Implicit memory stores forms of knowledge that are typically acquired without conscious effort and which guide behavior unconsciously. Priming is a type of implicit memory that operates in amnesic patients as well as healthy subjects, suggesting that it does not depend on medial temporal lobe structures.
Two types of priming have been proposed. Conceptual priming provides easier access to task-relevant semantic knowledge because that knowledge has been used before. It is correlated with decreased activity in left prefrontal regions that subserve initial retrieval of semantic knowledge. In contrast, perceptual priming occurs within a specific sensory modality, and according to Tulving and Schacter it depends on cortical modules that operate on sensory information about the form and structure of words and objects.
Damage to unimodal sensory regions of cortex impairs modality-specific perceptual priming. For example, one patient with an extensive lesion of the right occipital lobe failed to demonstrate visual priming for words but had normal explicit memory (Figure 65–9). This condition is the reverse of that found in amnesic patients such as H.M., and provides further evidence that the neural mechanisms of priming are distinct from those for explicit memory.
The right occipital cortex is required for visual priming for words.
(Adapted with permission, from Vaidya et al. 1998.)
A. Structural magnetic resonance imaging (MRI) depicts the near complete removal of the right occipital cortex in a patient, M.S., who suffered pharmacologically intractable epilepsy with a right occipital cortical focus.
B. Font-specific priming is intact in amnesic patients (AMN) and their controls as well as in the controls for patient M.S., but not in M.S. himself. Font-specific priming is a form of visual priming in which the individual is better able to identify a briefly flashed word when the type font is identical to an earlier presentation, compared to identification when the font is different (priming equals performance when the font is the same minus performance when the font is different). The patient M.S. has normal explicit memory, even for visual cues (data not shown), but lacks implicit memory for specific properties of visually presented words.
Visual priming is almost always correlated with decreased activity in higher-order visual (extrastriate) areas of cortex. Randy Buckner and his colleagues using fMRI found that activity in extrastriate cortex was greater during the initial exposure to an object than when the object was presented again later. These findings parallel the finding that activity in the left prefrontal cortex is reduced during conceptual priming. Most tasks include both perceptual and conceptual priming, and there probably are no sharp distinctions between the two.
Other forms of nondeclarative memory subserve the learning of habits, the learning of motor, perceptual, and cognitive skills, and the formation and expression of conditioned responses. In general, these forms of implicit memory are characterized by incremental learning, which proceeds gradually with repetition. The neural circuits that initiate habit, motor skill, and conditioned learning are independent of the medial temporal lobe system responsible for explicit memory. For example, H.M. is able to acquire new visuomotor skills, like the mirror-tracing task (see Figure 65–3).
New perceptual, motor, or cognitive abilities are also learned through repetition. With practice, performance becomes more accurate and faster, and these improvements generalize to learning novel information. Skill learning moves from a cognitive stage, where knowledge is represented explicitly and the learner must pay a great deal of attention to performance, to an autonomous stage, where the skill can be executed without much conscious attention. As an example, driving a car initially requires that one pay attention to each component of the skill, but after practice one no longer attends to the individual components.
The learning of sensorimotor skills depends in part on the basal ganglia, cerebellum, and neocortex. Dysfunction of the basal ganglia in patients with Parkinson and Huntington disease impairs learning of motor skills. Patients with cerebellar lesions also have difficulties acquiring some motor skills, and functional imaging of healthy individuals during sensorimotor learning shows changes in the activity of the basal ganglia and cerebellum. Finally, skilled behavior can depend on structural changes in motor neocortex, as seen by the expansion of the cortical representation of the fingers in musicians (see Chapter 67).
Perceptual learning improves the ability to make sense of novel sensory inputs, as in learning to read mirror-reversed text or recognizing novel objects by reference to familiar categories. Amnesic patients with damage to the medial temporal lobe can learn to read mirror-reversed text but this learning is mildly impaired in Huntington disease and variably impaired in Parkinson disease. Patients with cerebellar lesions have no difficulty with perceptual learning, even though the learning of motor skills is impaired.
A neuroimaging study by Russell Poldrack and his colleagues suggests that extensive practice with mirror reading produces a shift in the parts of the brain involved in the task. In this study performance of the mirror-reading task before practice was correlated with activity in ventral visual processing regions as well as extensive activity in the parietal cortex. After practice, activity decreased in the parietal cortex but increased in the left inferior temporal cortex, a region associated with representing visual form (Figure 65–10). These results reflect a transition from having to mentally rotate the mirror-reversed words to the ability to read directly the reversed letters. Different neural processes are involved once skilled performance moves from the cognitive to the autonomous stage. Similar neural changes have been observed in imaging studies of motor and visual-motor skill learning.
Perceptual learning involves a shift from cognitive to autonomous stages that use different neural pathways.
Subjects are asked to read mirror-reversed text, something most people rarely encounter. Prior to training individuals rely on the parietal cortex (red arrow) and to a lesser extent the inferior temporal cortex (white arrow). After extensive training the processing pathways involved in the task appear to be different. Individuals rely less on the parietal cortex and more on the inferior temporal cortex. (Reproduced, with permission, from Poldrack et al. 1998.)
Implicit memory also underlies habit learning or Pavlovian associative conditioning, the gradual learning about the predictive relationship between a stimulus and a response (discussed later). Habit learning in humans has been studied using the probabilistic classification task, where subjects attempt to predict accurately one of two possible outcomes based on the presentation of a set of cues, with each cue having a probabilistic relation to each outcome. For example, subjects may be asked to predict the weather (rain or sunshine) based on a set of cue cards (Figure 65–11).
Learning predictive relationships involves the neostriatum.
A. Subjects are instructed to predict whether the weather will be rain or sunshine based on a set of cue cards. Each cue card has a probabilistic relation to each weather outcome (eg, predicting sunshine either 75, 57, 43, or 25% of the time). Subjects attempt to learn these relations during training and they are told after each trial whether their prediction is correct or incorrect.
B. Performance on the prediction task across the first 50 training trials is plotted on the left; performance results on a declarative memory test are shown on the right. Amnesic patients (AMN) initially learn the prediction task at the same rate as healthy control subjects, although their performance on the declarative memory task is impaired. By contrast, patients with Parkinson disease (PD), who suffer impairments in basal ganglia function, perform poorly on the prediction task but perform as well as controls on the declarative memory task. PD* identifies a subgroup of the Parkinsonian patients with the most severe symptoms. (Reproduced, with permission, from Knowlton, Mangels, and Squire 1996.)
Because the associations between the cues and outcomes are probabilistic, thus requiring numerous trials to learn, explicit (conscious) memory of specific trials is not as useful for successful performance as the gradual accumulation of knowledge about the stimulus-outcome associations. Barbara Knowlton and colleagues have shown that, in contrast to patients with medial temporal lobe lesions, patients with basal ganglia disorders are severely impaired in this task.
Implicit Memory Can Be Associative or Nonassociative
Our consideration of implicit memory has so far focused on humans. But some forms of implicit memory can also be studied in nonhuman animals, and animal studies have distinguished two types of implicit memory: nonassociative and associative. With nonassociative learning an animal learns about the properties of a single stimulus. With associative learning the animal learns about the relationship between two stimuli or between a stimulus and a behavior.
Nonassociative learning results when a subject is exposed once or repeatedly to a single type of stimulus. Two forms of nonassociative learning are common in everyday life: habituation and sensitization. Habituation, a decrease in a response, occurs when a benign stimulus is presented repeatedly. For example, most people in the United States are startled when they first hear the sound of a firecracker on Independence Day, but as the day progresses they become accustomed to the noise and do not respond. Sensitization (or pseudo-conditioning) is an enhanced response to a wide variety of stimuli after the presentation of an intense or noxious stimulus. For example, an animal will respond more vigorously to a mild tactile stimulus after receiving a painful pinch. Moreover, a sensitizing stimulus can override the effects of habituation, a process called dishabituation. For example, after the startle response to a noise has been reduced by habituation, one can restore the intensity of response to the noise by delivering a strong pinch.
With sensitization and dishabituation the timing of stimuli is not important because no association between stimuli must be learned. In contrast, with two forms of associative learning the timing of the stimuli to be associated is critical. Classical conditioning involves learning a relationship between two stimuli, whereas operant conditioning involves learning a relationship between the organism's behavior and the consequences of that behavior.
Classical Conditioning Involves Associating Two Stimuli
Classical conditioning was first described at the turn of the century by the Russian physiologist Ivan Pavlov. The essence of classical conditioning is the pairing of two stimuli. The conditioned stimulus (CS), such as a light, a tone, or a touch, is chosen because it produces either no overt response or a weak response usually unrelated to the response that eventually will be learned. The reinforcement, or unconditioned stimulus (US), such as food or a shock, is chosen because it normally produces a strong and consistent response (the unconditioned response), such as salivation or withdrawal of the leg. Unconditioned responses are innate; they are produced without learning. Repeated presentation of a CS followed by a US gradually elicits a new or different response called the conditioned response.
One way of explaining conditioning is that repeated pairing of the CS and US causes the CS to become an anticipatory signal for the US. With sufficient experience an animal will respond to the CS as if it were anticipating the US. For example, if a light is followed repeatedly by the presentation of meat, eventually the sight of the light itself will make the animal salivate. Thus classical conditioning is the way an animal learns to predict events.
The probability of occurrence of a conditioned response decreases if the CS is repeatedly presented without the US. This process is known as extinction. If a light that has been paired with food is later repeatedly presented in the absence of food, it will gradually cease to evoke salivation. Extinction is an important adaptive mechanism; it would be maladaptive for an animal to continue to respond to cues that are no longer meaningful to it. The available evidence indicates that extinction is not the same as forgetting, but that something new is learned––the CS now signals that the US will not occur.
For many years psychologists thought that classical conditioning resulted as long as the CS preceded the US by a critical time interval. According to this view, each time a CS is followed by a US (reinforcing stimulus) a connection is strengthened between the internal representations of the stimulus and response or between the representations of one stimulus and another. The strength of the connection was thought to depend on the number of pairings of CS and US.
A substantial body of evidence now indicates that classical conditioning cannot be adequately explained simply by the fact that two events or stimuli occur one after the other (Figure 65–12). Indeed, it would not be adaptive to depend solely on sequence. Rather, all animals capable of associative conditioning, from snails to humans, remember actual relationships rather than simply sequential events. Thus classical conditioning, and perhaps all forms of associative learning, enables animals to distinguish events that reliably occur together from those that are only randomly associated.
Classical conditioning depends on the degree to which two stimuli are correlated.
In this experiment on rats a tone (the conditioned stimulus or CS) was paired with an electric shock (the unconditioned stimulus or US) in four out of 10 of the trials (red ticks). In some trial blocks the shock was presented without the tone (green ticks). The degree of conditioning was evaluated by determining how effective the tone alone was in suppressing lever-pressing to obtain food. Suppression of lever-pressing is a sign of a conditioned defensive response, freezing. (Adapted, with permission, from Rescorla 1968.)
A. Maximal conditioning occurred when the US was presented only with the CS.
B–C. Little or no conditioning was evident when the shock occurred without the tone as often as with it (40%). Some conditioning occurred when the shock occurred 20% of the time without the tone.
Lesions in several regions of the brain affect classical conditioning. A well-studied example is conditioning of the protective eyeblink reflex in rabbits, a form of motor learning. A puff of air to the eye naturally causes an eyeblink. A conditioned eyeblink can be established by pairing the puff with a tone that precedes the puff. The conditioned response (an eyeblink in response to a tone) is abolished by a lesion at either of two sites. Damage to the vermis of the cerebellum abolishes the conditioned response but does not affect the unconditioned response (eyeblink in response to a puff of air). Interestingly, neurons in the same area of the cerebellum show learning-dependent increases in activity that closely parallel the development of the conditioned behavior. A lesion in the interpositus nucleus, a deep cerebellar nucleus, also abolishes the conditioned eyeblink. Thus both the vermis and the deep nuclei of the cerebellum play an important role in conditioning the eyeblink and perhaps other simple forms of classical conditioning involving skeletal muscle movement.
Operant Conditioning Involves Associating a Specific Behavior with a Reinforcing Event
A second major paradigm of associative learning, discovered by Edgar Thorndike and systematically studied by B. F. Skinner and others, is operant conditioning (also called trial-and-error learning). In a typical laboratory example of operant conditioning a hungry rat or pigeon is placed in a test chamber in which the animal is rewarded for a specific action. For example, the chamber may have a lever protruding from one wall.
Because of previous learning, or through play and random activity, the animal will occasionally press the lever. If the animal promptly receives a positive reinforcer (eg, food) after pressing the lever, it will begin to press the lever more often than the spontaneous rate. The animal can be described as having learned that among its many behaviors (for example, grooming, rearing, and walking) one behavior is followed by food. With this information the animal is likely to press whenever it is hungry.
If we think of classical conditioning as the formation of a predictive relationship between two stimuli (the CS and the US), operant conditioning can be considered as the formation of a predictive relationship between an action and an outcome. Unlike classical conditioning, which tests the responsiveness of a reflex to a stimulus, operant conditioning tests behavior that occurs either spontaneously or without an identifiable stimulus. Operant behaviors are said to be emitted rather than elicited. In general, actions that are rewarded tend to be repeated, whereas actions followed by aversive, although not necessarily painful, consequences tend not to be repeated. Many experimental psychologists feel that this simple idea, called the law of effect, governs much voluntary behavior.
Because operant and classical conditioning involve different kinds of association—an association between an action and a reward or between two stimuli, respectively—one might suppose the two forms of learning are mediated by different neural mechanisms. However, because the laws of operant and classical conditioning are quite similar, the two forms of learning may use the same neural mechanisms. For example, timing is critical in both. In operant conditioning the reinforcer usually must closely follow the operant action. If the reinforcer is delayed too long, only weak conditioning occurs. Similarly, classical conditioning is generally poor if the interval between the conditioned and unconditioned stimuli is too long or if the unconditioned stimulus precedes the conditioned stimulus.
Associative Learning Is Constrained by the Biology of the Organism
Animals generally learn to associate stimuli that are relevant to their survival. For example, animals readily learn to avoid certain foods that have been followed by a negative reinforcement (eg, nausea produced by a poison), a phenomenon termed taste aversion.
Unlike most other forms of conditioning, taste aversion develops even when the unconditioned response (poison-induced nausea) occurs after a long delay, up to hours after the CS (specific taste). This makes biological sense, because the ill effects of infected foods and naturally occurring toxins usually follow ingestion only after some delay. For most species, including humans, taste-aversion conditioning occurs only when certain tastes are associated with illness. Taste aversion develops poorly if a taste is followed by a painful stimulus that does not produce nausea. Animals do not develop an aversion to a visual or auditory stimulus that has been paired with nausea.