['Animal Thought' © Stephen Walker 1983]
pdf file of this chapter
8 Memory—sustained and revived perceptions
Human memory is inextricably bound up with verbal reports and commentaries. It is rare to think that one remembers something without being able to say anything at all about it, and the experimental study of human memory can almost be defined as the measurement of verbal or quasi-verbal responses. (Someone can be asked to press a button instead of saying ‘yes, I remember seeing that picture before’, but if they could have said the phrase, the button-push is more than just a motor reaction.) Because of this, it may cause confusion to discuss the retention of sense impressions in animal nervous systems in the same terms as we apply to linguistically expressed recollections and judgments. But on balance, I would rather risk misunderstandings of this kind than to adopt the artificial alternative of applying new labels to all cases where past experience achieves memory-like status in unspeaking creatures.
I ought to make it clear, though, that I do not intend to attribute all effects of past experience on future behaviour to ‘memories’. Relatively clear exceptions are learned perceptual- motor skills and automatic habits. Our skill, or lack of it, at tying our shoes, getting dressed, or washing up, to say nothing of playing musical instruments, may demonstrably depend on accumulated previous practice without our necessarily being able to remember anything specific about important episodes of previous experience. There are plenty of cases of coordination of muscles, as important components of skilled movements, which we not only cannot remember, but do not know are taking place at the time. Similarly, there are components of perceptual skills, such as reading, or recognising speech, which are learned through experience, but are never available to introspection. And a dogma common to many forms of psychotherapy is that emotional adjustment,
or lack of it, is a consequence of prior experiences that are not available as memories in the normal way. It is always necessary to take account of the possibility that animal behaviour which reflects past experience is doing so by similar unconscious processes, or by even more rudimentary mechanisms that ought to be distinguished from those requiring more interesting forms of the internalisation of knowledge.
Rudimentary mechanisms which would manifest learning by experience in the absence of anything we should wish to call memory have been discussed in Chapter 3 as ‘stimulus- response’ devices. We can imagine a very simple animal that withdraws into its shell as a reflex response to the detection of vibration, and detects flashing lights without normally withdrawing in response to them. If such an animal were to be lightly jiggled once every minute, it would, if it were anything like a real snail, withdraw to begin with, but gradually cease to do so, after repeated jigglings. This in itself is a form of profiting from experience and retention of information, but the waning of a response to a repeated stimulus (known as ‘habituation’) can be interpreted as a temporary modification of a simple neural connection between sensory input and response output. As an extreme case, one may assume that a single sensory neuron, which functions as a jiggle-detector, synapses with a single motor neuron, which activates withdrawal (cf. Horn, 1967; discussed in Hinde, 1970; and Gray, 1975; also Kandel, 1974, for experimental work on the isolated ganglions and withdrawal reflexes in the sea-slug Aplysia). If this circuit becomes less sensitive as a function of use, the phenomenon of habituation would be mimicked. It could be said that the neural changes involved coded past experience, but the representation is confined to the stimulus-response circuit, and is not available when this circuit itself is not being used.
It would be consistent with what usually happens in real animals if our imaginary creature always withdrew into its shell in response to a very vigorous jiggle. Now suppose that it gets a spaced-out sequence of these very vigorous vibrations in which every jiggle is preceded by flashing lights; and, although it does not normally withdraw in response to flashing lights, we now find that it does so. This looks like a Pavlovian conditioned reflex, and some would say that our creature possessed a ‘capacity for associative learning’, in common with Pavlov’s dogs. But provided that a neuron from a light detector reaches the single synapse in the hypothetical jiggle-withdraw circuit, one could put forward the theory that conjunction of activity at the end of the light detector neuron with transmission across the synapse
due to a jiggle has somehow welded the light detector into the circuit, so that the light detector can operate the withdrawal neuron by itself. In this case one would hardly want to claim that the animal had any independent memory of the light- jiggle pairing, or was subject to jiggle expectancies when exposed to flashing lights.
To make life more complicated for this simple animal we might poke small pieces of food under its shell, but only when it had just withdrawn there after faint jiggles which would usually be ignored. If it now began to withdraw very promptly in response to mild jiggles, when it had been for some time without any food, we could argue that we had demonstrated instrumental learning and would have to construct a slightly more elaborate theory about its nervous system. Following Thorndike and Hull (see Chapter 3) we could suppose that our animal was equipped with a food detector, and that this reached back to the jiggle-withdraw synapse in such a way that the activation of the food detector just after the synapse had fired ‘stamped in’ the jiggle withdraw connection. To explain any decline in responsiveness when the animal was full of food, or when we stopped giving it food for withdrawing, we should have to add in extra inhibitory mechanisms, but clearly we could account for the changes in the creature’s behaviour, caused by its encounters with food, in terms of reflex-like response tendencies, without anything corresponding to a memory, or to an internal description of past experience.
I concluded in Chapter 3 that explanations in terms of altering strengths of behavioural responses, without recourse to inner memories and expectations, are distinctly implausible over the range of learned activities exhibited by the rats, cats and dogs which are the source of such explanations. However, it would obviously be useful if experiments could be designed which reduced ambiguities about the characteristics of internal representations which can be directly inferred from behavioural data. There is a fair amount of data from such experiments which I shall briefly review before returning to theoretical issues. For convenience the evidence can be put under the headings of: response to recent perceptions; comparison of current and recent perceptions; and extrapolation from remembered experience.
Response to recent perceptions
The question of whether sensory images persist for relatively short periods of time after their instigation by outside events is rather
different from that of whether representations of external stimuli can be revived or retrieved long after brain activities induced by the initial receipt of the stimuli have died away. Even without special-purpose design for retention of sensory information, the successive stages in, for instance, the visual pathways, would each allow for persistence of patterns of neural activity induced by external events. Then, given the number of loops in the visual pathways (from thalamus to cortex and back again, for instance; see E. G.Jones, 1974 and Gross, et al., 1974), it is easy to imagine that some visual information can be shunted around the loops for seconds or minutes after a given retinal input has ended. It is harder to imagine how activity in the visual pathways characteristic of a given retinal image could be reinstated at will, with the eyes closed: retrieval of information long gone is more likely to be partial and abstract, and the physical brain mechanisms by which inner descriptions and representations can be quietly put away and stored, to be brought back to life at a later date, are completely obscure still, although biological retention of information in the genetic code on the one hand, and the workings of computers on the other, provide appealing metaphors.
In terms of subjective analysis, and of behavioural experiments, it is also possible to distinguish between ‘short-term’ sustaining of immediate perceptions, and the retrieval of information which has lain dormant in ‘long- term’ memory. Remembering what we were doing ten minutes ago, or what has just happened in a game of football, is automatic and effortless, and still has the character of perceptions of present events: remembering the same sorts of things after time intervals of days of weeks is more arduous and uncertain, except for very rare and vivid experiences. Sometimes, it is difficult to draw any line between active remembering, and automatic knowing. We do not have to remember how to get from the kitchen to the bathroom in our own house, in the sense that we can do this without pausing for thought, but our knowledge of spatial relationships which allows us to do this is certainly based on past experience, and the storing of information. It is only in someone else’s house, which we may have visited once, months or years before, that our knowledge of where the bathroom is becomes sufficiently uncertain to require head scratching, and attempts at mental reconstructions.
In behavioural experiments with animals, it is only possible to guess at mental processes which may parallel such introspections. But
especially for ‘short-term memory’ or the retention of recent perceptions, fairly direct evidence can be obtained as to whether any sensory information at all persisted in the brain of the animal, and if so for how long. In experiments on ‘long- term’ spatial knowledge, such as that exhibited by maze-running or ‘homing’, it is clear that animals benefit from previous experience, but less clear that independent perceptual memories can be distinguished from habitual body orientation and hierarchies of response.
Delayed response methods
The paradigm for experimental demonstration of reference to recently perceived events is the ‘delayed response’ method pioneered by Hunter (1913). Animals can be easily trained to run to one of, say, four doors to find food, if the correct door is indicated by a consistent signal such as a light. If the animal is prevented from making its choice until a minute or so after the signal has been removed, but continues to respond correctly, then some information must have been retained over the span of the delay period. It is uncertain, though, whether what is retained is a perceptual representation—for instance, activity in the visual pathways which corresponds to the presence of a light over one of the doors—or a motor instruction, such as a tendency to move to the extreme left, or to the middle right door. It is usually stressed that the persistence of a motor instruction could take the form of ‘pointing’ at the correct location by bodily orientation (e.g. Ruggiero and Flagg, 1976) which would not require any internal kind of memory. This possibility makes the interpretation of results from some delayed response experiments difficult. Hunter (1913) himself concluded that rats and dogs in his experiments could only manage to respond correctly with the help of pointing postures, since preventing pointing by moving the animals, or moving the visual displays, induced mistakes. But racoons, as well as children, responded accurately despite these types of disruption. There is little doubt that animals frequently resort to postures, pointing of the eyes, and similar strategies to solve delayed-response problems they are set, since such strategies are successful. But this does not mean that the same animals might not use more interesting internal forms of memory if postures were not so much part and parcel of the procedure (Weiskrantz, 1968).
In tasks requiring the use of perceptual comparisons, discussed in the next section, postures do not help, and therefore this problem of interpretation is eliminated.
But there are many results obtained using simple forms of delayed response that also provide evidence for memory processes going beyond maintained body postures. Even if rats move about a good deal during delay periods of up to an hour or more they can select locations where food has been signalled prior to the delay (Maier and Schneirla, 1935; Ladieu, X944; Sinnamon et al., 1978), whether the movements during the delay were spontaneous (Maier and Schneirla, 1935) or because the animal has been carried about by the experimenter (Sinnamon et al., 1978). Similarly, although monkeys are sometimes confused if they are distracted during the delay between observing a visual cue in a certain position and manually searching for food in that position (Ruggiero and Flagg, 1976), they often appear to retain information about individual position cues over long intervals, while they are running about and looking at other things (e.g. Yerkes and Yerkes, 1928; Weiskrantz, 1968; Medin and Davis, 1974).
A systematic and substantial body of work on animal memory, which includes the use of delayed response methods, has been presented by Beritoff (1965, 1971). He was a student at St Petersburg in Pavlov’s day, and worked initially with conditioned reflex techniques. The story goes that he first began to search for more elaborate explanatory concepts after observing that a dog who saw a piece of meat thrown out of a second floor laboratory window reacted by running downstairs. In any event, Beritoff and his colleagues in Georgia, where he worked for many decades after 1919, used tests of the kind where a dog in a cage would be shown a piece of food, which was then hidden behind one of several screens and the dog let out of its cage several minutes later. If the animal then went directly to the hidden food, Beritoff (1971) attributed this to ‘image- memory’. It is not always clear from Beritoff’s accounts that postural pointing or the use of the sense of smell has been ruled out, but taken as a whole, this sort of experimental evidence suggests very strongly that some kind of inner memory of where food is, a memory which appears to dissipate with time, is present in dogs. The fact that an experienced dog would sometimes go to sleep during a memory-interval of thirty minutes, and then perform correctly when its cage was opened, shows that sustained pointing is not always necessary, and the finding that even the best animals made random choices if periods of more than one or two hours
had passed since they saw where the food was put demonstrates that the smell of food by itself was not a sufficient cue for correct choices under these conditions.
Beritoff (1971) attempts the difficult task of comparing the effectiveness of this ‘image-memory’ in a variety of vertebrate species including goldfish, frogs, turtles, lizards and pigeons. Only a very limited form of image memory could be demonstrated in goldfish. They were adapted to a large aquarium divided into three sections at one end where they were fed, and then netted and transferred back to the other end, from which they could swim back into one of the sections to be fed again. Although the fish were initially frightened by this procedure, Beritoff reports that they eventually came to ‘follow the net of their own accord’ (Beritoff, 1971, p. 92), a surprising result in itself. However, the main point was that if they were allowed to swim back to the feeding sections within 10 seconds of being placed back at the starting end of the aquarium they swam directly to the place where they had just been fed, whereas if they were held back in the starting end for more than 10 seconds, they usually swam down the middle of the aquarium, whether or not they had just been fed in the middle section. This may mean that in goldfish there is only a very brief persistence of internal brain activities sufficient to direct movements towards particular locations in space, but isolated results like this obviously need to be interpreted with caution.
It is rare for frogs and other amphibians to reveal even the most rudimentary effects of conditioning procedures, let alone ‘image memory’. Beritoff reports that frogs learned to go to a feeding tray in response to a light only after scores of feedings in which a worm was put in the tray 1o seconds after the lamp was turned on. Frogs also swam away from electrodes which delivered painful shocks, and selectively refrained from eating strips of meat after tasting one that had been dosed with oil of cloves. Such ‘conditioned aversions’ are probably the most reliable way of demonstrating responsiveness to past experience, but can be interpreted as special-purpose instincts for modifications of food selection (see Garcia et al., 1977).
For lizards and turtles, however, the behavioural evidence for ‘image-memory’ was similar in kind to that obtained from dogs. These reptiles were confined to a certain part of their living quarters, shown a piece of food being hidden within their sight, and then released to see if they could retrieve it. This they could do successfully if they were released within 2 or 3 minutes of seeing the food hidden, but not after
longer intervals. If some obstruction were deliberately placed in the most direct path to the hidden food, the animals climbed over it or detoured around it. These and other results indicating a considerable degree of flexibility of reptile behaviour in response to prior experience are reviewed by Burghardt (1977). There is far less experimental work on memory and learning in reptiles than there is on birds and mammals, and negative results are often obtained because many species feed infrequently and require high ambient temperatures for free movement, but it would be unwise to conclude because of this that the control of reptile behaviour is entirely a matter of simple reflexes and instincts.
According to Beritoff’s data, the time intervals over which pigeons and chickens can be restrained after being shown food being hidden, while still being able to locate the food when released, are roughly the same as those for the reptiles, that is 2 or 3 minutes. However, if the birds were allowed to eat part of the hidden food, carried back to their cage, and then released again, this second release could be delayed up to xo minutes with the birds successfully returning to the partly eaten bait. This effect did not occur with the turtles and lizards, which could only return to partially eaten food after the same interval of 2 or 3 minutes after an initial visual sighting of where food was hidden. In terms of the time intervals used in the visual sighting test, the birds seemed only marginally better than the reptiles. In all mammals tested, including rabbits, cats, and baboons, Beritoff says that appreciably longer intervals could be used, but, especially in the case of the rabbits, prolonged training with gradually increasing delay periods was necessary to establish this superiority, so the assertion should be treated with caution. It would be churlish to deny, however, that the image-memory of the baboons was remarkably robust, and appeared to survive longer time intervals and greater distraction than that in any other of the species tested by Beritoff. It took 2 or 3 months to acclimatise the baboons to living in a cage in the experimental room, and to moving around quietly outside the cage, with or without a collar and leash. When this had been done, they could be tested in the conventional way, by being shown a piece of apple in a basin, before the basin was placed behind one of six screens a few yards distant. They could be relied on to go directly to the hidden basin if they were released from their cage up to 30 minutes later. It is not clear, that there is a large quantitative difference between the baboons and the dogs and cats in the length of this interval, but Beritoff’s conclusion that the
baboons’ memory was ‘better developed’ may be based on slightly different tests given to the primates. For instance, if a baboon’s cage was covered after it had seen food being hidden, it might shake the cage and make prolonged vocal protestations, but still go directly to the hidden food when the cage was uncovered and it was let out. Or, if a baboon was taken out of its cage on a collar and leash and led about the room without being let near hidden food, but being given something else to eat instead before it was shut up again, it could still go to the previously hidden bait if it was now let out without a leash. This certainly rules out body orientation or pointing as the memory mechanism, and one presumes that olfactory cues were discounted by sometimes removing food after it had been hidden. (The deterioration of food finding after delays of more than 30 minutes is also evidence against the importance of olfactory cues.)
In the baboons, but in the dogs and cats as well, it was possible to demonstrate memories lasting weeks or months, it they were carried into a new room which they had not seen before, given food in a particular place, and then released at the door of this room days or weeks later. Going directly to sniff or search in the place where they had once been fed—under a table or behind a chair—was, very reasonably, taken as evidence for the long-term retention of information picked up at the previous feeding. The comparisons between species may be misleading, since it is difficult, if not impossible, to be sure that one can equate across species factors such as motivation—how interested particular species are in the rewards selected for them, at the levels of hunger used—and the degree to which a species is adapted to the conditions of testing, either in terms of its natural capacities or in terms of the amount of experience of the testing conditions. In the case of memory for food locations in new environments over long periods, Beritoff suggests that pigeons and chickens can manage only 5 days, by comparison with the several weeks observed in the larger mammals, but the quantitative difference is perhaps less important than the fact that in both birds and mammals it is possible to demonstrate retention over periods measured in days rather than minutes.
One can say, however, that the most elaborate memory for observed food locations reported in animals has been that found by Menzel (1973) in chimpanzees. In this case, a group of six wild-born chimps had lived in a one-acre field in which tests were conducted for over a year before the tests began. Just before an experiment, all six animals
were locked in a cage just outside the field. Then someone carried one of the chimps from the cage and around the field, accompanied by another person who, while the chimpanzee watched, placed 18 piece of fruit in different hiding places. This took about 10 minutes, and they the animal was put back in the outside cage for 2 minutes before being released with the other five to look for the fruit. The main outcome of the experiment was that a chimp who had just seen the 18 pieces of fruit hidden (four animals were given four tests each) managed, on average, to find 12 of them. Other chimps in the field at the same time, who had no prior knowledge of the whereabouts of the hidden items, occasionally came across them, but on average the five uninformed animals only discovered one piece of fruit between them during each trial when the informed chimp was picking up 12.
There is thus no doubt at all that the animals who watched the pieces of fruit being hidden retained information derived from this experience, in a form which allowed them to run in direct lines to the various food locations. Other aspects of their behaviour suggest that their memory could be described as a spatially organised description of what was hidden where. For instance, they did not pick up the fruit in the order in which it had been hidden, rather, they started with a piece fairly close to the position at which they were released into the field, then went to the hiding place closest to their current position, and so on. There was some variation in this, but they organised their search route more or less according to a principle of ‘least-distance’ between successive stopping places. Some of the variation may have been due to preferences for individual hidden items: in further tests 9 pieces of fruit and 9 pieces of vegetable were dispersed over the field, with the result that the chimps (who preferred fruit to vegetables) modified the ‘least distance’ principle in order to pick up fruit rather than vegetables first. They thus remembered not only that something was hidden in the various hiding places, but also, to some extent, knew what it was that was available at each location.
It should not be thought that the only form which animal memory can take is one which elicits movement towards food. Most of the tests discussed so far could be interpreted in this way, but spatial memory must include other landmarks besides food locations, and there are other large parts of the behavioural repertoire, in particular social interactions, which involve much more than spatial position. Even with food searches it needs to be pointed out that the chimpanzees in Menzel’s experiment had to remember not only the original locations
of hidden food, but also whether or not they had already harvested these locations during their current search. A simple rule of ‘move to where food has recently been experienced’ would lead to many wasteful returns to empty holes in the ground and unproductive tree stumps. It may be noted that Thorndike’s ‘stamping-in’ principle and other theories in which rewards are supposed to automatically strengthen preceding responses would predict much wasted effort in these circumstances.
A rule of not returning to places recently emptied of food seems to be readily adopted by rats as well as chimpanzees according to experiments reported by Olton and Samuelson (1976) Olton et al. (1977) and Olton (1978). These experiments make use of an apparatus known as a ‘radial maze’, in which a number of arms (usually 8) project out from a horizontal platform like the spokes of a cartwheel lying on its side. There is no outer perimeter, so the only way for rats to move about the maze is back and forth to and from the central platform and the end of any one of the projecting arms. The procedure is straightforward. If the spoke pattern is composed of alleyways on a floor, rats will explore every part of it spontaneously, and if elevated pathways raised a few feet above the floor are used, a little preliminary encouragement, in the form of bits of food visible on the arms, is all that is necessary. The experimental test is to place one food pellet, out of sight, in a hole drilled at the end of each arm. The most efficient thing for a rat to do in these circumstances is to traverse each arm in turn, picking up the food pellet, but never returning down any arm for a second time. In fact, rats do not always select arms in any systematic order, but if there are 8 arms, they almost never visit the same place twice during the first 8 choices (Olton and Samuelson, 1976). In other words, they seem to possess what Olton (1978) calls a ‘working memory’ of the previous choices they have made, plus a strategy of not repeating the same choice. If a similar maze was made more difficult, by including 17 arms radiating from a central platform rather than 8, then errors were made, but trained rats chose an average of about 15 different arms in their first 17 choices (Olton et al., 1977). This is somewhat less dramatic than it may seem, since choosing entirely at random would lead to traversing 11 different arms in the 17 attempts (e.g. going down 6 of these 11 twice). However, if there were no tendency to avoid previously chosen arms, making 15 or more different choices in the 17 attempts would be statistically extremely unlikely (p < .002) and so it is almost certain that there was some form of retention of what previous
choices had been made. Additional experimental procedures (e.g. confining the animals in the central platform to prevent systematic turning into a new alley as the animal emerges from one just traversed) suggest that neither body orientation nor the use of odour is necessary, and that previous choices are remembered by rats in terms of the directions they have gone in before (Olton, 1978).
Remembered perceptions directing delayed responses
The work of Hunter (1913), Beritoff (1965, 1971) and others suggests that what is perceived at one moment maybe used by animals to direct responses a few minutes later, and in some cases hours or days later. The nature of the information perceived and retained is a matter of inference and theory, but in most experiments its content has something to do with the location of food. This is partly a matter of experimental convenience—hungry animals will reveal their inner memories by external actions directed at getting food. But it also points to the utility of memory. Purely instinctive reactions which take no account of recent experience of food distribution may serve some species well, but being able to adjust foraging strategies according to circumstances might confer profound advantages on species equipped with brains which allow this (see Krebs and Davies, 1978).
It should be acknowledged that although it makes sense to assume that many vertebrate species have evolved mechanisms which enable them to incorporate both remote and recent experience into searches for food and other patterns of behaviour, and that delayed response methods provide clear experimental evidence for such abilities, parallels between animal-memory in this form and human memory, as it is subjectively known or experimentally tested, are still rather speculative. Current availability of items of information recently received is usually attributed to a particular theoretical component of human memory such as ‘primary memory’ (James, 1891), ‘short-term memory’ or ‘working memory’ (Gregg, 1975; Baddeley, 1976). A subvocal or ‘phonological’ method of temporary retention, such as an ‘articulatory loop’ is usually emphasised in these discussions, and clearly the ‘image-memory’, inferred by Beritoff from his delayed response experiments on animals, does not correspond to a speech-related process of this kind. On the other hand, it seems possible that Beritoff’s ‘image-memory’ has something in common with the ‘visual
short-term memory’ proposed by Baddeley (1976) on the basis of experiments with people.
Retrieval of spatial information from a longer-lasting store is suggested by the ability of Beritoff’s baboons to find hidden food after an interval of some minutes during which they had been distracted by eating other bits of food, as well as by their ability to go back to a room after a week’s absence and search in the exact place where they had seen food before. Menzel (1978) reports that his chimpanzees, after finding and eating 10 or so of the 18 pieces of fruit secreted in various parts of the field they lived in, would often lie down and rest, and appear to doze off. Then, suddenly, they would rise to their feet and run 10 or 20 yards to collect another bit of fruit that had been concealed (in their sight) close by. It is hard to resist the conclusion that, after a period of rest and digestion, they had just remembered another convenient food location, after it had been temporarily forgotten about, in much the same way that a human subject might remember things when participating in a similar exercise.
What may be true of the chimpanzee, whose brain and digestive system are very like ours, may not be true of other mammals, and almost certainly has even less in common with what is true of other classes of vertebrates, represented perhaps by the goldfish and pigeon, with brains more than a hundred times smaller than the ape’s. Additional behavioural evidence concerning relations between current actions and previous perceptions would be helpful, as well as data which might tell us something about how mechanisms inferred from behaviours are in practice accomplished by brain processes. But before going on to these further issues it is worth looking at some work by Krushinsky (1962, 1965) on what he terms ‘extrapolation reflexes’, which can be used as examples of the functional utility of remembering visual information about food location in order to direct food-seeking activities.
Extrapolation from remembered perceptions
In all the delayed response problems discussed so far, the solution required some form of memory of the locations of stationary food objects. For many animals, food objects are commonly not stationary, but moving. Predators who chase their prey may be observed to head off moving prey, and respond as if sophisticated calculations are made
about the likely intersections of their own and the prey’s movements. We can thus ask whether, if a dog is given a brief glance of a moving rabbit, and released a couple of minutes later, it will run to where it saw the rabbit, or to a point off’ in the direction in which the rabbit was running. Such an experiment would be difficult to set up and interpret, and Krushinsky, after having worked with dogs in the Pavlovian tradition, at Moscow, chose to use simple laboratory apparatus to investigate questions of this type.
One form of test which may involve very recent memory is the ‘Umweg’ or barrier problem (see Kohler, 1925) in which an animal can see food on the other side of a short screen, and must go around the screen to reach it. Possibly a dog which loses sight of the goal as it runs around a barrier for the first time retains an expectation that something will still be there when it arrives, and the fact that mammals seem to react more intelligently to such detour problems than, for instance, reptiles (Burghardt, 1977) is often taken to imply greater cognition in the sense that mammals can react more effectively than reptiles to objects not immediately present.
Krushinsky modified the barrier problem in the following way. An opaque screen roughly two meters wide had a slit down the middle through which animals of various species were allowed to poke their heads to eat from a bowl on the other side. There were in fact two bowls, to the left and right of the gap in the screen, one containing food and the other not. The important test was when, for the first time in the experience of an individual animal, both bowls began to move out of reach, away from the gap, and along the screen which separated the animal from the bowls. If this was done with pigeons, they usually withdrew their heads from the gap and walked a few steps in the direction in which the food bowl had moved, making poking movements at their side of the screen. By contrast, dogs, given the same test, but of course allowed a larger gap to get their heads through in the first place, reacted to the withdrawal of the bowls by running round the screen to the side where the food had been (Krushinsky, 1965). Out of 27 pigeons, only 2 ran around the screen; while out of 18 dogs tested, 14 ran around to the food side and 2 ran round the other side.
This could be taken as an expression of a natural aptitude for the tracking of moving objects in dogs, which is absent in pigeons, but also indicates an effective short-term retention of perceived events which is more marked in the dogs. Krushinsky’s detour test does not, however, cleanly separate the abilities of birds from those of mammals. Birds of
the crow family (crows, magpies and rooks) did just as well as dogs, about three-quarters of those tested immediately running around the screen on the correct side as soon as the bowls moved Out of reach. A few (6 out of 42 tested) ran around on the wrong side, in the direction of the empty bowl, and a couple of crows quite sensibly flew over the barrier to the point where the food bowl had moved to (Krushinsky, 1965). Over a hundred chickens were given the same test, and about half of them went around the screen when the bowls were moved, but, of these, as many went to the empty side as to the food side. This is rather better performance, in so far as such judgments can be made, than was obtained with rabbits, since only a minority of the rabbits used detoured around the screen, some to one side and some to the other. Cats, like the dogs and crows, could usually be relied on to go around the screen, but almost as many cats went in the direction of the empty bowls as detoured on the side of the full one.
All these results refer to the proportion of animals detouring around a barrier for the first time: this is important since it means the actions were directed by what had just been perceived, and were not stamped in by special training. With prolonged trial-and-error learning, Krushinsky found that pigeons would run around the screen, but in the same direction each time, while some of the chickens and rabbits learned to run in the direction of the food, which changed randomly from trial to trial. It was always easier, however, to train chickens and rabbits to run in a particular direction—going round to the left each time for instance—than to train them to run in the direction where the food was. It looks very much as though, for these species, it is easier to remember that there is food on the other side of the screen, than it is to remember whether the food is to the left or to the right on any particular occasion.
Various other food-finding tests were used by Krushinsky, all of which supported the idea that what he calls ‘extrapolation’ is easier for some species than for others, where the extrapolation is from very recent to current experience. It is worth mentioning one of his more unusual tests, which required a special apparatus rather like a toy railway set. A food bowl could be run down a straight track, at 8 cm per second, for a metre and a half in the open, then into a 3 metre tunnel, emerging again at the other end. The measure of an animal’s recent memory could be taken as how long it searched for food after it disappeared down the tunnel, and the distance it ran down alongside the tunnel. Pigeons, which followed the moving bowl and fed from it,
simply turned round and walked back along the track as soon as the bowl entered the tunnel. Chickens and rabbits stayed at the entrance of the tunnel (which was covered with a flap) for a few seconds or sometimes half a minute, while crows and magpies walked up and down outside the tunnel for about a minute. Even the crows did not initially walk all the way to the end of the tunnel but in a couple of cases where the tunnel was cut in the middle so that the birds could see the bowl going past, Krushinsky reports that the crows went all the way to the end and stayed until the food bowl emerged. This is not very much to go on, and the test was not tried with cats and dogs, but this kind of set-up reminds us again that even for a chicken scratching for insects, it might be extremely useful to be able to go on scratching in a place where an insect had just been seen, and, for many predators, being able to remember where a moving prey item has just disappeared to ought to have high priority as a psychological function.
Comparison of current and recent perceptions
Although the relatively straightforward methods of delayed response testing may correspond to the natural uses of retained perceptions in animals, it is often hard to decide whether responses to recent stimuli are controlled by the retention of sensory information as opposed to the persistence of motor reactions. The sight of an insect may elicit scratching reflexes in a chicken, and such reflexive behaviour, once begun, may continue for some seconds in the absence of any internal representation in the brain of the bird which could be considered to be a visual memory of the eliciting image. In many experiments the length of the interval for which responses are delayed, and variation in motor activities observed during this interval, may argue against such simple mechanisms of response persistence, but it is helpful if ways can be found to get clearer evidence for the retention of information in the form in which it is initially perceived. Experimental procedures which require animals to make comparisons between recent and current events appear to do this.
Delayed matching to sample
A technique which has been used for some time with primates (French, 1965; D’Amato, 1973) and occasionally with pigeons (Blough, 1959; Smith, 1967; Roberts and Grant, 1976) assesses knowledge of recent
stimuli by choice among two or more current comparison items. Typically there is a row of three panels, on which visual stimuli such a coloured lights can be displayed. A ‘sample’ image is first put on the centre panel—let us suppose that it is a horizontal line. Usually the animal has to touch the sample stimulus in some way to go any further. When it does, the sample goes out, and a fixed time later (say 20 seconds) the two outer panels are lit, one with a horizontal line, and one with a vertical line. In most experiments the procedure is such that the animal is given food reward if it touches whichever outer panel is the same as the centre panel was a few seconds previously; and usually there are only two possible stimuli to choose between. Even so, it would seem to be necessary for an animal to retain some perceptual knowledge of which of the two stimuli had just been presented in order to respond correctly at the end of the delay interval.
The experiment reported by Smith (1967) can be used as an example. In this case pigeons pecked at a centre display which was either a horizontal or a vertical line. A few seconds after the centre display had disappeared left and right displays came on, one with a horizontal line and the other with a vertical line. The crucial point about this type of experiment is that sometimes the horizontal line is on the left and the vertical line on the right, and sometimes vice versa, in a completely unpredictable way.; and the rule is that whichever display is the same as the preceding centre sample must be pecked by the bird in order briefly to operate a food dispenser. In Smith’s experiment, each individual pigeon had a daily session of 120 trials, one after the other. After some weeks of training, the birds responded correctly on more than 90 per cent of these trials if there was a minimal memory problem, that is, if the side choices came on just as soon as the centre sample went off. But there was a steady decline in accuracy, on days when there was a delay between the sample and the choice stimuli, as the delay increased from 1 to 51 seconds. With a 5-second delay, the three pigeons tested were only correct on about 6o per cent of the trials, which is not sufficiently different from the 50 per cent that would be achieved by chance to say that anything very much was being remembered. With only 1 or 2 seconds delay, performance at better than 70 per cent correct indicated that choices were being influenced in some way by the character of the preceding sample display.
Extensive research using similar experimental procedures, but with colours as the visual stimuli, has been summarised by Roberts and Grant (1976: see also Grant, 1976; Roberts and Grant, 1978). When
the sample colour has been seen for only one or two seconds, it is not usually possible to detect any influence of the sample on choices made more than 20 seconds later, and accuracy always drops off rapidly when the sample has to be remembered for more than 2 or 3 seconds. However, if each sample is seen for 10 seconds or more, the effects of the sample last for up to a minute. In other words if a pigeon sees red on a centre panel for 10 seconds, and then has to wait for a minute before the side panels are lit up red and green, it is still statistically more likely to choose the red panel, whichever side it is on (Grant, 1975, 1976).
Fairly similar experiments with New World monkeys have been reviewed by D’Amato (1973). On the face of it, the initial performance of these primates is not all that different from that of pigeons, since they appear to retain a memory of the sample sufficient to have a statistically significant effect on subsequent choices only for delays from 10 to 20 seconds. However the usual emphasis (D’Amato and Cox, 1976) is that after years of training these monkeys are very different from pigeons because they can remember the colour of a sample for 3 or 4 minutes. One monkey in particular was reported to perform at chance levels with delays of more than 10 seconds with less than 5,000 trials of training, but to have improved, after 30,000 trials spread over 6 years, so that the colour of the sample influenced choices made 9 minutes later. This is interesting, but one can argue that it shows that over a few weeks of training both pigeons and monkeys display ‘short- term memories’ which last roughly 10 seconds, which is a value in the range quoted for experimental studies of human capacities for things like remembering a telephone number after seeing it for the first time (see e.g. Gregg, 1975; Baddeley, 1976). The fact that monkeys get much better after years of specialised practice is difficult to compare with standard human experimental data, but people with specialised interests in remembering quickly presented information such as telephone numbers or the orders of restaurant customers should not have any difficulty in surpassing textbook figures, and remembering things seen or heard only once for to minutes or more.
It is likely that there are differences between species in the time intervals and richness of content possible in this sort of memory for recent events, and also that there are differences between sensory modalities. But there is little evidence available from experiments using modalities other than vision, and even for vision one is limited to experiments on pigeons and a handful of species of monkey. Herman and Gordon (1974), using three localised sounds in a procedure like the one described above using three visual displays, found that
bottlenose dolphins could select a sound which matched a sample heard 2 or 3 minutes earlier, and the experiments reported by Pavlov (1927: discussed in Chapter 3), in which dogs were trained to salivate only to certain sequences of musical tones, imply retention of auditory information for periods of seconds at least. Especially using sounds it would be an advantage to test memory for the location of a stimulus separately from memory for other auditory characteristics, but there is little evidence on memory for sound location as such, although Beritoff (1971), as part of the work on delayed food- seeking described above, included experiments in which blindfolded dogs and cats heard a food basin being banged in a particular location, and were able to go directly to it if released within 5 minutes.
A technique for avoiding the spatial aspect of choosing stimulus locations had been suggested by Polish researchers (see Konorski, 1967, and Wasserman, 1976). Two sounds are presented one after the other, with a short intervening interval. If they are the same, the animal is trained to make a response; if they are different it is trained not to. To the extent that this training is successful, over a wide enough range of sounds, one can infer that information about the first sound is being retained long enough for a comparison to be made with the second. The results quoted by Wasserman (1976) indicate that dogs can learn to move a paw when two low tones occur in sequence, or two high tones, but not to move their paw if there is a high-low, or low-high, succession. A similar task provided no difficulties for Old World monkeys tested by Stepien et al. (1960). One would assume, perhaps, that dogs or monkeys hearing the barks or shrieks of conspecifics do not lose all the information contained in these noises immediately the sounds cease. But memory for sequences of stimuli is not confined to hearing in mammals. It has been shown, for instance, that pigeons are influenced by the order of recent visual stimuli; they can be trained to respond to a vertical line which follows a sequence in which a green light preceded an orange light, but to respond instead to a horizontal line if the previous sequence included orange before green (Weisman, et al., 1980).
Recognition memory in monkey and man
On the grounds of phylogeny and brain anatomy, we should certainly expect that the characteristics of memory in monkeys and other primates approximate our own much more closely than those of the
pigeon ever could. The perennial question is whether the verbal coding of information by man radically alters capacities that may be measured by non-verbal methods. In the case of memory, it might be said that the predominant characteristic of human performance is the verbal description of past events—in particular the ‘recall’ or reproduction of words previously heard or seen. As species other than ourselves do not give verbal descriptions of anything at all, let alone descriptions of previous events, we could conclude that memory as we know it is, by definition, completely impossible in animals.
But some aspects of human memory are less bound up with linguistic description than others. We may remember a face, whether or not we can put a name to it, and perhaps it is not stretching the imagination too far to suggest that we might be able to point to what we had for breakfast yesterday, as a means of getting it again today, even if we were deprived of all the powers of language. Discriminating between the familiar and the unfamiliar, and perceiving objects and events as belonging to certain categories, are often referred to as ‘recognition memory’ in man. It is a matter of considerable dispute as to whether ‘recognition’ and ‘recall’ of verbal material are different manifestations of the same underlying capacity for memory, or require alternative kinds of theoretical explanation, in experiments on human subjects (e.g. Brown, 1976). Whatever the merits of these arguments, it is clear that something akin to recognition is relatively easy to assess in monkeys, even though the existence of anything remotely like verbal recall in apes is extremely dubious (see Chapter 9).
Detecting familiarity or novelty in visual images is subjectively one of the easiest ways to demonstrate the retention of previous experience. A self-administered test suggested by Davis and Fitts (1976) is to leaf through a well- illustrated magazine, idly looking at the pictures. Half an hour later, turn the pages over systematically one at a time, trying to decide for each illustration whether it was one you had looked at before or not. A test such as this may indicate a memory of a kind for hundreds or thousands of separate visual displays, even though, after the first thumbing through, one might not be able to give verbal descriptions of more than half a dozen (Standing, Conezio and Haber, 1970). Regular readers of illustrated magazines would expect to make a reasonable guess not only at whether a given photograph had appeared in the last year, but whether they had seen it in the last week or much longer ago. Inference about content would of course prevent this from being a fair estimate of visual memory alone, but few would
dispute that an enormous amount of information that is not normally available to our faculties of verbal recall can be revealed by tests of visual recognition.
Hints that a large capacity for recognition of visual images exists in monkeys are contained in experiments reported by Gaffan (1977a) and Davis and Fitts (1976). Gaffan (1977a) trained monkeys to watch a screen on which 25 different coloured slides were each presented twice during a testing session. The 25 were made up of 9 snapshots (for example, a beach, or a view from the Eiffel Tower) and 16 reproductions of figurative paintings (including landscapes and nudes). The point was that the slides were presented in a random sequence, and the animals had to notice whether a given slide was being presented for the first or second time during the test session. If they repeatedly pressed a response button during the second presentation of a slide, they were given a sugar pellet (via an automatic delivery device). But if they pressed this button during the first presentation of a slide, the effort was wasted, since nothing happened.
After a course of training, the two young rhesus monkeys used were able to distinguish with an accuracy of over 90 per cent between first and second showings of the 25 slides, when an average of 9 other slides and up to 18 others might intervene between the first and second showings. (Obviously, in the sequence of 50 presentations, the beginning ones had to be mostly first showings, and the last half-dozen or so had to be second showings, and correct performance here would not necessarily indicate visual recognition of individual pictures, but only the middle part of the test-sessions was used to assess the monkey’s accuracy.) All the pictures must by this time have been familiar to the animals, although this was not explicitly confirmed in the experiment. What the animals were doing was therefore not closely analogous to the illustrated magazine example just discussed, but more like remembering what cards have been played during a particular game of gin-rummy. Considerable training was required before the monkeys could perform the task accurately and they were no longer able to do much simpler forms of this recognition test after a brain operation (see below).
Davis and Fitts (1976) cut large numbers of illustrations out of magazines and pasted them on small boards which could be presented to monkeys as covers over wells in which titbits could be found—a standard way of testing the perceptual capacities of monkeys (a version of the Wisconsin General Test Apparatus or WGTA; Harlow, 1959).
The experimenters were not interested in finding out if their animals (rhesus and pigtailed macaques) could remember whether or not they had seen a particular picture before, but in testing the monkeys’ capacity for selecting a picture seen only once before according to whether that picture had been rewarded or not.
Their suggestion was that this ought to be much more difficult than responding according to familiarity, partly on the basis of the illustrated magazine analogy. Remembering whether individual pictures have been rewarded or not is a two-way classification, such as one may attempt to make by going through a series of pictures in a magazine and noting whether each one is on an odd or an even page. This is certainly much more difficult than just recognising recently seen illustrations as recently seen, but peanut versus no peanut is presumably a rather more vivid dichotomy for the monkey than odd versus even is for us. In any event the difference between familiarity, and remembered associations with rewards, was not actually assessed. The importance of Davis and Fitt’s results is that they provide separate evidence for remembered associations of particular pictures with reward, which, although it is not especially surprising, provides an experimental demonstration that emotional associations of recent events are retained, as well as the recent events themselves.
The procedure used by Davis and Fitts was to provide a monkey with one ‘sample’ picture, which the animal pushed away from a food well, thus discovering whether the sample was rewarded or not. Then, as quickly as possible (about one second later) or after a short delay, two pictures were presented at once, one of them the sample and the other a completely new one. The animal revealed its knowledge of the sample by choosing it, if it had been rewarded, but choosing the alternative, if the sample had previously covered an empty food well. (In two- picture choices, rewards were distributed according to this rule: that is, a rewarded sample was rewarded again, but the alternative choice to a previously disappointing sample covered the peanut.) The fact that the monkeys were able to do this at all is a considerable achievement, since each picture was seen only once (if it was an alternative) or twice (if it was a sample). This means that their visual memory functions reasonably well with unfamiliar material. In more stringent assessments of memory the monkeys were presented with ‘lists’ of two, four or sixteen sample pictures in succession, some covering bits of food and some not, before tests in which each sample was paired with an alternative picture. As might be expected by extrapolation from
human capacities, remembering whether individual items in these lists had been rewarded or not proved quite difficult, and there was a very marked decline in the accuracy of the animals’ choices according to the length of the list in which a sample had appeared. Again in line with what happens when human subjects are shown lists of words, position in the list made a difference. The main effect with lists of 2 or 4 pictures was that the most recently seen samples at the end of the lists were much better remembered than those at the beginning of the list. With a list of 16, the monkeys did not remember very much anyway, choices being only 6o per cent accurate on average, and so it was difficult to make reliable comparisons between the various positions, but it looked as though accuracy was still better with the more recent samples, with the first few items on the list being remembered a little better than those in the middle. These variations seem rather trivial aspects of the procedure, but are of interest because enormous amounts of effort have been expended on measuring the recognition of words which have appeared in different positions on lists used in laboratory experiments on human memory.
The crucial aspect of the procedure was whether or not a given sample picture was rewarded. If only one picture had been seen before a choice test, one can imagine that whether it had been seen before, and whether or not the monkey had just eaten a peanut, ought both to be easily remembered. But if two samples had been given before a test, one rewarded and the other not, then the monkey had to remember which was which. Davis and Fitts’s data show that this was much more difficult. By examining in detail the choices made after lists of two samples each, it was possible to see that if both samples had been rewarded, or both unrewarded, so that the animals did not have to worry about the distinction between them, choices were very accurate indeed, but if one was rewarded and one not, many mistakes were made. A curious aspect of these results was that the poor performance after the mixed samples was most obvious if the animals were tested within 5 seconds of seeing the samples—if the tests were delayed for just a little longer (20 seconds) choices after mixed samples actually became more accurate. Perhaps a little time to think was beneficial.
In these experiments (Davis and Fitts, 1976), what had to be remembered, over a few seconds, was whether novel visual patterns had covered full or empty food wells. What exactly it was about individual pictures that was remembered — colour, texture, or details of shape, is impossible to tell. The rule which needed to be followed was
to choose again a novel display which had just been rewarded, but avoid visual patterns recently seen without reward. Gaffan (1977b) used a technique in which colour, or spatial location, was definitely the visual cue which had to be retained, and monkeys had to respond according to whether or not a certain colour, or location, had just appeared in a short list.
Automatic apparatus was used to light up display panels, detect responses, and deliver sugar pellets to rhesus monkeys inside a special cage. For lists of colours, there was one display panel which could be illuminated with six different hues (red, green, yellow, magenta, blue and greenish- blue) and a response panel lit with plain white light. After some preliminary training with simpler versions of the task, individual animals were given test-sessions of the following sort. The display panel was lit with the first colour in a list of three (say red),—and the monkey had to touch the panel to make this colour go off. One second later another colour appeared (say blue), was touched, and after another second a third (yellow, perhaps) completed the sequence. Now there was a loud buzz, and a test colour appeared on the display panel, with the response panel plainly illuminated at the same time. This was the moment of choice. In order to get a sugar pellet, the monkey had to press the display panel if the test colour was one of the ones in the preceding list (in this example, red, blue or yellow), but needed to press the response panel if the test colour was one of the others (in this case, green, magenta, or greenish-blue). Assessed over several daily sessions each containing almost 200 lists, allowing for all the different permutations of the colours, the animals performed very creditably at this recognition task. Not surprisingly, they were rather better at correctly touching the test colour if it had just been seen as the last one in the list, than if it had been given first or second, and overall they made correct decisions after more than 70 per cent of the lists. This was not as good as the 90 per cent they achieved when only one sample colour was shown before each choice, but given the longer time intervals between the recognition tests and the earlier samples in the lists of three, and the possibilities for confusions between lists (it would be counted as an error if they touched a test colour that was in the last-but-one list instead of the most recent list) it would be very peculiar if they had not made a good many mistakes.
After this measurement of recognition of recently seen colours, the same monkeys were run through an identical procedure to assess memory for the spatial location of illuminated panels. In this case three
separate display panels from an array of six were illuminated (in orange light)- in succession, the monkey touching each one in turn. Then after the buzzer signal, just one of the six panels, and the response panel, was lit up at the same time, and the animals had to choose as before: touching the display panel if it was one of those seen in the previous list of three but selecting the response panel instead if the display was not one of the previous three. Again, they did very well, making few mistakes if the test display was the same as the last one on the previous list, and making about 75 per cent correct choices over all.
These experiments, and others like them (e.g. Overman and Doty, 1980; Sands et al., 1980) leave very little doubt that current visual perceptions of monkeys can in some way be matched against the animals’ recent experiences. What mental processes enable this to take place is a different matter. It seems possible that with very short memory intervals, of one or two seconds, a record of visual experience might persist automatically. But for successful performance at longer intervals, some special attention to the to-be-remembered cue, and a specific mental comparison of the choice stimuli with what is retained from recent experience, would surely be helpful. In other words, one tends to assume that monkeys exposed to these procedures adopted mental strategies which they might not employ otherwise.
Recognising exact repetitions of visual displays, especially if simple features such as colour or location are all that is needed, may be a particularly easy strategy to adopt, being to some extent built into perception as a means of establishing familiarity (see Sokolov, 1963, 1975). There are indications that monkeys may have more difficulty in utilising visual memories to make quite arbitrary connections between past and present perceptions. D’Amato and Worsham (1974) presented cebus monkeys with a sample display which was either a red disc or a vertical line. Some seconds later the monkeys were required to make a choice between a triangle and a white dot, according to the rule that the triangle should be selected if they had just seen the red disc, -but the dot should be selected if the preceding sample was the vertical line. They eventually managed to follow this rule, but months of daily training with the choice following the sample immediately was required. This is not very systematic evidence, but in the absence of anything better, we must assume that detecting the presence or absence of stimuli which match recent experience may be much easier than this kind of delayed and arbitrary association.
It should be emphasised that matching current with recent images is
not the only thing that animals can do which implies a form of memory. The monkeys which took a long time to adopt the rule of touching a triangle after a red disc but a dot after a line had no difficulty in stretching the ‘after’ intervals to a minute or more, once this very peculiar convention had been established (D’Amato and Worsham, 1974). It is interesting to speculate as to whether the monkeys did this by remembering, at the end of the minute, if the sample had been a red disc, or by remembering that they should be selecting the triangle (or by remembering both these things). Rather subtle experiments would be needed to distinguish between these possibilities. With a different sort of arbitrary association to be remembered, Gaffan (1977c) was able to design an experiment where the results suggested that it was the end product of the association, rather than the initial cue, which was remembered. This makes sense if one assumes that animals will tend to remember whatever it is that is going to be connected with their getting their reward, since in these kinds of experiment the rewards are the only reason the animals are paying any attention in the first place.
In this experiment (Gaffan, 1977C) monkeys saw briefly one of three sample colours (red, orange or blue). Ten seconds later, they had to press one of three plainly lit panels, depending on what the colour had been. (If it had been red, they should press bottom right; if blue, top right; if orange, top left.) The catch was that only one of these panels was lit, and sometimes it was the wrong one—in these cases the alternative choice of a specially marked ‘No’ panel was necessary. In itself, this is a revealing variation of the delayed-response problem, but the arbitrary connection between individual sample colours and the required response allows two possible strategies. The animals could retain knowledge of the colour of the sample, and work out after a panel was illuminated, if it was the correct one. Or rather more directly, they could select a panel as soon as the sample was seen, and retain this selection, switching to the ‘No’ response if it did not come up. Gaffan argues that this direct method of ‘response coding’ is what took place, because there were many more mistakes that looked like confusions between adjacent panels (presses on the lower right panel after a blue sample which would have required a press on the panel just above) than mistakes that looked like confusions between similar colours (lower right ‘red’ choices after an orange sample that would have prepared the animals to point upper left). These data are hardly conclusive, but there is certainly no reason to doubt that animals can
remember response orientations or spatial locations just as well, if not better, than discriminations of colour or shape (cf. data for pigeons in Smith, 1967). The least intellectually demanding strategy for the monkeys in this case would have been to hold their hand poised over the appropriate panel during the 10-second memory interval. Sitting still, even for such short periods, is not however a very popular activity for young monkeys, and Gaffan (1977c) reports that they jumped around the cage during the memory interval. Therefore he suggests that there must have been internal representations, corresponding to ‘I must press upper left this time’, held, or recovered, over the 10-second intervals.
Memory tests and memory functions
Experimental tests of capacities for memory in animals require information previously obtained to be brought to bear on choices made during laboratory procedures in which animals behave in unnaturally repetitive and arbitrary ways to obtain food rewards. Such tests are performed because they establish the presence of psychological processes which are not always obvious in the natural behaviours of the animals involved, or in other laboratory procedures. To some extent, it is obvious that details of the remembering that takes place are brought about by extensive and specialised training, and the character of the tests themselves. But equally, if special purpose tests demonstrate capacities for various forms of memory we ought to ask, first, how such capacities evolved in the context of a species’ natural life and, second, whether similar forms of memory might be utilised during tests of animal learning and perception which are not deliberately designed to involve remembering, but which do not necessarily exclude it (Shettleworth and Krebs, 1982).
No certain answers can be given to questions about the evolution of mental capacities, but it is reassuring that abilities shown in many delayed response tests would clearly be helpful in natural patterns of food-seeking, and in some cases would be logically necessary for animals to accomplish the strategies of foraging that actually occur in nature (Pyke et al., 1977; Krebs and Davies, 1978). It is less clear how the abilities to compare recent and current perceptions, that are manifested in tests of the ‘delayed matching to sample’ type, might help animals cope with their usual conditions of life. Being able to tell
whether a projection screen is illuminated with the same colour that it was 10 seconds ago or not seems at first sight far removed from any naturally occurring behaviour. It may be, however, that a general capacity for recognition of short- term Continuities is part and parcel of perception. If animal perception is a method of constructing internal representations of reality rather than merely picking up isolated sensations. as I argued in the previous chapter, there needs to be some correspondence between continuities or discontinuities in time in the outside world and the inner representations. The dating of perceptions in time sounds a complicated and mysterious matter, but if we suppose that a monkey surveying its surroundings identifies landmarks, and other monkeys, we should surely assume that as it glances from side to side it retains. if only briefly, the information detected in each glance, for incorporation into its assessment of where it is, where other individual monkeys are, and what is generally going on. Detecting external regularities over time may therefore be a rather fundamental aspect of brain processes serving perception. It is perhaps surprising that anything so fundamental should be revealed in tests in which animals are motivated by the delivery of food rewards, but the combination of persistence and ingenuity on the part of experimenters and greed on the part of laboratory animals should never be underestimated.
Discrimination learning sets
The switching in of information about recent events, as a function of experience, is one explanation for a phenomenon in animal learning known as Learning Set (Harlow, 1949; Mackintosh, 1974, pp. 610—14; R. C. Miles. 1965) This occurs most strongly in Old World monkeys (such as the rhesus) and anthropoid apes. If a monkey is repeatedly presented with a matchbox and a dry-cell battery lying on a tray in different positions. with a raisin always under the matchbox but never under the battery. it will gradually learn to always pick up the matchbox rather than the battery. But this takes time—even after choosing the matchbox and getting the raisin three times, and picking the battery and getting no raisin two or three times, the odds on the monkey choosing correctly next time would only be about 7 to 3. If it is now given a different pair of objects, say a ball of wool and a tobacco tin, with the raisin always under the tin, very much the same thing
happens: after half a dozen trials there will be some evidence of a preference for the tin, but no certain solution to the problem. In order to observe the learning set phenomenon, one has to keep on giving the same animal pair after pair of objects, presenting each pair six or more times before moving on to a new pair, for upwards of one hundred different pairs. The eventual outcome, after monkeys have experienced 250 successive pairs of objects, is rather different from the original tentative preferences (Harlow, 1949), If an animal is now given an entirely new pair of things to choose between—let us say a plastic bottle and an ashtray—it needs only one choice to establish a virtually certain solution. On once finding a raisin under the bottle, or not under the ashtray, there will be a chance of less than one in ten that the monkey will make a mistake on following tests with the same pair. This can be quantified by looking at the choices made for the second time with all the pairs from, for instance, the 250th to the 300th: it is a reliable result that more than 45 of these 50 should be correct (Miles, 1965).
Without going into details, there are grounds for believing that the accurate choices made at the end of learning-set experiments occur because the animal is then using its memory of preceding trials as a successful basis for choosing. This must involve what is referred to as a ‘Win- stay/Lose-shift’ strategy (Mackintosh, 1974), since if the animal remembers that the raisin was -under the bottle, it must choose the bottle again, whereas if it remembers that there was no raisin under the ashtray, it must leave the ashtray alone. Animals (even rats) are perfectly willing to adopt other strategies if they pay off (e.g. Olton, 1978): what interests us here is that no such strategy can work unless information is available at the time of choice which reflects which of the current objects was selected last time, and whether this selection resulted in a ‘win’ or a ‘loss’ (that is whether the animal ate a raisin or not). The availability of this information can be related to more direct experiments on memory because the normal procedure is for the animals to make successive choices within a few seconds. If the monkey successfully chooses the bottle but then has to wait half a minute or more before the next choice, it is much more liable to make a mistake (Bessemer and Stollnitz, 1971). What is not certain is whether, during the long training which results in more accurate choices, the animals are coming to terms with the ‘Win-stay/Lose-shift’ strategy in the light of already available memories, or whether the training is in some sense improving their memories, or encouraging the strategy of
paying attention to what they can remember about what they had just done. Probably a number of factors like this are important.
It is appropriate to acknowledge here that utilising recent memories is not the only mechanism which governs systematic choices made by animals. Innate preferences may influence choices made under experimental conditions as well as in natural environments. The standard explanation for choices of experimental stimuli which are brought about by experience of reward and punishment is of course automatic habits, in the form of strengths of reactions elicited by given sensations (Spence, 1936; See Chapter 3). The point is that one can have a tendency to make a certain choice without remembering the details of the previous experience responsible for that tendency. In human behaviour this is usually only obvious in the case of highly practised skills such as driving a car or playing tennis, where conscious control via remembered verbal instructions may have been crucial in the early development of habits, but in the case of animals the most parsimonious theories assume that certain classes of stimulus input acquire properties of eliciting response output, as a function of reward and punishment, and without the intervention of any mental processes outside individual stimulus-response circuits.
If a pigeon pecks at a green card which covers a hole containing grain, thereby displacing the card, and thus eats soon after the act of pecking something green, and is also allowed to peck at red cards which cover empty holes, it will very soon peck at green cards but not at red ones. One can construct several kinds of metaphorical mental processes which could account for this outer behaviour. Most simply, green cards merely could elicit an inner instruction ‘peck at that’, the subsequent uncovering of food coming as a pleasant surprise each time, or perhaps merely eliciting another instruction, ‘eat that’. Allowing for a slightly greater degree of cognition, we might suppose that either before or after pecking at green cards, the bird possesses some kind of expectation that eating is imminent, without any inner representation of previous events. Assuming, for the sake of argument, quite unrealistic mental capacities, then the sight of a green card might give rise to a train of thought along the lines of ‘I distinctly remember that I ate food from the hole underneath a green card like that a minute ago, and there has always been food underneath green cards on previous occasions’. The view I expressed in Chapter 3 was that a theory somewhere in the middle of this range is needed to cover most of the effects of previous experience on animal behaviour, but it must be
common to all theoretical positions that, especially after long training, previous experience is accumulated into automatic response tendencies, without any necessary access to individual episodes from the past.
The purpose of the elaborate experiments on delayed responses and delayed perceptual comparisons is precisely to ensure that automatic tendencies, of the type ‘always peck at green’ would not be sufficient to account for the data, because another rule has to be followed, such as ‘peck at green if green was present a minute ago, but not otherwise’ (Grant, 1976). The success of these experiments does not mean that animals never make automatic, reflex-like responses—they frequently do—but it adds weight to the view that automatic habits may be modified, and in some cases formed and broken, by the persistence over time of what has been previously perceived.
Memory and brain processes
If it is true that past experience can influence future behaviour in many different ways, including the formation of automatic habits and the revival of perceived features of previous events, there is reason to hope that we may be able to distinguish between types of brain, or between different parts of particular types of brain, which provide the physical mechanisms for accomplishing these psychological processes: For over a century there has been an almost political division between a party of localisers, who believe that psychologically defined mental processes can be assigned to distinct regions of the brain, and a party of unified action, which asserts that such theoretical divisions between brain activities arc illusory. The party of localisers has slightly disreputable roots in the early nineteenth-century phrenology of Gall and Spurzheim, which held that protrusions of parts of the brain concerned with such faculties as Hope and Time could be detected as bumps on the skull, but became more firmly established later in the century with the medical and experimental evidence of Broca and Ferrier.
Set against the localisers are those who are convinced that the physiological apparatus required for any particular psychological functions cannot be pinpointed at particular places in the brain because the brain acts as a whole, and that any identifiable working parts are diffused and spread about rather than neatly collected in anatomical units. In the first half of the last century, the most influential protagonist of these views was Flourens (see Chapter 5)
whose slogans referred to the ‘communal action’ and ‘unity of function’ of the various parts of the brain. His counterpart in this century is Lashley, who worked in the 1930s and 1940s, but who is still quoted as an authority on ‘mass action’ of the brain and the ‘equipotentiality’ of its different parts for accomplishing individual functions (see Boring, 1950). We should say first that this unity and indivisibility of function, for Lashley, and to a lesser extent perhaps Flourens as well, applies mainly to the cerebral hemispheres. Flourens himself was responsible for ascribing the higher aspects of perception and intention to the hemispheres, but the co- ordination of muscle movements to the cerebellum and basic maintenance of life to the medulla.
Even the most extreme diffusionists might tolerate, for instance, the proposition that memory in the form of detailed sequencing of muscle movements is located primarily in the cerebellum. There are modern theories (Bloomfield and Marr, 1970; Eccles, 1973) about how the neuronal structure of the cerebellum may be adapted for this sort of flexible motor- programming. It would probably also have been possible to persuade Flourens and Lashley that the midbrain and the medulla retain previous experience in the form of altered reflexes, while the cerebral hemispheres are able to do something over and above this.
Even such rudimentary distinctions between ways of storing information in the gross divisions of the vertebrate brain would, of course, be a novelty in the context of standard treatments of animal learning, but it would be powerful support for the separate consideration of automatic habits and independent perceptual memories if there were conclusive evidence that these hypothetical mechanisms are characteristic of different anatomical levels of brain action.
Neither Lashley, nor Flourens, nor anyone else, would dispute that, if other mammals possess anything remotely like human memory for past events, it will be located in the cerebral hemispheres rather than at lower brain levels. The arguments start if we ask whether the physical changes that allow for memories (what Semon, 1921, and Lashley, 1950, called the ‘engrams’) are diffused throughout the hemispheres, or collected in particular forebrain structures. However, the empirical evidence, if not the political momentum, is now very much with the localisers. Since Lashley’s time, the physiological mapping out of sensory areas of cortex has become very firmly established. The traditional doctrine of the localisers was that ‘association areas’ (more or less every expanse of cerebral cortex that was not known to have
specific sensory or motor functions) played an important role in memory (Diamond and Hall, 1969). There is still something to be said for this idea, although the trend has been to tie particular association areas more and more closely with adjacent sensory and motor projections (Diamond, 1979).
The most extreme plank of the localisers’ platform contends that, although long-term storage of mental associations can be attributed to neural connections within and between the various projection areas of cerebral cortex, a particular brain structure has something special to do with setting up such associations. This structure is the hippocampus, ‘old’ cortex characteristic of all mammals, and distinguishable as a forebrain region in even the most primitive vertebrates (Faucette, 1969; Isaacson, 1974). The initial suggestion that the hippocampus is a special organ of memory came from studies of human patients with loss of memory. The questions of whether, how and why human memory depends on this particular brain structure have implications for the theory of animal memory, since, once it has been claimed that any particular brain part serves memory in man, we can then ask ‘What is this structure doing in the brains of other animals?’ (Weiskrantz and Warrington, 1975; Weiskrantz, 1977).
The function of the hippocampus in the brains of monkeys and man
It is worth beginning with the issue of loss of memory due to brain damage in man and monkey before setting out to consider brain mechanisms and the retention of past experience more generally, since it illustrates most vividly the difference between remembered perceptions and retained habits. The study of human amnesia also illustrates the great variety of things which might go under the name of human remembered perceptions. The simplest distinction between forms of memory loss in human patients is between loss of previous memories (retrograde amnesia) and loss of the ability to form new ones (anterograde amnesia). Case-histories of the first kind, when for instance someone walks into a hospital unable to say who they are or what their address is, are very dramatic, but not very helpful for present purposes. Any brain trauma, such as a blow on the head, or electro-convulsive shock, is likely to blank out memory for events a few hours preceding the trauma, and it is possible to observe similar effects in animals (Miller and Springer, 1973). The second kind of ‘amnesic
syndrome’ in which people become permanently unable to report on the events of a few minutes past, is the one in which the hippocampus is implicated.
The most common cause of severe memory loss is alcoholism, or sustained heavy drinking. This may have a number of unfortunate consequences, one of which is the ‘Korsakoff syndrome’, which includes as a rule total amnesia for events during several years prior to its onset, and difficulties in retaining new information, as well as emotional disorders (Victor, Adams and Collins, 1971). The precise nature of the brain pathology underlying these deficits is uncertain. There are some individuals, however, in whom amnesia for current events has been apparently produced by misguided surgical intervention of a relatively circumscribed kind. These are people of normal intelligence, who suffered from epileptic attacks severe enough to persuade neurosurgeons in the 1950s that radical brain operations were worth trying. In a small number of patients, Scoville in Montreal deliberately removed the hippocampus from both cerebral hemispheres. The number was small because, although removing damaged tissue reduced the frequency of epileptic seizures, it was soon discovered that removal of the hippocampus also destroyed certain kinds of memory ability. One cannot be entirely sure about this, especially as rather a large proportion of what is said about the effects of the operation is based on a single patient, ‘H. M.’ (Scoville and Milner, 1957; Milner et al., 1968). But the evidence is fairly strong that interference with the inside surface of the temporal lobe, where the hippocampus is found in man and other primates, is likely to have very severe, and very bad, consequences for everyday memory capacities (Warrington and Weiskrantz, ,973).
The nature of H. M.’s memory after surgical removal of the hippocampus was as follows. He could report normally on his life before the operation (when he was 27) except for vagueness about the year or so immediately prior to it. He could perform previously acquired skills, such as reading, writing and his trade of motor-winding, and could recognise already known persons, including Scoville, who had done the operation. But he appeared to have difficulty in laying down any new memories, being unable to remember how to find the lavatory in the hospital where he convalesced, or to find a lawnmower he used regularly at the house he lived in afterwards (but whose address he never learned). In two important ways, however, it was apparent that new information could
be retained. First, short-term memory, or the persistence of immediate perceptions over a period of seconds and minutes, was normal. H. M. could carry on a conversation of sorts about his earlier life, and understand spoken instructions and questions, so that he could repeat back short strings of numbers and perform mental arithmetic. He could remember what he was looking for long enough to do jigsaw puzzles, even though he could remember little about a puzzle done the day before. Second, H. M. could show gradual improvement at simple learning tasks, such as tracing short finger mazes, or pencilling round shapes seen only as a reflection in a mirror. Given certain hints and cues about words or pictures in lists normally used to demonstrate the absence in long-term remembering, H. M., and similar patients, can remember about as such as normal subjects prompted in the same way (Weiskrantz and Warrington, 1975; Weiskrantz, 1977; Milner, 1970). One patient with an operation like H. M.’s happened to have played the piano, and could learn new pieces, and perform them later if given the opening bars, even though professing no recollection of the music, the time he had played it before, or the person he had then played it to (Starr and Phillips, 1970).
In these amnesic patients, then, previous experience can be incorporated into automatic routines and habits, without episodes of previous experience being accessible to conscious, verbally reportable, awareness. At its simplest, this amounts to retention of responses, and short-term persistence of perception, without any facility for reviving or retrieving recent perceptions. Also, mental skills, such as reading words written backwards, may improve in amnesic patients, even if they cannot remember having practised them (Cohen and Squire, 1980).
Is it possible to compare this syndrome with the behaviour of animals given experimental lesions of the hippocampal region? It was initially thought not, since one of the excuses for removing the hippocampus in human patients was that animals did not seem to suffer any ill effects after analogous operations. This is true only for the most cursory examination of the animal subjects, using the extremely ingenuous and naive assumption that if behaviour was modified in any way by experience, memory must have been all right. It is now the case that many parallels can be drawn between the behaviour of mammals with experimental lesions of the hippocampus and the clinical reports and systematic assessments of the abilities of amnesic patients (Weiskrantz and Warrington, 1975; Weiskrantz, 1977; Gaffan, 1976).
For some time it has been known that, although animals with hippocampal damage can be trained to perform simple habits to obtain food rewards, they differ from normal animals in persisting with the same habits long after the rewards have been discontinued (e.g. Douglas, 1967). This could have the interesting implication that memory of the recent absence of reward is one reason why the normal animals cease to perform previously rewarded habits, but alternative formulations, in terms of the automatic inhibition of responses, have been preferred (Kimble, 1968).
More direct evidence of the involvement of the hippocampus in the memories of animals can be obtained by the delayed-response and delayed-perceptual-comparison methods discussed earlier in this chapter (as I hope normal readers will remember). Sinnamon et al. (1978), for instance, tried a relatively straightforward delayed-response procedure with hippocampally damaged rats. Each day, the animals were allowed to find water behind one of four doors, which were recognisable by their position and surface appearance. Some time later, they were released to make a choice, water being given behind the same door on any given day. Rats that were normal in the sense of having had lesions to superficial cortical areas managed to make consistently correct choices at intervals of over two hours after initial discovery of the water location. But rats with certain kinds of interference to the hippocampus appeared unable to remember where the water was after delay intervals of 15 seconds or longer. O’Keefe and Nadel (1978) have published a monumental review of evidence suggesting that in rats the hippocampus functions as a repository for memories about places (The Hippocampus as a Cognitive Map). As places are the only things which rats usually can be demonstrated to have memories about, this is an adequate summary for that species. Where other sorts of memory can be shown, these too seem to depend on an intact hippocampus. The most powerful indications that this is so come from the procedures developed by Gaffan (1977a, 1977b) which I described earlier.
One of these procedures allowed a fairly close parallel to be drawn between tests given to human amnesics (Warrington, 1974; Huppert and Piercy, 1976) and the assessment of memory function in monkeys, since in both cases what was required was the classification of snapshots according to whether or not they had been seen before. Gaffan (1977a) succeeded in training rhesus monkeys to observe a sequence of coloured slides, and indicate when a given picture
occurred for the second time in the sequence (see pp. 307—11 above). Following this training, the animals were given a brain operation to eliminate output from the hippocampus. After the operation they were put through the same sequence of training as before. Like human amnesics such as H. M., they appeared to have satisfactory immediate recognition of what they had just seen, over very short intervals. If a view from the Eiffel Tower was shown twice in succession, or with only one intervening slide in between, the operated monkeys held back the first time, but pushed a panel to get their sugar pellet when it came up again, in the same way that they did before the operation. But, again like human amnesics, the brain-damaged monkeys seemed to have difficulties in identifying repeated slides when more than one other picture came between the initial showing and the repetition. Before the operation they had performed with few mistakes (misclassifications were less than 10 per cent) even when several other pictures might have been seen since the first showing of a repeated slide. Afterwards, in tests of only moderate difficulty in terms of the length of these intervening lists, errors reached about 20 per cent (the animals sometimes responded to the initial showing of a picture, when they should have kept still, and sometimes missed rewards by failing to respond to second showings).
In an alternative assessment of the retention of perceptions over short intervals, using the simpler stimuli of colours, very much the same thing was found with monkeys given the operation designed to disturb the functioning of the hippocampus (Gaffan, ,977b). In this case monkeys looked at a ‘list’ of one, two or three particular colours shown in succession, and then, when a test colour was presented a few seconds later, had to respond according to whether the test colour had been given in the previous list (see pp. 310—13 above). With only a single colour to remember, operated monkeys did just as well as normals, making few mistakes, but with lists of two or three, which were more difficult, the operated animals were distinctly worse than the others, especially at correctly identifying a colour which had been seen at the beginning of a previous list of three. (Overall, about half the mistakes took the form of responding as if a test colour had been on the last list, when it hadn’t, and in the other half colours which had been present were missed—this precludes simple response biases.) Exactly the same results were obtained when the spatial location of illuminated panels had to be recognised, rather than colours: monkeys whose hippocampus had been interfered with could remember which was the
last panel that they had seen lit up, and the last but one, but did much worse than the control animals when they had to remember the last but two.
These findings clearly suggest that the brain-operated monkeys had difficulties in detecting whether current perceptions matched or did not match those of the recent past, even though they were perfectly competent at detecting similarities between current perceptions and those immediately preceding them. A peculiarity is that the same sort of surgical intervention did not seem to have as much effect on a task which required monkeys to select objects on the basis of whether or not the objects had recently covered a desirable food object (a piece of sugared wheat; Gaffan, ,974). If 5 household objects had just been pushed away to uncover bits of sugared wheat, then monkeys with the brain operation were unable correctly to choose these 5 in preference to 5 others, when offered pairs composed of one of each, although unoperated control animals made very few errors. This would be expected if the operated monkeys had difficulties with recent memory. When 10 objects in succession were manipulated by monkeys, but 5 objects covered sugared wheat while the other 5 did not, this contrast in the availability of the food incentive meant that the 5 rewarded objects were more likely to be manipulated than the non-rewarded ones, when they were presented again singly. This preference was not very strong, but the peculiarity is that the operated animals had just as much of a preference as others.
One interpretation of the various abilities and disabilities seen in these ‘amnesic’ monkeys (Gaffan, 1974, 1976) is that they are bad at anything which requires a differentiation of the familiar from the unfamiliar that cannot be done on the basis of very recent, ‘short-term’ memories. This is a more limited deficit than one would expect on the basis of the human amnesic patients, but I must reveal now that, apart from the obvious problems of comparing human verbal memory with the choices of animals, all of the ‘amnesic’ monkeys discussed in this section had much more limited brain damage than that brought about in attempts to relieve epilepsy. In the monkeys, a small tract of nerve fibres that is one of the main connections between the hippocampus and other parts of the brain, the fornix, was cleanly transected. In H. M. and similar patients, the hippocampus itself was removed from both hemispheres, and it is likely that nearby parts of the inside of the temporal lobes were damaged as well. Especially given the prior history of tissue damage connected with epilepsy in the first place, it is
probable that anatomical abnormalities are more extensive in the human amnesic patients than in the experimental animals.
Problems of equating the extent of brain damage need not obscure the important fact that anatomical interference with hippocampal function has little or no effect on some behaviour, but produces a profound impairment of performance on certain tests. In animals, procedures which very explicitly involve moderately recent memory, memory for places, or assessments of the familiarity of visual displays, are sensitive to the effects of hippocampal damage. But response habits, and stimulus discriminations, trained up by selective reward and punishment, can be readily acquired by animals suffering from the same anatomical deficiencies. The implication is that sensory-motor skills and routines can develop in the absence of memory for previous perceptions, detection of familiarity, or whatever it is that the hippocampus usually does. In the human amnesic also, motor skills can be acquired, and with sufficient prompting it is evident that something about previous lists of words is retained (Weiskrantz, 1977), but the gaps in subjective memory are painfully obvious.
All this is useful in emphasising that memory is different from reflex-like habits. What we refer to as memory in man is usually subjective availability of mental content which we can, if we wish, comment on. The ubiquity of our verbal comments on our subjective states, and the absence of any such thing in experimental animals might be taken as an indication that animals possess only reflex-like habits, and have nothing which might correspond to our memories of previous events. But anatomical disturbances which cause subjective memory loss in man disrupt the ability of rats and monkeys to perform tasks designed to require some forms of memory. This strengthens the case that mammalian brains contain facilities for the retention of perceived information, which work in roughly the same way as the facilities which we ourselves may put to more elaborate uses. The practice of testing animal learning abilities by repeated and invariant routines, though perhaps convenient for other purposes, turns out to be singularly inappropriate for the study of memory, since it encourages automatic habits. These may form a substantial part of the mechanisms by which animals cope with experienced regularities in their natural environment, but the special tests developed to assess memory should not be regarded as necessarily unnatural. No one has tried it, but one would not be optimistic about the chances of a chimpanzee getting along in its natural environment after being given
the brain surgery which produces human amnesia. Certainly one would not expect such a chimpanzee to demonstrate memory for the locations of previously hidden food in experimental tests.
Even if we accept that there appear to be parts of the brain, of which the hippocampus is one, that seem to have something to do with memory in both monkeys and men, this does not take us very far towards specifying exactly what the hippocampus does. It has been suggested that it receives a version of events coded by the perceptual system, and serves mainly to distinguish between old and new perceptions (Sokolov, 1975; Gaffan, 1976). Individual hippocampal cells respond to novel, but not repeated stimuli (Vinogradova, 1975) so there may be something in this, although it is hardly an adequate explanation for the range of phenomena found in human amnesia (Weiskrantz and Warrington, 1975). However, a device for clocking in perceptions as they occur, tagging them for recency over minutes and hours, would in theory be useful in a number of ways, and if one considers the absence of any subjective sense of temporal order in the remembered events of the day, one can well imagine that other confusions would arise. It remains to be seen, however, whether Weiskrantz (1977) was correct in looking forward to the day when the same theory of brain function (and of hippocampal function in particular) can be applied to both man and monkey. There is considerable evidence that the hippocampus of rats, as well as that of primates, is used as a temporary or ‘working’ memory store, over intermediate time intervals (Rawlins and Tsaltas, 1983; Meck et al., 1984)
Phylogenetic development of memory and forebrain
The hippocampus is part of the limbic system of the forebrain, which is usually characterised as primitive, originally devoted to the sense of smell, and the seat of the baser emotions of man. The traditional story (Sherrington, 1906; Romer, 1949) is that the forebrain began as a centre of smell, acquired the dominant role of directing and organising the actions of early vertebrates because smell was the most important sense, and then attracted to itself the other sensory modalities, ultimately having to invent the cerebral neocortex, and grossly expand its physical proportions until, as in the human cerebral hemispheres, it enveloped and dwarfed all other parts. This story may have a happy ending, but the beginning and middle are suspect (see Chapter 5). It is
a fact that olfactory tracts go into the brain at the front, while sensory nerves for taste, touch and hearing go in at the back, and the output from the eyes goes to the middle. Originally this may have been a matter of anatomical convenience, given the layout of the respective sense organs. But in the lower vertebrates we know about (present-day sharks, other fish, amphibians and reptiles) it is not true that smell is the only input to the forebrain. Although sharks do indeed have an acute sense of smell, and large forebrains, only a small part of their forebrain appears to receive direct or indirect olfactory input (Ebbesson and Heimer, 1970; Graebner, r980)—about the same proportion as in the rat (Ebbesson, 1980). And at least one other sensory modality, vision, takes up some of the remaining space. The visual inputs to the forebrain telencephalon are less direct than the olfactory inputs, but we no longer need to localise brain function simply in terms of the point of entry of the sensory nerves.
The story of the telencephalon should thus be that once upon a time it received direct smell input and only indirect input from other senses. When it grew bigger and bigger in reptiles and birds, this plan stayed the same. And when it grew even bigger, in mammals and especially in man, the plan still stayed the same, but the indirect input of other sensory projections to the surface of the forebrain became just as good as the direct input to the smell centres (see Chapter 5, pp. 179—81).
Why then, did the forebrain grow at all, if it was not changing from a smell centre to a general purpose perceiver, having been general purpose all along? A nice story to finish this chapter with would be that the forebrain was always the best of all the parts of the brain at forming and storing memories, and so the forebrain grew bigger and better, as memories of what had happened before in the life of individual animals became more important than inherited instincts or modified reflexes.
Perhaps the forebrain was good at memories in the first place because of something about smells. Smells in the air (or water) tend to hang about, and do not usually change over time as rapidly as sights and sounds. One would think, on these grounds, that there would be less need for internal retention of olfactory sensations, though the slower pace of external changes might be matched in the receiving apparatus. Sherrington’s argument was that immediate smells may arise from objects at some distance in time and space, whereas for early vertebrates, living he supposes in muddy waters, vision, as well as touch, was only for the here and now. Smells needed to be used to direct actions towards their remote sources, and the receiving station
for smells was thus required to be a mechanism for retention and emotion rather than for reflex motor reactions. Such speculations take place in very muddy theoretical waters. My own preferred shot in the dark is that the cerebral hemispheres became important because they are not the first reception areas for other sensory modalities as much as because they are the point of entry for olfaction.
Just as, in perception, internal selection and organisation of sensory input can only take place beyond the sense organs themselves, so, for any kind of selective retention of sensory information over time, each extra stage of sensory processing has an advantage over the last. It is hard to imagine a monkey encoding a list of three colours in its retina, if the retina is designed to capture current visual images. Each new retinal image must wipe out the last—there has to be a later stage at which successive bits of visual information can be, metaphorically, ‘filed away’. Similarly the retina itself cannot retain the spatial location of an insect that has just disappeared out of sight, if at the same time it is registering a visual array in which there is an absence of insects. Something just beyond the retina might maintain orientation to a departed food object. But in general a memory store should be far removed from the immediate arrival of sensory messages, if the past is to be made independent of the present. Therefore there are theoretical grounds for assuming that the expansion of the cerebral hemispheres of the forebrain in birds and mammals should bring with it a greater capacity for storing perceptual descriptions and representations, in modalities other than olfaction. There is no doubt that in many mammals, and probably in the earliest ones, there are facilities for the storing of olfactory information as well, in the form of a sequence of olfactory processing within the hemispheres. But in birds the clear views of daytime flying, and the absence of restrictions imposed by nocturnal prowling, meant that the avian forebrain expanded while vision, rather than olfaction, became the dominant sense.
Whichever myth about the origins of forebrain powers one chooses to believe in, it is undeniable that our forebrain is very large, and we have supposedly unlimited, if fallible, memories, while an animal such as a toad has a very small forebrain, and shows few signs of being influenced at all by its previous experience. This in itself might initiate the idea that forebrains go with memory. It should be noted that there is no such thing as a large vertebrate brain with extremely small cerebral hemispheres (see Chapter 2). But we need behavioural evidence, and it would be best if we could show that measured
capacities for memory increased across species with forebrain size, and that these capacities can be selectively reduced by experimental interference with forebrain function.
Forebrain components and unity of function
I have already described, a few pages ago, evidence which suggests that in man, and probably in other primates, a small part of the cerebral hemisphere, the hippocampus, plays a special role in some kinds of remembering. Structures in the interior of the forebrain, which together with the hippocampus are identified as the limbic system, are usually thought to be primarily concerned with emotional drives and the regulation of bodily needs. (In a way, emotions function as a form of memory for provoking stimuli.) The surface cortex of mammalian hemispheres is largely taken up by areas devoted to specific sensory modalities, and subsequent perceptual analysis, as I emphasised when discussing perception (Chapter 4). Especially in primates, other regions of the cerebrum appear to be specialised for the planning and execution of motor acts, and this by no means exhausts the-list of duties which the forebrain may be supposed to discharge.
Concentrating on the relation between forebrain function as a whole and memory is thus a considerable oversimplification, especially since memory in this context is vague and undefined. The practical reason for being so vague is that some of the experimental evidence concerns such things as the effect of wholesale removals of most or all of the forebrain on the behavioural capacities of a given species. The theoretical reason for examining general as well as specific aspects of forebrain mechanisms is that distinctions between such things as perception, memory and emotion are to some extent artificial abstractions. One experimenter will lesion a part of the limbic system of rats and discuss the effects of this on subsequent aggressive behaviour, without otherwise assessing adaptation to familiar surroundings, while another will make systematic tests of delayed-response abilities in animals with similar forebrain damage, only mentioning in passing that the rats struggled and bit when handled. Animals do not have special parts of their forebrain designed for performance on experimental tests of memory and perception; abilities to perceive and remember, if they are present, exist because they are useful in the course of fighting, fleeing, feeding and
philandering. The assertion to be judged is that the integration o experience and action over time by the retention of perceptual information is a characteristic of forebrain function. This is hardly the only thing one would want to know about the brain, or about memory but it seems to me to be a reasonable claim to examine, even if the complexities of both anatomy and behaviour might appear to rule out anything so simple-minded.
Beritoff’s review of memory for food locations
A very clear claim was made by Beritoff (1971, p. 116): ‘image memory is a function of the forebrain’. I described some of the findings on which this claim is based earlier on in this chapter—the time intervals over which animals appear to remember the location of food appear to lengthen in accordance with a crude phylogenetic scale of forebrain development. Goldfish swim back to a place where they have just been eating for only up to 10 seconds after they have been removed from it, whereas baboons who watch a piece of apple being hidden outside their cage may go directly to fetch it if they are released up to an hour later. The details of comparisons between species may be dubious, and the exact time values should not be taken too seriously, but these tests provide some sort of support for the theory that memory is useful and species with large forebrains can make more use of it than others.
This is less likely to be a spurious connection if damaging the forebrain prevents all the animals remembering food locations in the usual way. Beritoff(,97,) says that this is in fact the case, and quotes experiments in which goldfish with their forebrains removed, though eating and swimming normally, no longer swam back to a place where they had just been fed. Turtles, lizards and chickens with substantial forebrain ablations could find food a few seconds after it had been shown to them, but not after the delays of two or three minutes which could be negotiated successfully by normal animals of these species. In cats and dogs ‘image memory is exclusively a function of the neocortex’— total removal of all sensory and motor areas of cerebral cortex stops the animals spontaneously running to a particular feeding place permanently, and lesions to secondary projection areas in a particular modality temporarily prevents the direction of movement toward previously experienced spatial cues in that modality.
This evidence is rather limited in scope. Beritoff, of course, goes into more detail than I have here, and discusses experiments on the
avoidance of places where animals have recently been given electric shocks, in which the results correspond roughly to those obtained with the food locations method, as well as various other behavioural tests of the conditioned-reflex type, whose results are more difficult to interpret. Many of the experiments are open to technical criticisms: for instance there does not generally seem to be any account taken of the possibility that changes in behaviour after forebrain damage could be due to impairments of olfaction. But there appears to me to be some sense in his overall conclusions: vertebrate memory takes various forms; ‘image memory’ depends on the forebrain; this and other kinds of memory vary from species to species and develop phylogenetically so that memory is better in monkeys than in fish.
Forebrain functions in non-mammals
Few would deny that human forms of memory depend on the integrity of the human cerebral hemispheres, and there is a considerable measure of agreement that particular parts of the human forebrain have particular functions that have to do with memory in one way or another. Awareness of even momentary visual experience is eliminated by damage to the visual cortex, satisfactory recall of recent visual experience is disrupted by interference with the hippocampus, and so on. Working backwards to other primates, it is reasonable to claim that there is some correspondence between human and monkey forebrain components (Weiskrantz, 1977). Although there is less experimental evidence to support detailed analysis of the function of individual forebrain components in other mammals, most current theories of the role of the neocortex of the mammalian hemispheres make little sense unless one assumes that some of the information collected and analysed by the cortex is stored for later use.
There is even less evidence, apart from Beritoff’s work, which is appropriate for applying to the hypothesis that the forebrain in other vertebrates functions so as to retain perceived information. For birds, the investigations of short- term memory in the pigeon, and the apparent use of strategies requiring detailed knowledge of previous events in the formation of learning sets by mynah birds and jays (Kamil and Hunter, 1970; Kamil et al., 1977) imply that significant perceptual retention takes place. There are experiments on the effects of forebrain lesions on forms of learning in birds (Stettner, 1974; Zeier, 1974; Macphail, 1975) but none that I know of that directly examine which, if any, structures in the hemispheres of birds are necessary for
such tasks as delayed perceptual comparisons (Roberts and Grant, 1975).
Even Beritoff reports no experiments on the effects of forebrain lesions in amphibia—there is very little sign of any form of behavioural modifications due to experimental experience in normal frogs anyway. There are, however, large numbers of papers on the effects of forebrain lesions in fish (see Chapter 5, pp. 185—8 and Savage, 1980) and a rather smaller number of similar investigations with reptiles (Peterson, 1980) none of which permit any strong theoretical conclusion. Generally speaking behavioural effects of forebrain lesions in lower vertebrates are notable by their absence, in standard tests of locomotion, feeding and simple forms of conditioning and learning. More complex natural behaviours, such as nest- building, courtship and aggressive social interactions, are more likely to be affected. But it is worth mentioning a few isolated results which suggest comparisons with mammalian forebrain function.
Forebrain lesions in reptiles
Lesions of the temporal lobe in mammals which involve the amygdala, a subcortical limbic structure associated with the hippocampus, produce placidity, lack of aggression and sexual and feeding activity involving inappropriate objects (Kluver and Bucy, 1937, 1939; Rosvold et al., 1954; Plotnik, 1968). Roughly similar results are obtained by damage to the putative analogue to the amygdala in mallard ducks (Phillips, 1964). Keating et al. (1970) and Tarr (1977) have extended the series to reptiles (Caimans and iguanid lizards). Caimans with amygdaloid lesions did nor show the usual attack or retreat reactions to the human experimenters except under extreme provocation (Keating et al., 1970). The lizards were studied in groups in small enclosures, in which they normally established dominance hierarchies (‘pecking orders’) maintained by aggressive encounters and displays and the adoption of restricted territories by individuals. After suffering lesions in the forebrain amygdaloid areas, a lizard would become socially indifferent, not making aggressive displays itself and ignoring those of others (Tarr, 1977). This perhaps ought to be interpreted as a disturbance of aggressive instincts, although I would prefer to put it down to a failure of social perception. Greenberg (1977) found that a slightly different type of forebrain lesion also stopped adult male lizards from issuing the normal challenge display to other adult males intruding into their personal territory, and deduced from
elaborations of the experiment that the lesions produced a failure to register the social cue rather than interference with the execution of display movements. These findings, among others, imply that the forebrain in vertebrates generally is an organising centre for inherited forms of perception and emotion.
In so far as personal territories are recognised by individual reptiles one could argue that the forebrain is the repository for cognitive maps, but it is a rare thing to find any direct indication that reptile hemispheres mediate any form of individually acquired knowledge. However, Peterson (1980), in the course of a review of all work on reptile hemispheric function since 1916, revealed that she had performed an experiment which suggested that certain forebrain lesions prevent maze learning in desert iguanas. (Burghardt, 1977, could quote only four reports of any experiments showing the learning of turns in mazes by lizards, so the lack of previous evidence of forebrain involvement in this task does not indicate negative results.) A large part of all methods of directing behaviour in lizards is devoted to temperature control: this Peterson turned to the advantage of psychological investigation by using a hot plate (at 40 degrees centigrade) as the goal for the desert iguanas. With this method, it had been shown that desert iguanas could rather laboriously learn simple mazes of the Hampton Court type (Julian and Richardson, 1968). If, after having done this, such animals had operations to destroy the dorsal cortex on the top surface of the forebrain hemispheres, they could no longer find their way around the maze they had previously learned, nor did they show any signs of recovering this ability in long training after the operation. But, operated animals could learn perfectly well to perform a simpler task of moving into a white rather than a black compartment, for the same reward of lying on a hot plate (Peterson, 1980). This may not be much to go on, but it provides a welcome chink in an otherwise blank wall. The dorsal cortex of reptiles has on anatomical grounds been thought to represent an early version of the mammalian cerebral cortex, and it would be encouraging if a reptilian dorsal cortex is necessary for some kinds of spatial learning, especially if it is required to direct movement toward relatively remote goals.
Forebrain lesions in fish
It is something of an embarrassment for any theory which proposes a gradual phylogenetic development of psychological abilities that
common aquarium fish are easy to train and perform well on most of the tests of learning capacity designed for mammals (e.g. Bitterman, 1965; Bitterman and Woodward, 1976). The ease of training fish has meant that (despite the ticklish nature of the surgery) there is a large amount of data concerning the effect of forebrain lesions on behaviours acquired as a result of individual experience. Savage (1980) suggests that no theoretical conclusions can be reached, but much of the data is consistent with the proposition that retention of recent perceptions is impaired by forebrain lesions (Savage, 1968; Flood and Overmeir, 1971; Flood et al., 1976). Fish with much of their forebrains (the telencephalon) removed have difficulty in learning mazes (Hale, 1956; Warren, 1961). In many experiments where there are delays within episodes of experience (after a choice, before the reward) or between these episodes (delays between successive runs in a maze, or between trials when fish are trained to swim away from a signal to avoid electric shock), forebrain-damaged fish are more affected by the delays than others. As a rule, fish with forebrain lesions are compared with others with intact brains, but disabled olfactory organs, to ensure that impairments are not due to lack of the sense of smell. In one such case, where the procedure used comes close to a direct test of memory, goldfish first received electric shocks preceded by a signal light (Farr and Savage, 1978). Normal and forebrain-ablated animals show Pavlovian conditioning in these circumstances, as measured for instance by heart-rate changes or swimming activity in response to the light (e.g. Overmeier and Curnow, 1969). Now, however, the fish were allowed to swim in a ‘T’ maze, the signal light being placed at one side of the ‘T’, but with no shocks or any other motivating events applied. Fish without forebrain removal learned to turn towards the side of the maze away from the light but those with forebrain ablations did not. It thus seems that the successful goldfish retained what had been learned from the association between the light and pain in a form that was capable of influencing subsequent choices, whereas forebrain lesions prevented this happening. There are many ways of putting this, and It can certainly be labelled as an aspect of conditioning (‘secondary reinforcement’; Flood et al., 1976). But it is inescapable that some information was retained from the time of the initial light shock pairings, until the time of the ‘T’ maze tests. The dispute is whether it is retained in the form of a response habit, an emotional connection or as a representation of events as they occurred. This may not be resolvable, but it appears that something can be retained with a
forebrain that cannot be retained without it. Since simple response habits appear to be formed and retained in the absence of the forebrain, it is arguable that, even in the goldfish, the cerebral hemispheres make possible the retention of something more complicated.
The retention of perceived information and the modification of habits
It is time to look again at the distinction between knowledge and habit. I have mentioned this several times already, and at the beginning of this chapter I especially invented a hypothetical animal that could alter its behaviour in various ways as an adjustment to life experiences, using only very rudimentary neural mechanisms of habit formation. With a circuit of only two or three neurons, it could habituate to repeated stimuli, develop Pavlovian conditioned reflexes, and react selectively on the basis of previous reward and punishment. That is, it could do these things under certain very restricted circumstances. I bring up this fairy-tale creature again now, in order to emphasise that these elementary categories of adjustment to past experience by themselves are of very limited value in examining psychological capacities. An elaborate central nervous system is not necessary for accomplishing the bare essentials of adaptation to repeated stimulation and stimulus-response associative learning. The waning of fixed instinctive actions as a consequence of repeated elicitation is found throughout the animal kingdom (Hinde, 1970). Changes describable as Pavlovian or Skinnerian conditioning occur in various worms and in the amputated legs of cockroaches (Fantino and Logan, 1979; Horridge, 1962; Eisenstein and Colton, 1965; Distenhoff et al., 1971).
It should not come as any surprise, then, that, as well as vertebrates created with very little forebrain, others, who have a lack of forebrain tissue thrust upon them by experimental surgery, are capable of demonstrating conditioned reflexes and limited motor habits stamped in by reward and punishment. Oakley (1979a) has given a most helpful account of his own and others’ findings of this kind with mammals. Some claim that the isolated spinal cord can ‘learn’. Mild electric shock to the thigh of a human paraplegic elicited urination after Pavlovian pairings with a current delivered to the abdomen strong enough to do this reliably (Ince et. al., 1978). Spinal rats appear to learn to flex a leg if stretching the leg causes shock (Chopin and
Buerger, 1975). Cars with the entire forebrain removed learn to blink their eyes at the sound of a high-pitched tone, if the high tone, but not a low tone, signals an electric shock to the cheek, Rabbits with the neocortex of the cerebral hemispheres removed (leaving the hippocampus, thalamus and basal ganglia) readily and flexibly form conditioned eye-blinking reflexes to tone or light signals paired with electric shocks to the cheek (Oakley and Russell, 1977). With sufficiently arduous or specialised training programmes, rabbits with this radical forebrain operation press a treadle many times for each food pellet received in a Skinner box (Oakley, 1979b; Oakley and Russell, 1978). Walking, climbing, swimming, grooming, and eating (including picking up food in the paws) is only slightly disturbed in rats with both neocortex and the hippocampal formation destroyed (Vanderwolf et al., 1978).
All of this shows that in mammals the forebrain, or a major part of it, is not absolutely essential for certain types of behavioural phenomena. It does not show what, if anything, the forebrain in mammals is absolutely essential for. Unfortunately, much more effort has been spent on showing what animals can do in the absence of forebrain structures than on what they cannot do. It is a fairly safe bet however that mammals deprived of cerebral cortex could not cope with anything approaching natural conditions. In decorticated rats, ‘food hoarding and social behaviour were essentially abolished’ (Vanderwolf et al., 1978). Decorticated rabbits trained to press a treadle for food fail to press the same treadle if it is moved a short distance from its original position (Oakley, 1978). Forebrain damaged mammals are thus only capable of forming very stereotyped motor habits and exhibiting instinctive response patterns such as locomotion and grooming. Oakley (1979) suggests that, unlike normal mammals, they do not know what they are doing, and cannot form conscious images which would allow them to reason. Although he is apologetic about using these terms, it would be foolish to beg the question by quibbling about their applicability. I do not believe that a normal rabbit has a subjective life very much like mine or yours, but we have to say something about how its mental processes differ from those of a spinal cord, or those of a rabbit with no sensory projections beyond the thalamus. I prefer to say that the normal rabbit has remembered perceptions, expectancies and so on, while an animal with little or no forebrain has only habits and reflexes. This may be inadequate, false or even meaningless, but it can hardly be more misleading than the
assumption that there is no psychological difference between a spinal cord and a complete brain, or between a goldfish and a chimpanzee, because all are capable of forming conditioned reflexes and habits (Fantino and Logan, 1979; Macphail, 1982).
Animal memory —conclusions
Vertebrate animals typically absorb information about the reality that surrounds them. It is possible to imagine creatures whose reactions can be exactly predicted by what they sense of the immediate present, but for real animals one would always have to know about previous events in order predict current reactions. Some sorts of predetermining conditions could clearly be important without raising the question of memory—a tired and well-fed beast is predictably different from an alert and hungry one. But if the tired animal finds its way back to a particular resting place, or the hungry one forages or hunts according to a previous pattern of success, we have to ask what psychological processes enable previous experiences of that individual to direct its behaviour.
I have drawn - a crude theoretical distinction between habit and memory, habits being automatic reactions whose form may have been determined by previous practice and past successes and failures, and memory being distinguished by its independence from particular habits. In terms of experimental tests, the best distinction is that memories may be unique whereas habits are always repetitive. If an animal responds according to exactly where food had been on one particular previous occasion, it usually requires us to assume that something has been remembered about the unique occasion (although we may have had to train the animal habitually to remember food locations in order to make this demonstration). There are various other experimental methods which make it possible to prove that, in principle, animals retain particular features of perceived events in the recent past.
If the perceptual apparatus of an animal is used to construct descriptions of what is perceived, in order to detect objects and things rather than isolated sensations, then one would certainly expect continuity and discontinuity with time to be a necessary part of the descriptions. However, more complicated brains would be needed to construct such descriptions, and to store them over time, than would be needed to react to a simple stimulus, such as a bright light, and to
modify reactions according to whether previous reactions to bright light had been strengthened by the ingestion of food, or such like. There is reason to believe that the cerebral hemispheres of vertebrate brains are particularly important for, among other things, the utilisation of memories as opposed to such simple habits. Behavioural evidence is distinctly patchy, but any vertebrate whose forebrain is removed is likely to be measurably less influenced by retained perceptions of previous events, although simple automatic habits may still be gradually acquired. In men certain parts of the cerebral hemispheres seem to be identified with particular aspects of memory: damage to the inner sides of the temporal lobes may leave intact information of long standing and the perceptions of the last few seconds, but prevent access to the events of previous days, hours or minutes. The same structures of the forebrain also serve memory functions in other mammals. It is probable that the memory capacity of animal species varies according to forebrain development, so that lower vertebrates such as fish and reptiles are very much less influenced by information retained from previous perceptions than are birds and mammals.