealing with verbal materials the “thinking” part (attributed to frontal cortex) is more demanding than the “storage of data” part (attributed to parietal cortex), whereas the reverse might be true when dealing with spatial materials. Storage of spatial data appears to take up lots of brain space, compared to storage of verbal data. This is only a speculation about human beings, but it does hold true for computers and smart-phones. Visuospatial materials take up a lot of storage room compared to text or speech.

It has probably not escaped you that Baddeley’s diagrams of functional architecture are backward with respect to the neuroanatomy—they place the phonological loop on the right and the visuospatial sketchpad on the left. Perhaps this is no more than left-right confusion, a common affliction, but it does allow us to make an important point about inspecting diagrams of functional architecture versus diagrams of functional anatomy. Functional architecture must capture the abstract information-processing activities and the patterns of information flow and process-to-process communication, but only at the computational level. It need not directly reflect the underlying brain structures that actually perform the computations or their locations relative to one another in brain geography. In contrast, a diagram of functional anatomy should aim to be true to brain geography.

It is clear that crucial functional components of working memory reside in prefrontal cortex. The amount of prefrontal cortex in the brain varies substantially from species to species, with humans up toward the top of the distribution. Given the role of prefrontal cortex in executive control and working memory, such differences lead almost inexorably to hypotheses about the existence and origins of species differences in planfulness and in learning of complex tasks. This in turn raises questions about the role of executive control and working memory in language.

Working memory and language

The importance of working memory in language has been a matter of debate. Most theories of language and reading comprehension give working memory a seat at the table, though what it is supposed to do can vary from theory to theory (Caplan & Waters, 1995; Daneman & Merikle, 1996; Snowling, Hume, Kintsch & Rawson, 2008; Perfetti, 1985). Much the same holds for language production (Eberhard, Cutting, & Bock, 2005; Bock & Cutting, 1992). Gruber and Goschke (2004) argue in favor of a domain-specific and heavily left-lateralized portion of the working memory system devoted specifically to language. Baddeley (2000) reviews a variety of evidence regarding the role of working memory in language learning, acquiring a second language, and how working memory might be compromised in language disorders such as specific language impairment.

What is currently agreed about the functional anatomy of the phonological loop—the specifically verbal storage and processing component of the working memory system—is shown in Figure 4-7.

Pursuing the question of whether the phonological loop is involved in any aspect of speech production, Acheson, Hamidi, Binder, and Postle (2011) have established a direct link between verbal working memory and the phonological/articulatory demands on one act of speech production—reading aloud. Using transcranial magnetic stimulation (TMS) to temporarily scramble neural signals from highly localized subregions of the cortex. They attempted to disrupt performance in three different tasks. One of these tasks—paced reading, in which participants had to read lists of phonologically similar pseudowords aloud—was chosen because it ought to draw on the specifically phonological and articulatory resources of working memory. Acheson and colleagues predicted from previous work on the functional anatomy of working memory that TMS administered to the posterior superior temporal gyrus (a region involved in a variety of reading functions and implicated in verbal working memory and also in the phonological loop) ought to harm paced reading. As can be seen in Figure 4-8, this is exactly what they observed.

Conflict monitoring and inhibitory regulation

In 1908, Walter Pillsbury—a black sheep who wandered off from the Pillsbury Flour family business to become a professor—contrasted two means by which attention might work. One involved facilitation of information needed to pursue a goal. The other involved inhibition—that is, attention might achieve much the same ends by inhibiting what was not desired. If so, then what was desired would be the last representations left standing (so to speak), and they would take control by default.

Pillsbury allowed that both might be at work, and this is the position taken by most current theories of attention. Sometimes inhibitory processes need not be engaged, but when conflict and confusion arise, or when the interpretation of the situation changes so that “old news” needs to be discarded in favor of new goals or new information, then inhibition comes into play. The engagement of inhibitory processes is thought to be triggered by conflict-monitoring operations that run in the background of all task performances (see Botvinick, Cohen, & Carter, 2004). Detection of conflict or ambiguity alerts the executive and supervisory control system, which slows performance so that more care can be taken, updates goals to be pursued, deletes now-irrelevant information so as to reduce “cognitive clutter” (Hasher, Zacks, & May, 1999), and allows the system to resolve the conflict or deal with the ambiguity.

A compelling and much-studied example of dealing with conflict comes from the Stroop Color-Naming Task (Cohen, Aston-Jones, & Gilzenrat, 2004; MacLeod, 1991; Stroop, 1935). This task presents words printed in different colors. One might see the word “blue” printed in blue, the word “red” printed in blue, the word “chair” printed in red or in blue, or one might simply see a patch of color. In all cases, the job is to name the color of the stimulus—not to read it if it is a word.

Relative to naming the color of a patch of color, which is the simplest color-naming task imaginable, any stimulus that includes a word slows performance, at least for skilled readers. It is as if just having a word available to perception creates dual-task interfence in which it is difficult to ignore the word or to try to perform its automatically associated task of reading (Brown, Gore, & Carr, 2002).

What if the stimulus is a word that spells out the name of the color that must be produced (“red” printed in red), or alternatively, spelling out the name of a different color (e.g., “red” printed in blue)? These contrasting conditions produce the classic “Stroop Effect.” Naming times are much slower when the word spells a different color-name than the one that must be produced, and errors are more frequent. Most errors arise from reading the word rather than naming the color in which the word is printed.

In neuroimaging studies of the Stroop Effect, an important locus of the neural activity elicited by conflict, ambiguity, and error is the anterior cingulate cortex (ACC), which consists of two large gyri running side-by-side from front to back along the bottom of prefrontal cortex in the middle of the brain. ACC communicates widely with other prefrontal regions, with motor cortex, and with the motor control and arousal/alerting pathways involving the basal ganglia.

Work on communications patterns among different brain regions—called “effective connectivity” or “functional connectivity”—shows that ACC participates in a network with other prefrontal regions and portions of parietal cortex to process conflict and also to deal with surprising stimuli, as when an unexpected or rare stimulus occurs that requires a response (Wang, Liu, Guise, et al., 2009). It appears that surprising stimuli are recognized in parietal cortex, especially the intraparietal sulcus, which signals ACC and dorsolateral prefrontal cortex (DLPFC) with the news that something unexpected requires scrutiny and possibly action. Recognizing and adjusting for conflict works in the opposite direction—ACC and DLPFC serve as sentinels and interact to determine what processing adjustments need to be made in executive and supervisory control. This prefrontal collaboration includes signaling parietal cortex about changes to be made in selective attention. Thus, parietal cortex initiates network activity in the processing of surprise stimuli, whereas ACC and DLPFC interact with each other to resolve conflict by changing the deployment of attention during task performance.

As might be expected from earlier discussion of partial modality-specificity and partial modality-overlap between visual and auditory attention, monitoring and adjusting for conflict also depends to some degree of the sensory modality in which stimuli are being processed. Roberts and Hall (2008) compared a standard visual Stroop color-naming task, as described above, to an auditory Stroop-like task with similar demands for management of conflict. The auditory Stroop task presented participants with the words “high,” “low,” or “day”—the neutral equivalent of a patch of color. These words were spoken in a high-pitched or a low-pitched voice. The participant’s job was to indicate the pitch of the voice. Thus, the congruent condition was a high-pitched voice saying the word “high” or a low-pitched voice saying the word “low,” whereas the incongruent condition eliciting conflict was the high-pitched voice saying “low” or the low-pitched voice saying “high.” Imaging data from fMRI conducted during task performance showed shared regions of activation corresponding to the basic conflict monitoring and adjustment already described, plus additional modality-specific regions appearing only in the visual Stroop task or the auditory Stroop task.