Associative and perceptual learning and the concept of memory systems

https://doi.org/10.1016/S0926-6410(96)00042-0Get rights and content

Abstract

An introductory review is followed by some new experimental data and a final discussion. The primate temporal lobe contains multiple qualitatively distinct memory systems. The functional properties of these memory systems can be explained by reference to the nature of the afferent information which they process, rather than by reference to any putative specialization in memory processing. In this way, the plasticity of `memory systems' in associative memory is probably similar in principle to the plasticity of `perceptual systems' in perceptual learning. Therefore, it is important to consider the relationship between perceptual and associative learning. Two experiments investigated perceptual learning in the Rhesus monkey (Macaca mulatta). Substantial perceptual learning was observed both with complex scenes and with simple colours. Two hypotheses as to the basis of perceptual learning are discussed. A physiological hypothesis is that training with a particular set of stimuli expands the cortical representation of those stimuli. This can explain the effects in both experiments. A psychological hypothesis is that perceptual learning is produced by learned associations among the multiple features of complex stimuli. This can explain the effects in Expt. 1 but not in Expt. 2. The psychological associative hypothesis is therefore redundant. Furthermore, associative learning can itself be viewed as an expansion of the cortical representation of a complex event. Thus, the distinction between perceptual and memory systems will need to be abandoned as deeper understanding of cortical plasticity is achieved.

Introduction

At the cellular level, the capacity for long-lasting functional changes is widespread in the central nervous system. Analysis of the cognitive effects of brain damage shows, however, that in some ways memory appears to be not simply a diffuse property of nervous tissue, but instead a specialized function of discrete anatomical systems in the brain. The evidence in favour of the concept of memory systems in the brain came in the first place from patients with acquired memory disorder (`organic amnesia'). Memory disorder is often produced as one component of a general loss of mental ability (`dementia'), but in some patients amnesia is a discrete disorder, leaving intact many other cognitive abilities such as reasoning, perception and intelligence. Relatively pure amnesia of this kind is caused by relatively circumscribed brain damage, as opposed to the diffuse damage which is seen in dementia. Neuropathological analysis of the brain damage responsible for producing discrete amnesia suggests that the critical lesion for producing this disorder is any bilateral interruption of a system of interconnected structures which was first clearly identified by Delay and Brion in 1969 [2]. The Delay-Brion system consists of the hippocampal formation in the medial temporal lobe, including the subiculum as well as the hippocampus proper; the fimbria and fornix; the mamillary bodies; the mamillothalamic tract; the anterior group of thalamic nuclei; and the cingulate bundle and cingulate cortex. In different patients, amnesia can be caused by different lesions which are spatially quite distant from each other in the brain: for example, by temporal lobe infarctions due to posterior cerebral artery occlusion in one patient, and in another patient by lesions to the mamillary bodies, adjacent to the midline and far from the temporal lobe. Delay and Brion proposed that this apparent heterogeneity of the causative lesion in amnesia could be explained by the neural connections of the structures involved. For example, the subiculum of the hippocampus sends a heavy projection to the mamillary bodies through the fornix, and interruption of the fornix also causes amnesia 9, 21.

Modern evidence on the neuropathology of amnesia has largely supported the interpretation put forward by Delay and Brion 9, 19, 21. However, since their time our understanding of memory systems in the brain has significantly advanced in two important ways, which are further discussed below. Experimental work with nonhuman primates has given a clear picture, a much clearer picture than clinical evidence alone can give, as to the effects of particular localized brain lesions upon memory. At the same time, psychological analysis of these effects has shown that the type of memory deficit which is seen in the organic amnesic syndrome is only one kind of memory impairment, and that impairments in other kinds of memory can follow lesions outside the Delay-Brion system.

The functional relationship between brain systems and memory is revealed only imperfectly in clinical material. The interpretation of such material has necessarily remained to some extent controversial or open to doubt. For example, the causative lesions are produced by the vagaries of disease and trauma; post mortem neuropathological analysis does not necessarily reveal the whole extent of functional disorder; and the consequent memory impairments must be assessed against the background of wide individual variation in memory ability, in patients whose pre-morbid memory ability is usually unknown. For these and other reasons, a considerable effort has been devoted to the investigation of experimentally produced amnesia in nonhuman primates.

The results of early studies were puzzling in that macaque monkeys (Macaca mulatta and M. fascicularis) appeared to show little impairment in some simple tests of object memory after experimentally produced ablations in the Delay-Brion system. However, more recent experiments, investigating a wider range of memory tasks, have shown that these ablations produce a severe impairment in monkeys' memory for complex spatially organized scenes. These impairments are revealed in a wide variety of tasks, including spatial memory tasks in mazes [24], memory for the location of hidden food rewards [10], memory for complex naturalistic scenes such as frames from a cinema film [6], and memory for artificial scenes generated by a formal algorithm on a computer-driven display [8]; this last kind of experiment is particularly valuable in allowing the effects of brain lesions on memory for objects, backgrounds and spatial organization to be separately analysed. After interruption of the Delay-Brion system monkeys are not impaired in remembering objects independently of the scenes in which the objects were presented, but they are impaired in scene-specific object memory. In terms of cognitive psychology, the `episodic memory' in which amnesic patients are severely impaired – that is, memory for specific personally experienced events – is analogous to the monkeys' memory for experienced events involving some specific objects set in a specific background or scene. This type of memory for complex specific events is contrasted with `semantic memory' – that is, acquired general knowledge about objects, independent of any specific background or any specific discrete event. The experimental and clinical evidence indicates that the Delay-Brion system is specially involved in memory for events, as opposed to general knowledge which is not specific to some discrete event in a particular scene. This discovery leads naturally on to the idea that some other memory system or systems, independent of the Delay-Brion system, might be responsible for the kind of memory which is manifested in general knowledge about objects, independent of any one specific scene.

Patients with lesions in the neocortex of the anterior temporal lobe, sparing the Delay-Brion system, can show a syndrome which is in some respects complementary to the classical amnesic syndrome. Unlike amnesic patients, these patients do not show striking impairments in remembering discrete events such as the visit of a family member or some particular meal; but, again unlike amnesic patients, they are deficient in general knowledge (`semantic memory') as tested, for example, by the ability to identify and describe a family member or a well-known food item [17]. The anatomical basis of this type of impairment has been elucidated by experiments with macaque monkeys.

Monkeys with discrete ablation of the perirhinal cortex (a strip of cortex in the anterior temporal lobe, lateral to the rhinal sulcus) show disorders of knowledge about objects. These disorders are revealed in discrimination learning tasks, in which the animal learns for food reward to choose the objects which the experimenter has arbitrarily designated as correct, in preference to other objects which the experimenter has designated as incorrect [7]; a similar disorder is also revealed in the matching-to-sample task with objects, in which the monkey learns to choose the object which is identical to a given sample object [12]. Both of these impairments occur when the animal has to deal with a large population of objects, requiring an accurate specification of individual objects in order to perform the task correctly; the impairments in the same tasks disappear if only one pair of objects is repeatedly used throughout the experiment [3]. This kind of impairment was first discovered in experiments with delayed matching (or nonmatching) to sample, and with large ablations of the medial temporal lobe, involving not only the perirhinal cortex but also the hippocampus and amygdala [23]. It is now known, however, that damage to the perirhinal cortex alone, sparing the hippocampus and amygdala, produces in these tasks an impairment of almost equal severity to that produced by large medial temporal ablations [22]. It is also now known that the delay component of delayed matching-to-sample, as tested in early experiments, is not crucial since an impairment is also seen in matching-to-sample with no delay [3].

The memory impairment produced by discrete perirhinal cortex ablation in the monkey is qualitatively dissimilar from that which follows discrete interruption of the Delay-Brion system, by fornix transection for example. Monkeys with fornix transection are severely impaired in memory for the spatial organization of scenes, and only mildly impaired in matching-to-sample with objects, whereas monkeys with perirhinal cortex ablations show the opposite pattern [7]. Thus, there is clear evidence for at least two distinct specialized memory systems in the primate temporal lobe, and the functions of these discrete anatomical structures correspond, at least in part, to the discrete cognitive functions of episodic memory and object knowledge.

The Delay-Brion system and the perirhinal cortex are two of the most important and best understood memory systems of the primate brain, but there are also other identifiable systems with specialized functions, including the amygdala and the prefrontal cortex [7]. In general, therefore, the primate brain should be thought of as containing multiple (that is, more than two) specialized memory systems. Memory has frequently been classified in a binary way, for example into `memory' and `habit'; but any such binary classification appears to be too simple as a summary of the multiple diversity of the memory systems in the primate brain.

The developments reviewed above may seem straightforward and uncontroversial. On the contrary, however, there has existed a degree of tension between two different conceptual approaches to the study of memory mechanisms. The approach taken above is that of cognitive neuroscience. In this approach, multiple specialized cognitive mechanisms are supposed to exist in the brain, and the task is to identify those systems. In this task, cognitive neuroscientists have given themselves a free hand in hypothesis formulation, and the number of proposed specialized memory mechanisms such as `recognition memory', `episodic memory', `working memory', `declarative memory', and so on is therefore far too large to review here. The experimental basis for these proposals has been the discovery of double dissociations between different kinds of memory impairment. At its logical extreme, this approach has been taken to imply that, if one can find two neurological patients and two memory tasks such that patient A is impaired on task 1 but not task 2, while patient B is impaired on task 2 but not task 1, then one can announce the discovery of a new dissociation among memory systems, named for whatever cognitive process appears to be specifically required by the two tasks. For example, within the realm of semantic memory it is widely believed that knowledge about living things (plants and animals) is an anatomically separate system from knowledge about nonliving things (human artifacts).

The opposite approach is that of neurobiology. According to this view, most recently expounded in a lucid and forceful review by Vanderwolf and Cain [31], the experimentalist's task is not to look for specialized memory mechanisms in the brain, but rather to look for continuity between memory or knowledge, on the one hand, and other biological mechanisms of behaviour, such as instinct and innate reflexes. The expectation should be, according to this view, that the human brain should perform what we call memory or learning by some mechanism that is similar to other adaptive mechanisms, since many examples of learning capability in nonhuman animals can be seen as "a refinement and further development of instinctive behavior" ([31], p. 265) and neural plasticity can be viewed as a general property of nervous systems. Thus, according to this view, one should be sceptical about the idea that specialized memory systems exist in the human brain.

There can be no doubt that the scepticism advocated by Vanderwolf and Cain [31]is justified up to a certain point. Merely the discovery that patient A is impaired on task 1 but not task 2, while patient B is impaired on task 2 but not task 1, is surely insufficient to warrant the claims of separate brain systems for tasks 1 and 2. Rather, one should first look for more parsimonious explanations of the data. For example, the fact that impairments of semantic memory are often more severe in the realm of living things than in the realm of nonliving things might simply result from the fact that living things are more confusable with each other, on average, than nonliving things are [11]. However, one still needs to explain the fact that some specific brain lesions produce specific memory impairments, as reviewed above. Contrary to the claims made by Vanderwolf and Cain ([31], pp. 272–273), it is beyond doubt that amnesia and other kinds of specific memory impairment are reliably produced by specific brain lesions and not by diffuse brain damage.

It seems that some compromise is required between the theoretical prolixity of cognitive neuroscience and the scepticism of the neurobiological approach. One such compromise is to accept that plasticity is a general feature of neural systems, and thus to deny that memory systems have any special role in memory processes such as consolidation or retrieval. Instead, according to this compromise view, the multiple specializations of memory systems arise from the different kinds of afferent information which they receive. To store and retrieve episodic or whole-scene memories, which as we have seen involves associative memory for multiple aspects of specific events, requires the convergence of many kinds of afferent information from different modalities, including spatial information. The hippocampus–fornix system may thus owe its specialization in episodic memory to the fact that it receives multiple inputs from high-order polymodal cortical areas. Similarly, the perirhinal cortex may owe its specialization in object individuation to the fact that it receives multimodal information about objects but not spatial information. Of course, one might equally argue that the evolution of the anatomical connectivity of the primate hippocampus has been driven by the adaptive benefits conferred by whole-scene or episodic memory. The important point remains, that perhaps the specialization of memory systems is conferred hodologically, that is by specialization of their anatomical connections to other structures, rather than by any specialization in specific memory processes such as retrieval, consolidation or working memory.

If this hodological hypothesis of memory systems is true then one should expect that the plasticity of memory systems, the intrinsic memory processing which operates universally upon the different kinds of information that are determined hodologically in qualitatively different memory systems, should be similar to the plasticity of other brain systems which are not thought of as memory systems, but instead as perceptual systems, for example. One clear example of plasticity outside of memory systems is perceptual learning. It is therefore important to consider the relationship between associative memory and perceptual learning.

In associative learning, improvement in performance is attributed to the acquisition of an association between some stimulus that acts as a retrieval cue and some separate event with which that stimulus is correlated in time. Perceptual learning, by contrast, is said to occur when an improvement in performance cannot be attributed to any such association formed to a stimulus, but results instead from an improvement in the perceptual processing of the stimulus itself. One way to demonstrate such an improvement is by physiological measurement of the neuronal activity in sensory pathways. For example, monkeys given extensive practice in a tactile discrimination, in which one particular digit of one hand was always used to make the discrimination, showed an expanded representation of that digit in the somatosensory cortex [27], and human Braille readers have a similar expanded somatosensory representation of the digit used for reading [25]. Another way to demonstrate perceptual learning is by behavioural measures. For example, stimuli which have benefited from a perceptual learning process support unusually rapid learning in a subsequent associative learning task, presumably because of their enhanced discriminability [14]. The connection between these two approaches to the study of perceptual learning, the physiological and the psychological, has rarely been explored. In a full review of psychological treatments of perceptual learning, Hall [15]made no mention of any physiological measurements of it.

McLaren et al. [20]put forward an explanation of perceptual learning in terms of associative learning. The associative learning that takes part during perceptual learning, according to this view, is not the acquisition of associations between a stimulus and some separate event, but the acquisition of associations among the elements or features of the stimulus itself. According to this account, when an animal repeatedly perceives two different stimuli A and B, the elements of A become associated with each other, the elements of B become associated with each other, elements common to A and B receive more latent inhibition than elements present in only one of the stimuli, and inhibitory associations are formed between elements that are present in one of the stimuli but not in the other. Each of these three associative changes that are intrinsic to the stimuli A and B has the effect that the stimuli will support more rapid extrinsic differential associative learning when, in a subsequent phase, they are associated with different events such as A with food and B with no food, as compared with a control pair of stimuli C and D which had not benefitted from perceptual learning. Thus, perceptual learning is explained as a product of a particular example of associative learning, that is, the formation of associations between the elements of stimuli.

McLaren et al.'s account is particularly plausible as an explanation of perceptual learning with complex stimuli such as the extra-maze cues in a spatial learning experiment with rats [28]. Intuitively, the stimuli to which a rat is exposed in the arm of an open radial maze contain many independent elements which can enter into association with each other. The purpose of the first of the present experiments (Expt. 1a) was to measure perceptual learning in the monkey with similarly complex stimuli, composed of multiple independent elements. The stimuli were artificially generated complex scenes [8]. Each scene consisted of a multi-part background which was unique to that scene, together with two foreground objects which occupied constant spatial positions within that background. The monkey's task was to choose one of the two foreground objects by touching it. In each daily session of training the monkeys learned, concurrently in 20 such scenes, which of the two objects in each scene was correct (rewarded with food) and which was incorrect (unrewarded). However, the determination of correct and incorrect objects by the experimenter changed randomly from day to day. Thus, if the monkey's learning proficiency improved from day to day when the same scenes were used every day, this improvement could not be attributed to the extrinsic associations of the foreground objects with food and no food (which changed randomly from day to day), but instead could be attributed to perceptual learning about the scenes, which according to McLaren et al. consists in the acquisition of associations among features of the scenes. To check that this improvement was specific to the particular scenes that the animals had experienced, rather than some general improvement in the strategy of learning, the set of scenes which each animal learned about was changed every four days. A further experiment along similar lines (Expt. 1b) was designed to test whether experience with particular individual scenes could lead to improvement specific to those individual scenes, as opposed to the improvement in a whole set of scenes which was measured in Expt. 1a.

However, McLaren et al.'s account is less intuitively attractive if applied to perceptual learning with simple stimuli. Expt. 2 therefore explored monkeys' perceptual learning with colours. Within each day the animals learned which of two colours was rewarded that day. As in Expt. 1, the reward associations of the stimuli could not explain between-day improvement when the same colours were used every day, since the reward associations changed from day to day. To check that the improvement was specific to the trained colours, rather than reflecting some general improvement in the strategy of learning, the animals were subsequently switched to a different pair of colours. For example, one monkey became expert at learning about colours that were almost pure reds, and was then switched to colours that were almost pure blues, while another monkey first became a green expert and was then switched to reds, and so on. If the within-day learning rate suffers from the switch, perceptual learning can be inferred. It is not clear intuitively what the multiple features of red or green are, unlike the multiple features of a complex scene in Expt. 1.

Section snippets

Subjects

These were three experimentally naive adult male Rhesus monkeys (Macaca mulatta). Before beginning the main experiment as described below, each monkey was trained by conventional procedures to touch foreground objects in scenes of gradually increasing complexity (for definition of foreground objects and scenes, see below). Scenes used in this preliminary training were different from any scenes used in the main experiment.

Apparatus

The monkey was brought to the training apparatus in a wheeled transport

Subjects, apparatus, and stimulus material

These were the same as in Expt. 1a. New scenes were used. The animals proceeded to Expt. 1b shortly after completing Expt. 1a, with no other intervening training.

Procedure

Six new sets of scenes were presented one after another to each animal. For animal S1 there were 20 scenes in each set. For animals S2 and S3 there were 50 scenes in each set, because the error rates of these two animals had been rather low with only 20 scenes in each set (as shown in panel C of Fig. 1), threatening the measurement

Subjects

These were three adult male Rhesus monkeys. One of the subjects from Expts. 1a and 1b (S3) performed the present experiment immediately after completing the previous experiment, and the remaining two subjects in the present experiment were experimentally naive (S4 and S5). Before beginning the main experiment as described below the two naive subjects were trained by conventional procedures to touch visual stimuli displayed on the touchscreen.

Apparatus

This was the same as in Expts. 1a and 1b.

Stimulus material

The stimuli

General discussion

The results of Expt. 1a (Fig. 1) showed that the monkeys improved their within-day learning rate substantially when the same set of scenes was used for four successive days. One might have predicted that a between-day improvement would be seen particularly with scenes in which the animal's initial preference on any day was for the object that happened to be correct on that day, but there was no sign of a difference in the between-day improvement according to the animal's initial preference

Acknowledgements

This research was supported by the Medical Research Council.

References (31)

  • Bessemer, D.W. and Stollnitz, F., Retention of discriminations and an analysis of learning set. In A.M. Schrier and F....
  • Delay, J. and Brion, S. Le Syndrome de Korsakoff, Masson, Paris,...
  • Eacott, M.J., Gaffan, D. and Murray, E.A., Preserved recognition memory for small sets, and impaired stimulus...
  • Fiorentini, A., Ghez, C. and Maffei, L., Physiological correlates of adaptation to a rotated visual field, J. Physiol.,...
  • Gaffan, D., Monkeys' recognition memory for complex pictures and the effect of fornix transection, Q. J. Exp. Psychol.,...
  • Gaffan, D., Amnesia for complex naturalistic scenes and for objects following fornix transection in the Rhesus monkey,...
  • Gaffan, D., Dissociated effects of perirhinal cortex ablation, fornix transection and amygdalectomy: evidence for...
  • Gaffan, D., Scene-specific memory for objects: a model of episodic memory impairment in monkeys with fornix...
  • Gaffan, D. and Gaffan, E.A., Amnesia in man following transection of the fornix: a review, Brain, 114 (1991)...
  • Gaffan, D. and Harrison, S., Place memory and scene memory: effects of fornix transection in the monkey, Exp. Brain...
  • Gaffan, D. and Heywood, C.A., A spurious category-specific visual agnosia for living things in normal human and...
  • Gaffan, D. and Murray, E.A., Monkeys (Macaca fascicularis) with rhinal cortex ablations succeed in object...
  • Gaffan, D., Saunders, R.C., Gaffan, E.A., Harrison, S., Shields, C. and Owen, M.J., Effects of fornix transection upon...
  • Gibson, E.J. and Walk, R.D., The effect of prolonged exposure to visually presented patterns on learning to...
  • Hall, G., Perceptual and Associative Learning, Oxford University Press, Oxford,...
  • Cited by (0)

    View full text