Barbara J Knowlton. Handbook of Cognition. Editor: Koen Lamberts & Robert L Goldstone. Sage Publication. 2005.
The study of learning and memory is an area of high convergence between neuroscience and cognitive science. How the brain acquires, stores and later retrieves information are among the most fundamental questions in neuroscience. Learning is ubiquitous in the animal kingdom, and we have gained a great deal of knowledge about brain learning mechanisms through the study of animal models (see Squire, 1992, for a review). These studies have provided information about the neural substrates of learning and memory, and it appears that the circuitry devoted to learning and memory in monkeys, rats and mice is quite similar to that found in the human brain. The extensive study of human learning and memory in the past century has allowed researchers to begin to understand the psychological processes that arise from this circuitry.
The basic logic of cognitive neuropsychology is to examine the performance of patients with brain injury, and to infer the structure of cognition through the pattern of impaired and intact performance. This discipline relies on the development of clever tests that can isolate specific components of cognition. If a particular process is impaired in a patient against a background of relatively normal function in other areas, it may be taken as evidence for this particular process as a distinct psychological entity. A similar logic exists in the field of cognitive development. If young children of a particular age have trouble with a specific cognitive task but not with other similar tasks, it would support the idea that this task requires a specific cognitive process that only emerges later in development. Both developmental psychology and cognitive neuropsychology are aimed at discovering the structure of human cognition, but the neuropsychological approach can offer additional information about the brain structures underlying these processes.
The Study of Amnesia
Nearly every lecture or chapter on the human amnesic syndrome begins with the discussion of the case of patient HM. This patient underwent surgical removal of most of the medial temporal lobe on both sides of the brain for the treatment of epilepsy that was intractable to drug treatment. Although this surgical procedure apparently removed the epileptic foci, it was quickly noted that HM suffered a profound memory impairment as an unexpected result (Scoville & Milner, 1957). HM was no longer able to remember specific events that occurred post-surgery, and he was effectively unable to learn new facts, such as the names of new people. Despite his profound amnesia, HM’s other cognitive abilities, such as perception, language and reasoning, were relatively spared. Thus, the primary lesson to be taken from the case of HM is that memory is a mental faculty with distinct underlying brain circuitry. Although this seems quite self-evident today, at that time there was a strong feeling that brain systems were equipotential when it came to higher cognitive functions (e.g. Lashley, 1929). The case of HM focused attention on the medial temporal lobes as playing a specific role in learning and memory.
The case of HM was not only informative regarding the neural substrates of learning and memory, but it also helped to shape psychological theory. The fact that HM was able to retain information for several seconds, and could maintain information indefinitely if not distracted, supported the idea that short-term memory is a stage of processing distinct from long-term memory. Another important feature of HM’s condition is that it was not specific to a particular modality. His memory was impaired for words, faces, melodies, pictures, etc. Memory processing thus appears to be general across materials, at least in part. In addition, the fact that HM retained information acquired before surgery indicated that the retrieval of memories does not rely on the structures required for forming new memories. Thus, HM could remember events occurring throughout his childhood and retained vocabulary and semantic knowledge that was learned before the surgery (Milner, Corkin, & Teuber, 1968).
Perhaps the most important insight gained from the study of HM was that memory is not unitary. Although HM’s memory deficit was extremely profound, it appeared that he was able to show some new kinds of learning. For example, HM was apparently able to learn perceptuo-motor skills like rotor pursuit normally. In this task, the subject must learn to maintain contact with a rotating disk using a stylus. HM showed improvement across trials as he practiced the skill, despite an inability to recall previous training episodes or to recognize the testing apparatus (Corkin, 1968). Thus, it appeared that the learning that occurs in perceptuo-motor skill tasks is accomplished using brain circuitry that is independent of the circuitry required for learning about facts and events. We may therefore infer that these types of learning differ in terms of behavioral properties as well.
The intensive interest in the case of HM has led to numerous further studies of the cognitive abilities of amnesic patients. Although the study of HM has been of utmost importance to the field of neuropsychology, one could argue that his severe childhood epilepsy limits the generalizability of these findings to our understanding of how cognition is organized in the normal brain. However, the results of studies of additional amnesic patients have corroborated the results from HM. Most of these patients have become amnesic not through surgical resection of the medial temporal lobes but through various incidents that have resulted in damage to medial temporal lobe structures or damage to structures in the diencephalon that are interconnected with medial temporal lobe structures. Examples of the various etiologies of amnesia are anoxic episodes, encephalitis, Korsakoff’s disease and thalamic infarcts (Damasio, Tranel, & Damasio, 1991; Guberman & Stuss, 1983; Oscar-Berman, 1980). Although amnesia can arise in many ways, studies of these patients have supported the finding that there are forms of memory that are spared. In addition to the learning of perceptuo-motor skills, amnesic patients have been shown to exhibit intact priming (Cermak, Talbot, Chandler, & Wolbarst, 1985; Schacter, 1985). Priming refers to the fact that people are able to process a stimulus a little more quickly and accurately if they have been presented with it before. For example, subjects are able to identify a word or picture at a shorter duration if they have seen that word or picture recently. Subjects are also better able to identify a stimulus in noise if they had been presented with the stimulus recently. Amnesic patients exhibit priming to the same extent as normal subjects, although their memory for the presentation of the items to be primed is impaired, or even at chance.
The finding that learning on some tasks is spared and on some tasks it is impaired leads one to attempt to find the defining characteristics for the two types of learning. Performance is impaired in amnesia when subjects are aware of learning in a task. That is, when subjects have conscious knowledge of the information learned, amnesic patients are severely impaired relative to controls. For example, when one experiences an event, one is aware of learning about and of using this knowledge when later asked about the event. When a certain face looks familiar, one is aware that the face is evoking a past encounter. When we retrieve facts we are also aware that we have this information stored in memory. This type of memory has been referred to as explicit, or declarative, in that one is able to ‘declare’ the information stored as memory (Squire & Cohen, 1984). Explicit memory is characterized by its flexibility (Eichenbaum, Mathews, & Cohen, 1989). Many different cues can be used to retrieve explicit memories, and explicit memories can be used to guide behavior in new situations. The declarative nature of explicit memories allows this information to be transmitted across individuals. Explicit memory is what people mean when they casually refer to memory.
In contrast to the information stored in explicit memory, people are not necessarily aware of the information stored in implicit memory. For example, people may be able to recognize that they are showing improvement in a motor skill with practice, but they are not aware of what the particular changes in movement were that afforded improved performance. Likewise, participants are not aware that a primed stimulus will be processed more readily in a future encounter; priming only emerges during this future encounter.
The psychological difference between explicit and implicit memory is further supported by the dissociation seen in the performance of amnesic patients. The fact that amnesic patients exhibit normal skill learning and priming has led researchers to better define the role of the medial temporal lobe in terms of its specific role in conscious memory for facts and events. It has also led to extensive study of the learning capabilities of amnesic patients to investigate whether implicit learning can support performance in other tasks. Studies of amnesic patients have been useful in studying mechanisms of priming, sequence learning, preference learning, habit learning and category learning. The impact of neuropsychological work in each of the areas will be discussed in turn.
As described above, priming can lead to an item being perceived more fluently. The fact that amnesic patients show normal levels of this ability demonstrates that the mental representations that give rise to priming are formed independently of the neural circuitry that supports the formation of explicit recognition memory for the items. Behavioral evidence also indicates that recognition and priming are supported by different representations. Unlike recognition, priming is not influenced by depth of processing at study. Thinking about the meaning of words and counting the vowels in words leads to similar levels of priming on a visual identification task, while semantic processing is clearly superior for later recognition (Roediger, Weldon, Stadler, & Riegler, 1992). Priming and recognition also differ in terms of the effects of stimulus change between study and test. Seeing pictures results in less priming for words than if the words themselves had been studied (Weldon & Jackson-Barrett, 1993). This is particularly true if the subject does not have much time to think about the verbal label of the picture. However, the opposite is true for recognition: viewing pictures sometimes leads to better recognition of the corresponding words than if the words themselves had been studied (Snodgrass & Asiaghi, 1977)! Interestingly, it is not the case that the representation supporting recognition is less specific than the representation supporting priming, since subjects are often able to explicitly recognize if the word was presented in the same form or modality as it was during study. The explicit memory representation may contain a great deal of information about the study episode that can be accessed flexibly depending on the task demands (Cooper, Schacter, Ballesteros, & Moore, 1992).
Priming can also exist for conceptual, or semantic, representations. Prior presentation of a semantically related item will allow for more fluent processing of a stimulus (e.g. reading ‘doctor’ will cause one to read or identify ‘nurse’ as a word faster than if ‘window’ had been read; e.g. den Heyer, Briand, & Dannenbring, 1983). Semantic priming can also be shown in a category exemplar generation task. If the word ‘pineapple’ had been presented to you recently, you would be more likely to give the response ‘pineapple’ when asked to name four fruits that come to mind than if you had heard another word (Graf Shimamura, & Squire, 1985). In these cases, what is primed is not a perceptual, but rather a conceptual, representation. When ‘doctor’ is read, knowledge associated with this term is activated, and then comes to mind more readily than if it had not been primed. Similarly, if the concept of ‘pineapple’ is activated, it will come to mind readily when one considers fruits. Importantly, amnesic patients show intact conceptual priming, demonstrating that this activation can occur independently of explicit memory. Indeed, even in neu-rologically intact subjects, neither perceptual nor conceptual priming seems to require that subjects explicitly remember that the primed items had been presented.
One curious aspect of the intact priming ability of amnesic patients is that they appear unable to use the fluency that they presumably experience when they see old items to help them in recognizing these items. If amnesic patients were readily able to use perceptual fluency as a cue that an item was old, one would expect their recognition memory to be reliably better relative to their recall. In fact, the recall and recognition abilities of amnesic patients seem impaired to the same degree (Haist, Shimamura, & Squire, 1992). Thus, these data appear to argue against a substantial contribution of perceptual fluency to recognition memory.
Evidence from the serial reaction time task (SRT) indicates that subjects are able to implicitly learn sequences. In this task, subjects see a stimulus such as an asterisk that appears in different locations on a computer screen and are asked to respond by pressing a key below the location where the asterisk appears. As the asterisk jumps from location to location, the subject continues to press the key beneath the asterisk as fast as he or she can. Unbeknownst to the subject, the asterisks are not appearing in random locations, but rather are appearing according to a complex sequence. It can be shown that subjects are learning this sequence, because when the asterisk locations are surreptitiously switched to random, their reaction time increases significantly. Subjects need not be aware of the sequence, or even that there was a sequence at all, to show this effect. Amnesic patients also show normal sequence learning in this task even though they may have no recognition of the sequence and little memory for the testing episode (Nissen, Willingham, & Hartman, 1989; Reber & Squire, 1998).
Rather than a series of motor movements, it appears that subjects are learning a series of locations. They are thus able to implicitly anticipate where the asterisk will appear and can shift their attention more rapidly than when the asterisk is appearing in random locations. Subjects do not show much of a decrement in performance when they transfer to using a new set of motor effectors if the sequence of locations is the same. However, if a new set of locations is remapped to the old ones, performance suffers substantially, even if the motor responses do not change (Willingham, Wells, Farrell, & Stemwedel, 2000).
Another task that measures the learning of sequential dependences is the artificial grammar task (Reber, 1967). In this task, the subject views a series of letter strings formed according to a rule system that allows only certain letters to follow each letter. Subjects are not told that the letter strings are formed according to any rules until after viewing the letter strings. At that point, subjects are told that their task will be to decide for a new set of letter strings whether each one does or does not follow the rules. Subjects are understandably perplexed at these instructions because they generally feel that they did not notice any rules. Nevertheless, they are reliably able to classify new letter strings at a level significantly above chance. Amnesic patients also show normal artificial grammar learning despite having extremely poor memory for the letter strings that were shown during training (Knowlton, Ramus, & Squire, 1992).
A great deal of research has focused on exactly what subjects are learning in the artificial grammar learning task. In making their classification judgments, subjects appear to be very sensitive to the frequency with which letter bigrams and trigrams had been presented among the training strings. Subjects are likely to endorse test items that include letter combinations that had been frequently repeated during training, and will tend not to endorse items that are comprised of letter combinations that had not been repeated frequently during training. The endorsement of items containing high-frequency bigrams and trigrams may be similar to what happens during priming (Knowlton & Squire, 1994). Those test strings with components that had been repeatedly presented during training might be processed more readily than those containing mostly low-frequency or novel components. Subjects may be using this processing fluency in their classification decisions. As is the case with priming, amnesic patients seem to be using this bigram and trigram frequency information to the same extent as controls, so it appears that this type of learning is not dependent on explicit memory for training strings.
There is evidence that subjects are learning more abstract information about the rules of the artificial grammar as well. If subjects are trained using one letter set, and then are tested using strings formed with a new letter set that maps to the first, subjects are able to classify these ‘translated’ letter strings at a level significantly above chance (Knowlton & Squire, 1996). Amnesic patients also exhibit normal letter-set transfer. There have even been demonstrations of transfer across stimulus domains in which symbols are mapped onto tones (Altmann, Dienes, & Goode, 1995). Subjects seem particularly sensitive to the repetition structure of training items. Test items that contain a pattern of repetitions that was seen frequently during training are more likely to be endorsed (Gomez, Gerken, & Schvaneveldt, 2000). The artificial grammar learning data suggest that both exemplar-based and abstract information about the stimuli can be learned implicitly, and that this information can influence judgments.
The idea that ‘familiarity breeds contempt’ is generally not supported in studies of preference learning. In fact, subjects generally will express preferences for items presented before as compared to novel items. This ‘mere exposure’ effect has been demonstrated with a wide variety of stimuli, including geometric shapes, faces and melodies (see Zajonc, 2001, for a review). Subjects do not need to be aware that the preferred stimuli were shown previously. Subjects simply feel that they are expressing a preference, not that they are recognizing items. In fact, the mere exposure effect can sometimes be stronger for subliminally presented stimuli (Bornstein & D’Agostino, 1992). It is as if explicitly remembering that an item had been seen before gives one a reason for whatever enhanced processing normally gives rise to the preference effect, and thus the apparent preference can be discounted. If processing is enhanced in the absence of explicit memory, then this enhanced processing can be attributed to a preference.
Two studies with amnesic patients have demonstrated that attitude change can occur without explicit memory for previous experience with items. The mere exposure effect has been shown in amnesic patients for Korean melodies that were novel for the subjects (Johnson, Kim, & Risse, 1985). In another experiment, Lieberman, Ochsner, Gilbert, and Schacter (2001) showed that amnesic patients changed their preferences for art prints as a result of having ranked them. In this study, subjects ranked a series of prints in order of preference. Later, they were given a series of pairs of prints two at a time, and they were asked to choose which pair they preferred. Among these pairs were the subject’s 4th and 10th ranked prints and 6th and 12th ranked prints. When subjects subsequently ranked the initial set of prints again, both amnesic patients and control subjects tended to rank the prints in the selected pair higher than previously and the items in the rejected pair lower than previously. Amnesic patients, however, were very poor at remembering which pairs of prints they had chosen between. These results suggest that attitude change resulting from one’s previous behavior does not require explicit memory for that behavior.
The term ‘habit’ has traditionally been used in the field of animal learning to refer to a gradually acquired stimulus-response (S-R) association. Examples of this type of learning in rats include certain visual discrimination tasks in which the animal must make a response (e.g. pressing a bar) when a cue is present to receive a reward (Reading, Dunnett, & Robbins, 1991). Another habit-learning task is the win-stay task in which a rat learns to run down lit alleys of a maze to receive food (Packard, Hirsh, & White, 1989). It appears that in both of these cases, the rat is learning to associate a cue (e.g. a light) with a response (e.g. pressing a bar or running down an alley). The role of the reward in these tasks is to ‘stamp in’ the S-R association. Studies with experimental animals indicate that habit learning can proceed normally if the hippocampal system is damaged. These data raise the question of whether this type of learning occurs in humans, and whether it also is independent of the hippocampus. If so, one might expect that these S-R associations would be formed without awareness of what has been learned. Also, because the learned associations are between cues and responses, the reward would not be necessarily accessed when the cue is present–the response would simply be elicited. Such a situation may occur in everyday learning when a particular response is overlearned, such as the route to your university. The stimuli that you encounter along the way (such as the site of a particular intersection) will elicit particular responses (e.g. turn left). The fact that driving an overlearned route can be based on S-R associations is demonstrated by the phenomenon of finding yourself halfway to your university when in fact you know it is a holiday and you had intended to go elsewhere. You had been relying on S-R habits!
Can we measure human habit learning in a laboratory setting, and examine whether it is independent of the brain system that gives rise to explicit memory? One difficulty with this endeavor is that the tasks used to measure habit learning in rats would be approached quite differently by humans. It is unlikely that a human subject would need several sessions of training to learn that pressing a bar when a light is on will deliver food. A human subject would probably explicitly test the hypothesis that something will happen when the bar is pressed, and after a single reinforcement, the relationship between the light, the bar press and the food would be induced by the subject. In order to circumvent the ability of subjects to use their memory for explicit trials, we used a task in which the association between stimuli and outcomes was probabilistic (Knowlton, Squire, & Gluck, 1994). Memory for individual trials would not be nearly as useful as knowledge gleaned across many trials. We designed a computer game in which a set of cues were probabilistically associated with two outcomes (rainy or sunny weather). The association probabilities ranged from about 80% to about 60%. Subjects saw a combination of the cues on each trial, and decided whether the outcome would be sun or rain. They were given feedback on each trial as to whether their choice was correct, although because of the probabilistic nature of the task there were many trials in which they actually picked the most associated outcome but were given error feedback. Although subjects generally felt that they were not doing very well, they nevertheless generally exhibited learning across 50 trials of the task in that they gradually tended to select the most associated outcome for each cue combination. Amnesic patients showed the same degree of learning as control subjects, although the patients had poor memory for facts about the testing episode, such as the appearance of the cards and the computer screen. Thus, it appears that this task may be an analog of the type of habit-learning tasks that have been used in studies of experimental animals.
One of the most important cognitive abilities is the capacity to classify new items based on experience. Category learning is nearly ubiquitous and thus cannot be explained by a single mechanism. People are able to classify based on the similarity of new items to previous items stored as explicit memories. People may also use rules to classify new items. These rules may have been acquired by direct hypothesis testing, or they may be implicitly acquired through exposure to exemplars, as in the case of artificial grammar learning. Implicit rule learning has also been demonstrated using intersecting-line stimuli in which a complex (non-linear) rule determines the relationship between the lengths of two lines. Amnesic patients have been shown to learn this classification task normally (Filoteo, Maddox, & Davis, 2001).
For most categories in the real world, however, it appears that we use similarity to a central tendency, or prototype, or a category to make classification judgments. People are much faster at classifying a robin as a bird than a chicken, even though both of these adhere to the rules of what defines a bird. The robin, however, is closer to the prototype of a bird that has presumably been formed across the multiple exposure to birds across one’s lifetime (see Posner & Keele, 1968; Rosch, 1988). Prototype formation has some of the defining characteristics of implicit learning: people are not aware that they are forming a prototype during exemplar exposure, they may not necessarily be able to tell you what their prototype is for a particular category (knowledge is non-declarative), and knowledge of the prototype only emerges through the pattern of classification of new items.
The idea that prototype formation can occur implicitly and independently of explicit memory of exemplars is supported by findings from amnesic patients (Knowlton & Squire, 1993). In a series of experiments, amnesic patients and control subjects were shown a set of dot patterns that had been generated by making large random distortions of a ‘prototype’ dot pattern that had been randomly selected. Subjects were not told that the training patterns were generated from a single dot pattern. Later at test, subjects viewed new dot patterns generated by the prototype, some dot patterns that were generated from other prototypes, and some examples of the prototype itself. Control subjects demonstrated evidence of a prototype abstraction effect, in that they endorsed the prototype as being in the category the most often of any trial type, even though the training stimuli were actually high distortions of this prototype, and it was not seen during training. The low distortions were endorsed the next most frequently, followed by the high distortions. Amnesic patients exhibited this same pattern of responses, and endorsed the items in the category as accurately as control subjects did. These results support the notion that general knowledge categories may have a distinct status in the brain, separate from knowledge of specific exemplars.
The Logic of the Double Dissociation
Intact learning in amnesic patients is taken as compelling evidence that the system engaged is independent of the system that is required for explicit learning. However, it is possible that such dissociations could be obtained if there was only one learning system, with different tasks drawing on it differentially. For example, if the weather prediction task described above requires only weak, fuzzy declarative memory for the cue-outcome relationships, it is possible that amnesic patients could exhibit nearly normal learning on this task, even though they would exhibit severe impairments on tasks that draw heavily on explicit memory, like recognition tests. One approach to address this issue is to look for double dissociations, in which there is a second group of patients with different brain pathology who are intact on an ability impaired in the first group, but impaired on an ability intact in the first group. If a double dissociation is demonstrated, it is unlikely that intact learning performance by amnesic patients on a task is simply due to low explicit memory demands on that task. Rather, it would suggest that learning is dependent on a different system that is specifically affected in the second patient group. The logic of the double dissociation in neuropsychology has been attributed to Hans-Lukas Teuber (1950), who advocated its use in assigning functions to specific brain systems.
A concrete example of a double dissociation exists for perceptual priming. There have been a few case studies of patients with diffuse occipital lobe damage who exhibit impaired perceptual priming, even though they appear to recognize stimuli normally (Fleischman, Vaidya, Lange, & Gabrieli, 1997; Keane, Gabrieli, Mapstone, Johnson, & Corkin, 1995; Samuelsson, Bogges, & Karlsson, 2000). Thus, the normal performance of amnesic patients on priming tasks cannot be explained by assuming it is a more sensitive measure of residual declarative memory, since patients with damage elsewhere in the brain find priming to be the more difficult task. The fact that patients with occipital lobe damage exhibit impaired perceptual priming is consistent with neuroimaging studies showing a decrease in activation in occipital regions for primed items (e.g. Koutstaal, Wagner, Rotte, Maril, Buckner, & Schacter, 2001). This decrease in activation, typically measured as blood flow in functional magnetic resonance imaging paradigms, could be seen as the neural substrate of perceptual fluency. Those items previously presented require fewer neural resources for subsequent processing.
Another example of a double dissociation in memory exists for habit learning and explicit memory. Patients with Parkinson’s disease, which affects the basal ganglia system of the brain, are impaired on the weather prediction task described above. Despite this impairment in habit learning, patients with Parkinson’s disease can exhibit completely normal recognition memory for the stimuli used to train the habit (Knowlton, Mangels, & Squire, 1996). This is the exact opposite of the pattern seen in amnesic patients. This double dissociation argues against the idea that habit learning is simply an easy explicit memory test. Rather, it suggests that habit learning depends on a different brain system than explicit memory, namely the basal ganglia. As is the case with priming, neuroimaging data support this idea. Activation in the basal ganglia is seen when subjects perform the weather prediction task (Poldrack, Prabhakaran, Seger, & Gabrieli, 1999). It appears that activity in the medial temporal lobe actually decreases when subjects perform this task (Poldrack et al., 2001).
In addition to difficulties in forming new memories, patients with the amnesic syndrome typically exhibit some impairment in remembering information that was acquired before the onset of amnesia. Typically, retrograde amnesia occurs for a period just preceding the event that precipitated amnesia, while memory for more remote time periods remains intact (see Squire & Zola-Morgan, 1996, for a review). The duration of this period varies substantially across amnesic patients. For some, retrograde amnesia may only extend for months or a few years before the onset of amnesia. For other patients, retrograde amnesia can last for decades with only childhood memories spared. The extent of retrograde amnesia is typically related to the severity of anterograde amnesia, although many exceptions have been reported in which retrograde amnesia is more or less severe than the severity of anterograde amnesia (Kopelman & Kapur, 2001).
The existence of retrograde amnesia demonstrates that the brain structures damaged in amnesia are not only involved in the storage of new memories, but also may play a role in the retrieval of memories for some period of time after learning. The temporally graded nature of retrograde amnesia is consistent with the idea that the medial temporal lobe contributes to a consolidation process whereby memory traces gradually undergo a transformation such that they can ultimately be retrieved independently of medial temporal lobe structures. This process would appear to be a slow one, lasting years.
The idea that memories gradually ‘consolidate’ is puzzling in a sense. Why should memories that are perfectly retrievable at one time undergo such a transformation? One possible reason why memories should consolidate may arise from the fact that medial temporal lobe structures including the hippocampus have limited storage capacity that makes them inadequate to hold the vast amount of memories that one gains across one’s lifetime. Rather, the vast expanse of the cerebral cortex is more likely to provide such space. It is generally thought that the hippocampus has the capacity for plasticity that affords extremely rapid learning. In contrast, the formation of permanent memory traces in cortex may take considerably more time. Although some types of cortical plasticity may occur rapidly (e.g. priming), the formation of long-lasting declarative memory traces that involve multiple semantic interconnections and can be accessed flexibly is likely to occur more gradually. According to one view, it is the role of the hippocampal system to acquire memories and then ‘train’ the cortex until the memory is fully instantiated there (McClelland, McNaughton, & O’Reilly, 1995). This process of reinstantiation of memories is hypothesized to be gradual: the interleaving of reinstated memories in the cortex fosters the formation of robust, integrated representations that are less susceptible to interference than if memories were stored in cortex successively.
Patients who have retrograde amnesia that extends for several decades have usually sustained extensive damage to cortical regions in the temporal lobes (Cermak & O’Connor, 1983; Reed & Squire, 2000). This finding is consistent with the notion that cortical regions are the ultimate storage site for memories, since these patients are likely to have sustained damage to these sites. Only childhood memories exist in these patients, which perhaps are represented in a very distributed manner because they were acquired while the cortex was maturing. There have been a few case studies of patients who lose all memories for events, including those occurring during childhood (Damasio, Eslinger, Damasio, Van Hoesen, & Cornell, 1985; Tulving, Schacter, McLachlan, & Moscovitch, 1988). In these patients, there is likely to be damage to both frontal and temporal lobes, which might broadly affect the strategic ability to recall as well as memory storage sites. Interestingly, much semantic knowledge is generally intact in these patients, such as their language abilities and knowledge of game rules. It appears that well-learned semantic knowledge may be a particularly robust form of memory.
The finding that retrograde episodic memories are particularly affected in some cases of brain damage has led to the view that these memories may be stored differently than semantic memories. Patients with retrograde amnesia are particularly poor at recounting autobiographical memories; that is, memories of events that have happened to them. These patients are much better at personal semantics. For example, they may be able to remember the date of their wedding but not have a first-person recollection of the event. It is unfortunately difficult to definitively separate true episodic memories from semantic memories. Many of our childhood memories are in the form of stories that we have told many times, and thus may lack the episodic character of memories where we can feel that we can travel back to another place and time. As time passes, an episodic memory for an event becomes a memory for the fact that the event happened. Soon after a lecture, a student may remember the episode when a professor mentions a particular experimental finding. Much later, knowledge of the finding may persist without remembering the exact moment when the information was learned. It may be that structures in the medial temporal lobes play a persistent role in supporting true episodic retrieval for memories throughout the lifespan, while older memories become ‘semantisized’ and thus independent of this system (Rosenbaum, Winocur, & Moscovitch, 2001). However, it is also possible that the difference between episodic and semantic memories is one of degree and not kind. By definition, episodic memories occur only once, whereas memories for facts can occur across multiple study occasions. Also, memories for episodes tend to include many multimodal components. Thus, these memories may be more fragile by their nature and could be more susceptible to damage to storage sites in cortex than would semantic memories.
Alzheimer’s Disease and Dementia
The study of amnesia receives the greatest attention in the neuropsychology of memory because of its theoretical importance to the study of mnemonic processes and their neural substrates. However, the majority of patients who exhibit memory impairments as a consequence of a neurological disorder suffer from Alzheimer’s disease and not the amnesic syndrome. Alzheimer’s disease is a degenerative disease that strikes older adults. A memory deficit is the most commonly associated symptom of Alzheimer’s disease. However, Alzheimer’s disease can only be diagnosed if there is at least one additional cognitive deficit present. This is often a problem with naming objects (Price, Gurvit, Weintrub, Geula, Leimkuhler, & Mesulam, 1993). Visuospatial problems, difficulties with executive function and attention deficits are also common in Alzheimer’s disease. Patients with Alzheimer’s disease may also exhibit psychiatric symptoms such as apathy or agitation, which can be particularly difficult for caregivers to manage (Mega, Cummings, Fiorello, & Gombein, 1996). At the very earliest stages of the disease, however, the memory impairment may be the only significant deficit.
Another primary characteristic of Alzheimer’s disease is that cognitive abilities decline as the disease progresses. The non-selective and progressive nature of Alzheimer’s disease makes it less well suited than the amnesic syndrome for the study of the brain areas involved in memory processing. However, the study of Alzheimer’s disease has enormous clinical significance given that people are living longer into old age.
Alzheimer’s disease is associated with neuronal loss and characteristic neuropathology in the brain. Plaques composed of amyloid protein and neurofibrillary tangles are readily visible through a microscope in the post-mortem brains of patients with Alzheimer’s disease, especially in medial temporal lobe regions (Arnold, Hyman, Flory, Damasio, & Van Hoesen, 1991). It is not known whether these pathological changes directly cause the cell loss, or if they are only markers of a disease process that is killing cells. The vast majority of cases of Alzheimer’s disease are not linked to the inheritance of a particular gene. However, there are genetic risk factors associated with Alzheimer’s disease (see Holmes, 2002, for a review). For example, people with the gene that codes for a particular form of apolipoprotein E have a greater risk than those who have genes that code for other forms of this protein. This risk is even greater when this gene is present on both alleles. Apolipoprotein E is involved in the transport of lipoproteins into cells. Thus, it may be important in order for neurons to maintain health and normal synaptic contacts (see Poirier, 2000, for a review). Particular forms of apolipoprotein E may not be quite as efficient at these tasks as other forms, and this may hasten the disease process. Other risk factors that have been suggested include head trauma and low education level (Lye & Shores, 2000; Small, 1998). Both of these factors relate to the idea that there is a certain amount of ‘cognitive reserve’. If individuals become symptomatic at the point at which cognitive abilities decline beyond the level at which daily life activities are affected, then confronting old age with a higher level of cognitive reserve would delay the onset of Alzheimer’s disease, perhaps past the point of the end of life. Head trauma may induce brain damage that may not have a functional effect until the brain is further challenged by aging and the Alzheimer’s disease process. Education appears to have a protective effect such that patients with higher education levels receive the diagnosis of Alzheimer’s disease later (although the mortality rate does not appear to be affected: Qui, Backman, Winblad, Aguero-Torres, & Fratiglioni, 2001). Higher educational achievement is likely to be associated with continued intellectual stimulation throughout life. It may not be the case that such intellectual stimulation prevents the Alzheimer’s disease process, but rather it might prolong the capacity to function within a normal range.
The cognitive reserve idea is particularly relevant if one accepts the view that Alzheimer’s disease is an inevitable consequence of aging in the nervous system. By this view, if we each lived long enough, we all would develop Alzheimer’s disease eventually. The strategy for developing treatments is not necessarily one of finding a way to prevent the disease, but of putting it off sufficiently.
The memory deficits in Alzheimer’s disease patients are similar to those of amnesic patients. Patients with Alzheimer’s disease exhibit declarative memory deficits accompanied by normal motor skill learning and perceptual priming (see Gabrieli, 1998, for a review). This is consistent with the anatomical findings that the basal ganglia and primary motor and sensory cortical regions are relatively spared in Alzheimer’s disease. Patients with Alzheimer’s disease differ from amnesic patients in that they exhibit deficits in forms of priming that depend on more complex representations. For example, patients with Alzheimer’s disease have been reported to exhibit poor semantic priming, as in category exemplar generation tasks. These deficits may be due to a breakdown in the semantic networks that support such priming (Vaidya, Gabrieli, Monti, Tinklenberg, & Yesavage, 1999). Because semantic deficits are somewhat variable in Alzheimer’s disease, deficits in semantic priming do not seem to be as consistent as the declarative memory deficits. Patients with Alzheimer’s disease also often have deficits with word stem and fragment completion tasks that are performed normally by amnesic patients (Russo & Spinnler, 1994). These deficits point to disruption in lexical-level representations in Alzheimer’s disease.
Patients with Alzheimer’s disease exhibit retrograde amnesia that becomes increasingly dense as the disease progresses. More remote memories are generally preserved (Beatty, Salmon, Butters, Heindel, & Granholm, 1988). It is common for patients in later stages of Alzheimer’s disease to eventually forget such pertinent information as the names of their children or their spouse. This loss is not confined to episodic memories. As mentioned above, patients also lose semantic knowledge as the disease progresses. The loss of past memories is likely related to involvement of cortical regions that are the sites of memory storage.
Although Alzheimer’s disease is the most common form of dementia, other progressive dementia disorders have been described. One of these disorders, frontotemporal dementia, is particularly interesting to neuropsychologists because the constellation of symptoms can be fairly selective early in the disease, and the symptom pattern can be contrasted with that of Alzheimer’s disease. Frontotemporal dementia appears to exist in at least two subtypes (see Hodges & Miller, 2001, for a review). In the frontal variant, personality changes, social disinhibition and problems with executive function are the earliest signs, and degeneration of prefrontal cortex is present, especially in cell layers involved in communication between cortical areas. In the temporal variant, there are differences between patients depending on whether the left or right temporal lobe is affected. Patients with left temporal lobe degeneration exhibit a decline in semantic knowledge (also called semantic dementia). This would manifest itself as problems in answering general knowledge questions about the world (e.g. What continent is Egypt located in?), defining vocabulary words, or naming pictures (Hodges, Patterson, Oxbury, & Funnell, 1992). The deficit is not specific to language output, since these patients also have difficulty in displaying semantic knowledge non-verbally. For example, they have difficulty in recognizing incompatibilities in the pyramid and palm trees test, such that they may not identify that a palm tree, and not a pine tree, belongs in a desert pyramid scene. Degeneration in the left temporal lobe appears to disrupt semantic networks that had been built up across the lifespan. Patients with severe semantic dementia may be able to identify only simple familiar items learned early in life (like ‘dog’ or ‘cup’: Lambon Ralph, Graham, Ellis, & Hodges, 1998). Right-sided temporal lobe degeneration results in problems with empathy and social cognition (Perry, Rosen, Kramer, Beer, Levenson, & Miller, 2001). It may be that the right temporal lobe has a similar function to the left in storing semantic knowledge, but specific to social interaction.
The profile of frontotemporal dementia is somewhat different from that of Alzheimer’s disease. Frontotemporal dementia typically begins in late middle age, while Alzheimer’s disease affects older adults with risk increasing with age. Many patients with the frontal variant are diagnosed initially with a psychiatric condition. There is a much stronger genetic component to frontotemporal dementia than there is to Alzheimer’s disease. In frontotemporal dementia, deficits in declarative memory are not as pronounced as they are in Alzheimer’s disease. Patients with semantic dementia may have difficulty with memory tasks when control subjects are able to benefit from deep (semantic) encoding. Semantic dementia patients can perform normally on recognition memory tasks using meaningless stimuli or simple objects that do not benefit much by elaborative encoding. Likewise, patients with the frontal variant may have difficulty with memory tasks that require strategic processing (see section below on frontal lobe contributions to memory). Finally, the plaques and tangles that are characteristic of the neuropathology in Alzheimer’s disease are not present in frontotemporal dementia. In Alzheimer’s disease, medial temporal lobe structures are in general affected earliest, while in frontotemporal dementia frontal lobes or polar and temporal cortex is affected at the onset. Structures in the medial temporal lobe tend to be spared initially. In both of these disorders, degeneration is progressive and eventually includes extensive regions of cortex.
A comparison of the performance of patients with semantic dementia and Alzheimer’s disease on retrograde memory tests yields an interesting double dissociation. Patients with Alzheimer’s disease have greatest difficulty recalling more recent events, while the opposite is true for semantic dementia patients. These patients generally have more difficulty relative to control subject in recalling old memories than new memories (Graham & Hodges, 1997). These results support the view that medial temporal lobe structures are important for the acquisition of information into declarative memory, but that the eventual storage site of these memories is in networks in the temporal cortex. In the semantic dementia patients, new memories could be stored in remaining neocortex, but cortical degeneration had impaired access to older memories.
Frontal Lobes and Memory
Damage to prefrontal cortex is relatively common, and can result from strokes, removal of tumors or head injury. Because of their position on the head, the frontal lobes are particularly susceptible to damage during vehicular accidents. Although medial temporal lobe structures are firmly identified with declarative memory processing, the frontal lobes are clearly part of the circuit that contributes to this cognitive ability. While subjects are performing declarative memory tasks, activation is typically seen in regions of prefrontal cortex using neuroimaging techniques (for reviews see Cabeza & Nyberg, 2000; Maguire, 2001; Yancey & Phelps, 2001). In fact, these prefrontal activations are often more pronounced than activations in the medial temporal lobe. Patients with frontal lobe damage do not exhibit an amnesic syndrome, in that they are not globally impaired on all recall and recognition tasks. However, deficits in memory function are readily apparent in patients with prefrontal damage, and these deficits have an impact on their daily life activities.
Activation in left inferior prefrontal cortex is present while subjects are asked to encode information for a later memory test, and this activation is correlated with the ability of subjects to later remember words and pictures. It seems that this region is involved in semantic processing of items, and those items that are deeply processed are remembered better later (see Gabrieli, Poldrack, & Desmond, 1998, for a review). Although patients with prefrontal damage are not necessarily impaired on semantic tasks, such as tests of vocabulary, they do seem to have trouble using this information to benefit memory encoding. Patients with prefrontal cortical damage do not benefit from the inclusion of semantically similar items in the study list (Hirst & Volpe, 1988; Incisa della Rocchetta & Milner, 1993). For example, if control subjects are given a list of fruits (apple, pear, orange), they will recall the list better than a list of unrelated words. Patients with prefrontal damage do not show this benefit. It is as if they are not able to actively form the inter-list associations that are required in order to benefit from the semantic relatedness of the words.
This is also shown by differences in the recall output of controls and patients with prefrontal damage. If given a list with multiple intermixed categories, control subjects will tend to cluster items within a category when they are recalling the list. Patients with prefrontal damage show this clustering to a lesser extent, suggesting that they are not making use of semantic relatedness at encoding (Gershberg & Shimamura, 1995). Patients with prefrontal damage who also have memory problems have been reported to show a lack of release from proactive interference (Freedman & Cermak, 1986). For example, if control subjects learn successive lists of items from the same semantic category (e.g. animals), their performance will decline due to proactive interference from successive lists. When a new list of items from a different semantic category is given (e.g. cities), there is a release from this proactive interference, and performance improves. Patients with prefrontal cortical damage show proactive interference even for non-related lists because they have general difficulty with interfering stimuli (Shimamura, Jurica, Mangels, Gershberg, & Knight, 1995; Smith, Crane, & Milner, 1995). Frontal lobe patients may have particular difficulty benefiting from a category switch across lists when memory traces are weak.
The deficits in semantic encoding in prefrontal patients can be attenuated to some extent by providing encoding support (Gershberg & Shimamura, 1995; Incisa della Rocchetta & Milner, 1993; Vogel, Markowitsch, Hempel, & Hackenberg, 1987). For example, giving a category label or clustering items in a mixed list seems to help these patients encode semantic information. These manipulations also benefit control subjects to some extent, although in most cases control subjects already use semantic encoding well enough for them not to receive additional benefit from making semantic relatedness more transparent. The fact that prefrontal patients can use semantic encoding underscores the idea that their semantic networks are largely intact, but are not employed effectively in memory encoding.
Patients with prefrontal damage also have difficulty with memory retrieval. Their recall performance is typically much poorer than their recognition performance (Dimitrov, Granetz, Peterson, Hollnagel, Alexander, & Grafman, 1999; Jetter, Poser, Freeman, & Markowitsch, 1986; Wheeler, Stuss, & Tulving, 1995). This contrasts with the performance of amnesic patients, in which recall and recognition are impaired to the same extent. The fact that prefrontal patients can perform normally on recognition tasks suggests that they may be able to encode new information, but they may not be able to effectively use strategies to retrieve information. The less support given to recall, the greater the deficit in prefrontal patients. For example, these patients fare the most poorly with tasks of free recall and exhibit relatively smaller deficits with cued recall. Recognition memory can be quite normal, especially forced choice recognition in which the subject must choose between a novel and a familiar item. With free recall, the subject must essentially form a retrieval ‘plan’ in which items are strategically retrieved that will aid in retrieving the desired information. Thus, to recall the name of one’s third-grade teacher, one might attempt to retrieve memories of one’s third-grade playmates and important events that happened that year in order to activate the memory of the teacher. Deficits in the ability to form this plan would impair effective recall but would leave recognition intact. Patients with prefrontal damage have particular difficulty with assessing the source of memories or the temporal order with which memories occurred (Janowsky, Shimamura, & Squire, 1989; Mangels, 1997; McAndrews & Milner, 1991). Even for items that are correctly recognized, prefrontal patients have difficulty with discriminating where and when the memory was acquired. This stands in contrast to the performance of amnesic patients, who generally exhibit proportionate impairments in item and source memory. Source memory and temporal order judgments require the recall of specific information that requires a great deal of self-generated cueing. Thus, prefrontal patients may find it particularly hard to access the specific context in which learning took place.
The role of the prefrontal cortex in memory is also supported by neuroimaging evidence. Prefrontal activation is commonly seen when subjects retrieve items (Cadoret, Pike, & Petrides, 2001; Henson, Rugg, Shallice, & Dolan, 2000; McDermott et al., 1999). It is not known whether this activation is related to the search of memory contents, the evaluation of the products of the retrieval process (such as whether a retrieved item actually exceeds threshold for being in the study phase) or some other retrieval-related process. Given the fact that the prefrontal cortex is a large, heterogeneous region in humans, it is likely that it is involved in multiple retrieval-related subprocesses.
The idea that prefrontal patients have trouble with monitoring the products of retrieval has been suggested in order to explain the phenomenon of confabulation in these patients. Prefrontal patients, particularly those with memory disorders, have been reported to answer questions with impossible, inconsistent answers (Fischer, Alexander, D’Esposito, & Otto, 1995; Stuss, Alexander, Lieberman, & Levine, 1978). For example, a man might claim that he is a newly-wed despite the fact that he had been married for several decades. These confabulations appear to be distinct from deliberate untruths, in that patients do not seem to have any motivation to lie. Also, it does not seem that the confabulations are simply due to poor memory for facts and events in the past. Amnesic patients who have profound retrograde memory deficits do not typically confabulate. It may be that prefrontal patients have a deficit in their ability to judge whether what has been retrieved is appropriate and to inhibit retrieved information that is inappropriate. According to this view, a query into memory would result in the retrieval of several different pieces of information that were activated. For example, when asked to recount a recent vacation, one might initially broadly retrieve memories about the recent vacation as well as previous vacations, and even vacations imagined or described in recent television programs. The products of retrieval are immediately filtered for credibility. Also, as specific temporal information for the memories is retrieved, memories of older vacations can be eliminated from contention. In this way, memory for the recent vacation can emerge as the correct memory in this context. Even in control subjects, many recounted memories are blended with elements of related memories. Prefrontal patients may not be able to effectively inhibit inappropriately retrieved memories and thus they may come up with odd confabulations that are really blends of various bits of retrieved information. A similar difficulty in retrieval monitoring may contribute to the finding that prefrontal patients make numerous intrusion errors from previous lists when given successive lists to remember.
All of the memory processes that are disrupted by frontal lobe damage appear to draw heavily on working memory resources. Because neuroimaging and neuropsychological studies have linked working memory to the function of cortical loops that include prefrontal cortex (for a review see Baddeley & Della Sala, 1996; Carpenter, Just, & Reichle, 2000), it is possible that long-term memory deficits are a direct consequence of working memory impairment in these patients. For example, deficits in semantic encoding could result if subjects were unable to hold and manipulate information in working memory. It would be difficult to form inter-list associations or extract common semantic content across items if working memory capacity or the ability to manipulate or update items in working memory were impaired. Providing prefrontal patients with some semantic structure during learning may help to overcome working memory limitations.
Working memory deficits would also have a major impact on strategic abilities such as planning and problem solving. Patients with prefrontal cortical damage are notoriously poor at planning (see Owen, 1997, for a review). Planning requires the generation and simultaneous consideration of multiple subgoals that should be followed in order to execute a plan. For example, in order to plan a meal one must form subgoals, such as shopping for food and preparing the various dishes. Each of these subgoals can be further broken down into additional subgoals. Multiple subgoals must be integrated in order to form an efficient plan. One must start cooking the planned dishes at different times, and one must make sure that if two dishes require the same ingredient that there is enough for both. Prefrontal patients often perform tasks like shopping inefficiently because of their difficulty in forming and/or using goal hierarchies.
This strategic deficit could also extend to the formation and execution of retrieval plans. The generation of effective retrieval cues requires the simultaneous consideration of the ultimate retrieval goal as well as retrieval subgoals for relevant information. For example, if I want to remember what I had for breakfast last Tuesday, I would keep this main goal in mind while I tried to recall last Tuesday’s date, which might help me remember if I had given a lecture that day. If I could remember the lecture, I could remember if I was in a hurry that day, and thus I could remember if I ate at home or not. If I did, remembering when I had gone shopping would give me cues as to what I might have had in the house at the time. Performing a difficult recall task such as this requires several pieces of information to be held in mind at once, and thus requires substantial working memory. Such tasks may be difficult for prefrontal patients because of working memory deficits. Retrieval based on familiarity, as in many recognition tasks, does not require such strategic processing, and is relatively intact after prefrontal damage.
The concept of working memory does not merely refer to a passive buffer of information held in mind, but also to the manipulation of the information in the service of cognition. The delineation of the processes that comprise working memory is currently a ‘hot’ topic in cognitive science. Patients with prefrontal damage may have limitations in working memory capacity, or they may have problems with inserting items into working memory, integrating multiple propositions in working memory, or inhibiting irrelevant items from flooding working memory. If the inhibition of irrelevant information is considered as part of working memory, one could also interpret intrusion errors and confabulations as stemming from working memory dysfunction. Recent neuroimaging studies have suggested that different components of working memory may be subserved by different regions within the prefrontal cortex. It is thus likely that working memory is affected differently in different prefrontal patients depending on the locus of damage. Because the prefrontal cortex represents almost one-third of the human brain, it is likely to support multiple cognitive operations that contribute to complex thought.
Memory and Normal Aging
Neuropsychology has traditionally focused on the effects of brain injury, but it is clear that even normal, healthy, aging results in brain changes that affect cognition. Older adults, on average, exhibit a number of deficits compared to young adults. One of the most striking differences is in reaction time, suggesting that processing is slowed with age (see Salthouse, 1996, for a review). Older subjects also exhibit poorer declarative memory than young adults. Non-declarative memory appears to be relatively spared, although mild deficits are also reported. Working memory deficits have also been reported, which appears to have a negative impact on executive function in aging (Li, Lindenberger, & Sikstrom, 2001). Many brain systems are affected by aging, and thus it follows that many cognitive systems are affected as well. However, it is not the case that aging involves an across the board decline in cognitive function. Semantic knowledge such as vocabulary and fact knowledge seem to be spared, or even superior in older adults (Eustache et al., 1995; Horn & Donaldson, 1976). Older subjects have had more time to acquire semantic knowledge over the course of their lifetime. Thus, older subjects can perform very well on tasks that require knowledge of the world, but have relatively more trouble when participants must quickly generate solutions to novel problems.
Memory problems are a typical complaint among older adults. Older adults seem to have particular difficulty with source memory, with relatively fewer problems with recognition memory (Schacter, Osowiecki, Kaszniak, Kihlstrom, & Valdiserri, 1994). This pattern supports the idea that the deterioration of memory with aging has much to do with prefrontal involvement. The working memory problems seen in aging are also consistent with prefrontal degeneration. There is also neuropathological evidence that the frontal lobes are particularly affected in aging. The frontal lobes continue to develop late into childhood, and this late development is mirrored at the opposite end of the developmental spectrum by relatively early degeneration in old age (Scheibel, Lindsay, Tomiyasu, & Scheibel, 1975; Thatcher, 1991).
Although aging is associated with cognitive deficits on average, equally striking is the increase in variability in performance that accompanies aging (Christensen, Mackinnon, Jorm, Henderson, Scott, & Korten, 1994). A significant number of older adults perform well into the range of younger subjects on declarative memory tasks. There are probably several reasons for this variability. First, older adults are likely to differ in the extent of age-related neuropathology. Second, some older adults may be able to use alternative strategies to solve problems, such as relying on well-developed semantic knowledge to enhance encoding. A third factor may be the extent to which older subjects have continued to develop their cognitive capabilities. It appears that individuals who continue social and physical activities into old age maintain cognitive function longer (Laurin, Verreault, Lindsay, MacPherson, & Rockwood, 2001; Stuck, Walthert, Nikolaus, Bijla, Hohmann, & Beck, 1999). Increased activity may protect against neurodegeneration, or it may allow more efficient use of remaining neural substrate. Social and physical activities are likely to convey specific benefits, and these activities generally involve cognitive stimulation as well (e.g. joining a bridge club, or noticing new things along a daily walk). Maintaining an active lifestyle into advanced age appears to have many interactive health benefits. As medical advances prolong the lifespan, it is becoming increasingly important to find ways to enhance the quality of life of older adults. Enhancing opportunities for cognitive, social and physical activity among the aged may be an effective means toward this goal.
Neuropsychology and Functional Neuroimaging
The most fundamental discoveries in neuropsychology were made before the development of modern brain-imaging techniques. Studies of brain-injured patients provided key support to concepts such as cerebral lateralization, executive function and levels of representation in visual perception. Neuropsychology relied on the careful study of patients in whom the site of brain damage was only generally known. Even when a post-mortem analysis of the brain revealed the extent of damage, it was difficult to pinpoint which of the multiple regions that were damaged were truly responsible for the behavioral deficit. In some sense, this was not critical since the goal of neuropsychology was to understand the components of cognition and not necessarily link structure and function. More recently, magnetic resonance imaging of the brain has allowed neuropsychologists to identify the extent of brain damage with excellent resolution in a non-invasive manner, allowing more information about the specific structures damaged. However, the problem remains that the brain damage that occurs in patients often encompasses more than one neuroanatomical region. For example, in the case of patient HM described above, several structures were removed so it was not clear which of these structures were critical for declarative memory.
Functional neuroimaging techniques can avoid some of the ambiguity that arises in the study of patients in whom damage is typically non-specific. Techniques such as positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) allow one to examine brain regions that are activated during performance of a cognitive task. Rather than inferring structure-function relationships by examining functional losses when a structure is damaged, a direct link can potentially be made between a structure and its function. PET and fMRI do not directly measure the medium of communication in the brain (neuronal action potentials), but rather cerebral metabolism (PET) and blood oxygenation level (fMRI), which are thought to be reasonable proxies for neural activity. The resolution of PET is about 5 mm, while that of fMRI is much better (as low as 1 mm) which can allow, for example, the differentiation of hippocampus from surrounding medial temporal lobe cortices.
Functional neuroimaging also offers a degree of temporal resolution not possible in neuropsychological studies. For example, in patients with amnesia, damage is present during the encoding, consolidation and retrieval phases. Thus, it is not known how the medial temporal lobe structures damaged in amnesia may play different roles in these processes. Unlike studies with PET, in which brain activation over blocks of several minutes is measured, fMRI allows the measurement of activation associated with specific single events, such as correct recollection of stimuli or the study of stimuli that will be later remembered. Such event-related fMRI allows us to assess brain activation associated with different hypothesized stages of cognitive processes.
Another difficulty with inferring structure-function relationships from neuropsychological studies is the possibility of functional recovery. Plasticity in the nervous system is well documented. It is quite possible that a given structure may typically contribute to a cognitive ability, but other structures may be able to take over after damage, leading to no loss of function. Thus, the study of brain-injured patients may underestimate the contribution various structures make to cognitive function.
Despite these advantages of the neuroimaging approach, neuropsychology arguably offers some complementary advantages. For example, there are many constraints one must make in the design of neuroimaging experiments. In PET, blocks of one trial type must last several minutes, and thus it is impossible to intermix stimuli in different conditions. For example, if brain activation to recognized words was compared to brain activation to unrecognized words, one would need to block old and new words separately at test. In event-related fMRI, one can measure activation on intermixed trials, but one must design the experiment so that there are many trials of each type to measure reliable activation. In addition, subjects are very constrained in terms of the response they can make (generally button presses), since subjects are confined to the scanning apparatus and movement must be minimized.
Another issue in the interpretation of neuroimaging data is the fact that these techniques are contrastive. Thus, activation is always measured relative to some other condition. Often this condition is rest or the performance of a cognitively undemanding task. However, such contrasts may lead to misinterpretation (Newman, Tweig, & Carpenter, 2001; Stark & Squire, 2001). For example, comparing the activation associated with the recall of a list of words with the activation during ‘rest’ might not reveal structures truly involved in recall if those are also actively involved in recalling the subject’s shopping list during the ‘rest’ condition!
Finally, neuropsychological studies can help answer the question of whether or not a region of the brain is necessary for a function or simply active during performance. For example, neuroimaging studies of declarative memory have in the past shown much more robust activation in prefrontal structures than in the medial temporal lobe. Medial temporal lobe activation may simply be more difficult to see because of its position in the brain, its vascularization, or because it may be constantly active even during baseline conditions. The frontal lobe is clearly active in strategic aspects of encoding and recall as can be seen in the performance of prefrontal patients on memory tasks. However, the findings from the case of patient HM endure as to the primary and critical role of medial temporal lobe structures in declarative memory.