4 THE STUDENTS GUIDE TO COGNITIVE NEUROSCIENCE the heart rather than the brain.he believed that the brain acted as a coolant system:the higher the intellect.the larger the cooling system peeded.In the Roman age.Galen(circa AD 129-199)observed brain injury in gladiators and noted that nerves project to and from the brain.Nonetheless,he believed that mental experiences themselves resided in the ventricles of the brain.This idea went essentially unchallenged for well over 1 500 vears for example when Vesalius (1514-1564).the father of modern anatomy of dissected brains the ventricles were dran dewspaes eas the cortex was drawn crudely and schematically (see Figure 1.2).Others followed in this tradition.often drawing the surface of the brain like the intestines.This situation probably reflected a lack of interest in the cortex rather than a lack of in It is of Gall and Spurzhein ures of the ern eyes FIGURE1.2:Drawings of the brain from Vesalius(1543)(top).de Viessens(1685)(bottom left)and Gall and Spurzheim (1810)(bottom right).Note how the earlier two drawings emphasized the ventricles and/or misrepresented the cortical surface
4 THE STUDENT’S GUIDE TO COGNITIVE NEUROSCIENCE the heart rather than the brain. He believed that the brain acted as a coolant system: the higher the intellect, the larger the cooling system needed. In the Roman age, Galen (circa ad 129–199) observed brain injury in gladiators and noted that nerves project to and from the brain. Nonetheless, he believed that mental experiences themselves resided in the ventricles of the brain. This idea went essentially unchallenged for well over 1,500 years. For example, when Vesalius (1514–1564), the father of modern anatomy, published his plates of dissected brains, the ventricles were drawn in exacting detail, whereas the cortex was drawn crudely and schematically (see Figure 1.2). Others followed in this tradition, often drawing the surface of the brain like the intestines. This situation probably reflected a lack of interest in the cortex rather than a lack of penmanship. It is not until one looks at the drawings of Gall and Spurzheim (1810) that the features of the brain become recognizable to modern eyes. FIGURE 1.2: Drawings of the brain from Vesalius (1543) (top), de Viessens (1685) (bottom left) and Gall and Spurzheim (1810) (bottom right). Note how the earlier two drawings emphasized the ventricles and/or misrepresented the cortical surface
INTRODUCING COGNITIVE NEUROSCIENCE 5 received a bad histori pres KEY TERMS ng,because o. erent regi of the br Phrenology The failed ea that perform an second,that the size of these regions produces ons of the correlates with individual differences in cognition and personality. Taking these two ideas in turn,the notion of functional specialization within the shape. brain has effectively endured into modern cognitive neuroscience, having seen off a number of challenges over the years(Flourens,1824 1929).The observations of Penfield and co-workers on the electrically brain are specialized for stimulated brain provide some striking examples of this principle.However different functions. the functional specializations of phrenology were not based on controlled experiments and were not constrained by theories of cognition.For example. Fowler's famous phrenologist's head had regions dedicated to "parental love,”“destructiveness”and“firmness”(Figure 1.3).Moreover,skull shape has nothing to do with cognitive function Although phrenology was fatally flawed,the basic idea of different FIGURE 1.3:The parts of the brain serving different functions paved the way for future phrenologist's head was developments in the nineteenth century,the most notable of which are Broca's used to represent the (1861)reports of two brain-damaged patients.Broca documented two cases in which acquired brain damage had impaired the ability to speak but left the brain other aspects of cognition relatively intact.He e Photos.com/Thinkstock concluded that language could be localized to a particular region of the brain.Subsequent studies argued that language itself was not a single entity but could be further subdivided Wernicke 1874).This was motivated by the observation that brain damage can lead eithe N to d p at le two in the and that be inde red by bra age.This of w knarad forw rd in ter think ng bei ca ng use is nguag single ro and d th models of id not ma h that spe cogni ut nece brain they were loc or underlying neurons brought these processes
Introducing cognitive neuroscience 5 Gall (1758–1828) and Spurzheim (1776–1832) received a bad press, historically speaking, because of their invention and advocacy of phrenology. Phrenology had two key assumptions: first, that different regions of the brain perform different functions and are associated with different behaviors; and second, that the size of these regions produces distortions of the skull and correlates with individual differences in cognition and personality. Taking these two ideas in turn, the notion of functional specialization within the brain has effectively endured into modern cognitive neuroscience, having seen off a number of challenges over the years (Flourens, 1824; Lashley, 1929). The observations of Penfield and co-workers on the electrically stimulated brain provide some striking examples of this principle. However, the functional specializations of phrenology were not based on controlled experiments and were not constrained by theories of cognition. For example, Fowler’s famous phrenologist’s head had regions dedicated to “parental love,” “destructiveness” and “firmness” (Figure 1.3). Moreover, skull shape has nothing to do with cognitive function. Although phrenology was fatally flawed, the basic idea of different parts of the brain serving different functions paved the way for future developments in the nineteenth century, the most notable of which are Broca’s (1861) reports of two brain-damaged patients. Broca documented two cases in which acquired brain damage had impaired the ability to speak but left other aspects of cognition relatively intact. He concluded that language could be localized to a particular region of the brain. Subsequent studies argued that language itself was not a single entity but could be further subdivided into speech recognition, speech production and conceptual knowledge (Lichtheim, 1885; Wernicke, 1874). This was motivated by the observation that brain damage can lead either to poor speech comprehension and good production, or good speech comprehension and poor production (see Chapter 12 for full details). This suggests that there are at least two speech faculties in the brain and that each can be independently impaired by brain damage. This body of work was a huge step forward in terms of thinking about mind and brain. First, empirical observations were being used to determine the building blocks of cognition (is language a single module?) rather than listing them from first principles. Second, and related, they were developing models of cognition that did not make direct reference to the brain. That is, one could infer that speech recognition and production were separable without necessarily knowing where in the brain they were located, or how the underlying neurons brought these processes KEY TERMS Phrenology The failed idea that individual differences in cognition can be mapped onto differences in skull shape. Functional specialization Different regions of the brain are specialized for different functions. FIGURE 1.3: The phrenologist’s head was used to represent the hypothetical functions of different regions of the brain. © Photos.com/Thinkstock
6 THE STUDENTS GUIDE TO COGNITIVE NEUROSCIENCE KEY TERMS about.The approach of using patients with acquired brain damage to inform theories of normal cognition is called cognitive neuropsychology and Cognitiv remains influential today (Chapter 5 discusses the logic of this method in The study of brain detail).Cognitive neuropsychology is now effectively subsumed within the term"cognitive neuroscience."where the latter phrase is seen as being less damaged patients to m theones of norma restrictive in terms of methodology cognition Whereas discoveries in the urosciences continued apace throughout cholog as a Information processing tury took the study of the An approach in which ehavioris。 from its biological underpinnings This did not nitive stages dualism.It was due in part,to some pragmatic constraints.Early pioneer such as Willia m Ja dSigmu and Freud. and per g to say abo the ism ber gy a nd bi ogy notio that ohe logy lie in th of the h and the itive psycho 10s mu impl that tra memory Shi 1968 awn a series of b arrow grams (e.g.,Figure The implication was that one could understand the cognitive system in the same way as one Output patterns nternal FIGURE 14:Examples of box-andarrow and co ectionist models of coenition.Both represent ways of describing cognitive processes that need not make direct reference to the brain
6 THE STUDENT’S GUIDE TO COGNITIVE NEUROSCIENCE about. The approach of using patients with acquired brain damage to inform theories of normal cognition is called cognitive neuropsychology and remains influential today (Chapter 5 discusses the logic of this method in detail). Cognitive neuropsychology is now effectively subsumed within the term “cognitive neuroscience,” where the latter phrase is seen as being less restrictive in terms of methodology. Whereas discoveries in the neurosciences continued apace throughout the nineteenth and twentieth centuries, the formation of psychology as a discipline at the end of the nineteenth century took the study of the mind away from its biological underpinnings. This did not reflect a belief in dualism. It was due, in part, to some pragmatic constraints. Early pioneers of psychology, such as William James and Sigmund Freud, were interested in topics like consciousness, attention and personality. Neuroscience has had virtually nothing to say about these issues until quite recently. Another reason for the schism between psychology and biology lies in the notion that one can develop coherent and testable theories of cognition that do not make claims about the brain. The modern foundations of cognitive psychology lie in the computer metaphor of the brain and the information-processing approach, popular from the 1950s onwards. For example, Broadbent (1958) argued that much of cognition consists of a sequence of processing stages. In his simple model, perceptual processes occur, followed by attentional processes that transfer information to short-term memory and thence to long-term memory (see also Atkinson & Shiffrin, 1968). These were often drawn as a series of box-and-arrow diagrams (e.g., Figure 1.4). The implication was that one could understand the cognitive system in the same way as one FIGURE 1.4: Examples of box-and-arrow and connectionist models of cognition. Both represent ways of describing cognitive processes that need not make direct reference to the brain. KEY TERMS Cognitive neuropsychology The study of braindamaged patients to inform theories of normal cognition. Information processing An approach in which behavior is described in terms of a sequence of cognitive stages
INTRODUCING COGNITIVE NEUROSCIENCE 7 KEY TERMS with t reference The that the brain contains differen of functional Modularity specializati The notion that certain has been arour However, one particular variation on this theme cognitive processes (o and controversy- -namely Fodor's (1983, 1998) theory of modularity. restricted in the type of First,Fodor makes a distinction between two different classes of cognitive information they process process:central systems and modules.The key difference between them relates to the types of information they can process.Modules are held to demonstrate domain specificity in that they process only one particular process (or brain region type of information (e.g.,color,shape,words,faces),whereas central is dedicated solely to systems are held to be domain independent in that the type of information ne part processed is non-specific(candidates would be memory,attention,executive functions).According to Fodor,one advantage of modular systems is that faces.words). by processing only a limited type of information,they can operate rapidly ractivity efficiently and in isolation from other cognitive systems.An additional claim is that modules may be innately specified in the genetic code.Many tages are complete. of these ideas have been criticized on empirical and theoretical grounds.For example,it has been suggested that domain specificity is not innate,although op-down proce the means of acquiring it could be (Karmiloff-Smith,1992).Moreover, es on the or ine of systems like reading appear modular in some respects but cannot be innate because they are recent in evolution.Others have argued that evidence for arlier ones (e.g..memory nfluences on perception interactivity suggests that modules are not isolated from other cognitive processes(Farah,1994). The passage of informa The idea of the mind as a computer program has advanced over the years along with advances in computational science.For example,many tion from simpler (e.g.. e comp cognitive models contain some element of interactivity and parallel e.objects) processing.Interactivity refers to the fact that stages in pro Parallel pro sing stages are complete.Moreover.later stages can influence the outcome of Different information is early ones (top-down processing.in contrast to bottom-up processing) time (ie in parallel) Parallel processing refers to the fact that lots of different informati on e)Altho ugh thes mputers tionally in whic licit models P ear processing occurs using do not e contact ith many interconnected the ce literature always n nodes COMPUTATIONAL AND CONNECTIONIST MODELS OF COGNITION In the 1980s,powerful computers became widely accessible as never before.This enabled cognitive psychologists to develop computationally explicit models of cognition (that literally calculate a set of outputs given a set of inputs)rather than the computationally inspired.but underspecified.box-and- arrow approach.One particular way of implementi omputa onal models has been very influential a mely the neural network,connectionist or paral el distributed processing(PDP)approac (McClelland et al.,1986).These models are considered in a number of places throughout this book notably in the chapters dealing with memory,speaking and literacy
Introducing cognitive neuroscience 7 could understand the series of steps performed by a computer program, and without reference to the brain. The notion that the brain contains different regions of functional specialization has been around in various guises for 200 years. However, one particular variation on this theme has attracted particular attention and controversy—namely Fodor’s (1983, 1998) theory of modularity. First, Fodor makes a distinction between two different classes of cognitive process: central systems and modules. The key difference between them relates to the types of information they can process. Modules are held to demonstrate domain specificity in that they process only one particular type of information (e.g., color, shape, words, faces), whereas central systems are held to be domain independent in that the type of information processed is non-specific (candidates would be memory, attention, executive functions). According to Fodor, one advantage of modular systems is that, by processing only a limited type of information, they can operate rapidly, efficiently and in isolation from other cognitive systems. An additional claim is that modules may be innately specified in the genetic code. Many of these ideas have been criticized on empirical and theoretical grounds. For example, it has been suggested that domain specificity is not innate, although the means of acquiring it could be (Karmiloff-Smith, 1992). Moreover, systems like reading appear modular in some respects but cannot be innate because they are recent in evolution. Others have argued that evidence for interactivity suggests that modules are not isolated from other cognitive processes (Farah, 1994). The idea of the mind as a computer program has advanced over the years along with advances in computational science. For example, many cognitive models contain some element of interactivity and parallel processing. Interactivity refers to the fact that stages in processing may not be strictly separate and that later stages can begin before earlier stages are complete. Moreover, later stages can influence the outcome of early ones (top-down processing, in contrast to bottom-up processing). Parallel processing refers to the fact that lots of different information can be processed simultaneously (by contrast, serial computers process each piece of information one at a time). Although these computationally explicit models are more sophisticated than earlier box-and-arrow diagrams, they, like their predecessors, do not always make contact with the neuroscience literature. In the 1980s, powerful computers became widely accessible as never before. This enabled cognitive psychologists to develop computationally explicit models of cognition (that literally calculate a set of outputs given a set of inputs) rather than the computationally inspired, but underspecified, box-andarrow approach. One particular way of implementing computational models has been very influential; namely the neural network, connectionist or parallel distributed processing (PDP) approach (McClelland et al., 1986). These models are considered in a number of places throughout this book, notably in the chapters dealing with memory, speaking and literacy. COMPUTATIONAL AND CONNECTIONIST MODELS OF COGNITION KEY TERMS Modularity The notion that certain cognitive processes (or regions of the brain) are restricted in the type of information they process. Domain specificity The idea that a cognitive process (or brain region) is dedicated solely to one particular type of information (e.g., colors, faces, words). Interactivity Later stages of processing can begin before earlier stages are complete. Top-down processing The influence of later stages on the processing of earlier ones (e.g., memory influences on perception). Bottom-up processing The passage of information from simpler (e.g., edges) to more complex (e.g., objects). Parallel processing Different information is processed at the same time (i.e., in parallel). Neural network models Computational models in which information processing occurs using many interconnected nodes
8 THE STUDENTS GUIDE TO COGNITIVE NEUROSCIENCE Connectionist models have a number of architectural features.First,they are composed of arrays of simple information-carrying units called nodes.Nodes are information-carrying in the sense that they respond to a particular set of inputs (e.g..certain letters.certain sounds)and produce a restricted set of outputs.The responsiveness of a node depends on how strongly it is connected to other nodes in the network(the"weight"of the connection)and how active the other nodes are.It is possible to calculate,mathematically,what the output of any node would be.given ase of input activations and a set of weights.There are a number of advantages to this type of model For example.by adiusting the weights over time as a result of experience.the model can develop and learn.The parallel processing enables large amounts of data to be processed simultaneously. A more controversial claim is that they have"neural plausibility."Nodes,activation and weights are in many ays analogous to ne rons,firing rates and neural connectivity,respectively. However these models have been criticized for being too powerful in that they can learn many things tha real brains cannot (Pinker Prince.1988).A more moderate view is that connectionist models provide examples of ways in which the brain might implement a given cognitive function,and they generate new predictions that can then be tested.Whether or not the brain actually does impl ement cognition in that particular way will ultimately be a question for empirical research ir cognitive neuroscience KEY TERM The birth of cognitive neuroscience It was largely advances in imaging technology that provided the driving force The basic units of neura for modern-day cognitive neu roscience.Raichle (1998)describes how brain network models that are imagino was in a 'state of indifference and obscurity in the neuroscie activated in response t community in the 1920s"and might never have reached nrominence if it wer not for the involvement of cognitive r psychologistshad psychologists in the 1980s Cognitiv imental designs and infor models that otentially fit well with thes ethods.It is i ological advan nt of f rain but also e isely in s that we er po (except at postmortem) neuro ence is c osed of a broad div t of deta ch apters t th The distinct ompa and tive ne rect electri brain in it ha (e.g.,in Par c th skull on TMS tra ranial y).This considered in Chapte alongside the effect of organic brain lesio Electrophysiological method s (EEG/ERP and single-cell recordings)and magnetophysio methods (MEG)record the electrical and magneti properties of neurons th nselves I he ethods are considered in Chapter 3.In contrast,functional imaging methods(PET,fMRI and fNIRS)record
8 THE STUDENT’S GUIDE TO COGNITIVE NEUROSCIENCE The birth of cognitive neuroscience It was largely advances in imaging technology that provided the driving force for modern-day cognitive neuroscience. Raichle (1998) describes how brain imaging was in a “state of indifference and obscurity in the neuroscience community in the 1970s” and might never have reached prominence if it were not for the involvement of cognitive psychologists in the 1980s. Cognitive psychologists had already established experimental designs and informationprocessing models that could potentially fit well with these emerging methods. It is important to note that the technological advances in imaging not only led to the development of functional imaging, but also enabled brain lesions to be described precisely in ways that were never possible before (except at postmortem). Present-day cognitive neuroscience is composed of a broad diversity of methods. These will be discussed in detail in subsequent chapters. At this juncture, it is useful to compare and contrast some of the most prominent methods. The distinction between recording methods and stimulation methods is crucial in cognitive neuroscience. Direct electrical stimulation of the brain in humans is now rarely carried out as a research tool, although it has some therapeutic uses (e.g., in Parkinson’s disease). The modern-day equivalent of these studies uses stimulation across the skull rather than directly to the brain (i.e., transcranially). This includes transcranial magnetic stimulation (TMS) and transcranial electrical stimulation (tES). These will be considered in Chapter 5, alongside the effect of organic brain lesions. Electrophysiological methods (EEG/ERP and single-cell recordings) and magnetophysiological methods (MEG) record the electrical and magnetic properties of neurons themselves. These methods are considered in Chapter 3. In contrast, functional imaging methods (PET, fMRI and fNIRS) record Connectionist models have a number of architectural features. First, they are composed of arrays of simple information-carrying units called nodes. Nodes are information-carrying in the sense that they respond to a particular set of inputs (e.g., certain letters, certain sounds) and produce a restricted set of outputs. The responsiveness of a node depends on how strongly it is connected to other nodes in the network (the “weight” of the connection) and how active the other nodes are. It is possible to calculate, mathematically, what the output of any node would be, given a set of input activations and a set of weights. There are a number of advantages to this type of model. For example, by adjusting the weights over time as a result of experience, the model can develop and learn. The parallel processing enables large amounts of data to be processed simultaneously. A more controversial claim is that they have “neural plausibility.” Nodes, activation and weights are in many ways analogous to neurons, firing rates and neural connectivity, respectively. However, these models have been criticized for being too powerful in that they can learn many things that real brains cannot (Pinker & Prince, 1988). A more moderate view is that connectionist models provide examples of ways in which the brain might implement a given cognitive function, and they generate new predictions that can then be tested. Whether or not the brain actually does implement cognition in that particular way will ultimately be a question for empirical research in cognitive neuroscience. KEY TERM Nodes The basic units of neural network models that are activated in response to activity in other parts of the network