The Vertical Lobe and Touch Learning in the Octopus

Since the work of Sanders & Young (1940) on Sepia it has been known that the vertical lobe is concerned in learning, and a considerable number of studies has been made to define its function more closely (reviews, see Wells, 1962; Young, 1961). It is probably fair to say that its function is still not known, though several theories of the nature of its action have been produced. The present account is a review of the situation with regard to touch learning, and includes a number of previously un-published experiments. The results are discussed in relation to current theories about the function of the vertical lobe in visual learning.

All the experiments were made with Octopus vulgaris, Lamarck, from the Bay of Naples. The animals varied from about 250 to just over 500 g. in weight; each was kept by itself in a 100 × 60 × 40 cm. asbestos tank with circulating sea water. For all but a few ‘repeated presentation’ experiments (see p. 248 below) the animals were blinded by section of the optic nerves or, in a few of the earlier experiments, by removal of the optic lobes. The latter operation, although it removes a considerable proportion of the supraoesophageal brain, does not appear to affect touch learning (Wells & Wells, 1957a). Some of the animals had other parts of the supraoesophageal lobes removed. Operational techniques have been described elsewhere (Wells & Wells, 1956, 1957a). Fig. 1 shows the relative position of lobes mentioned in the text. The extents of lesions to these parts were assessed from serial sections prepared using a modification of Cajal’s silver method given in Sereni & Young (1932, method B). These sections show: (1) that there is no detectable regeneration of parts removed within the period of the experiments ; (2)that it is almost impossible to remove the wedge of vertical lobe lying immediately behind the median superior frontal without damage to the superior frontal; (3) that while it is relatively easy to estimate the proportion of vertical lobe removed because this region has well-defined boundaries and nuclei that, in contrast with those from surrounding tissues, stain characteristically dark red or black with Cajal, it is another matter to achieve accurate estimates of lesions in, for example, the sub-vertical lobe. The distortions that often follow gross removals mean that estimates by comparison with controls of similar size are unreliable, and the cell types are sometimes insufficiently different to be identified with certainty. Something of this difficulty presents itself in estimating lesions to the lateral superior frontal. Here, the cells bounding the lobes are distinct and the neuropil differs from that of the median superior frontal in being more obviously patterned in organization; but the two are broadly confluent and there is no clear boundary between them where the neuropils merge. In Table 1, which summarizes the lesions in the animals used for the experiments, estimates of vertical lobe removal can therefore be regarded as reliable, while the extent of lesions in the median superior frontal is less certain, though probably accurate within the stated limits. Lesions to the lateral superior frontals are rated as being greater or less than 50% removed, and damage to other parts is simply recorded as having occurred (it was normally slight in these experiments), without attempts to assess its extent. For present purposes these estimates are accurate enough.

After the operations the octopuses were allowed several days to recover, during which they were fed upon crabs or pieces of fish several times a day. They were then (again with the exception of the ‘repeated presentation’ experiments) trained to discriminate between two cylindrical Perspex objects, Pi and P4. Pi was roughened by a series of grooves cut along its length and across the ends; P4 was left smooth. The octopuses had no pre-experimental experience of the objects and there was no pre-training to take the test object used as ‘positive’ in subsequent discrimination training. In a survey of the scores in the early trials of forty-two training experiments with blinded but otherwise unoperated controls it was not possible to detect any general preference for the rough or the smooth object, although the performance of individual animals sometimes indicated a predilection for one or the other, perhaps as a result of previous experience with rough or smooth bivalves in the sea before capture.

In training, the test objects were presented to the blinded octopuses by touching one object at a time against one of the arms. An animal was rewarded with a piece of fish for grasping and passing the ‘positive ‘object under the interbrachial web towards the mouth, and punished by means of a 6-9 V. a.c. shock to the arms when it did the same with the ‘negative’ object. Shocks were given through electrodes attached to a probe, the negative object being pulled away by means of a line attached to a screw in the top of the cylinder as soon as the shock had been given. Pieces of fish impaled on a wire were presented to octopuses taking the positive object, the object itself being pulled away as soon as the octopus had taken the reward off the wire and passed it under the interbrachial web to the mouth. Typical animals took all of the objects presented at the start of an experiment and learned to reject the negative object by thrusting it away. In the course of training, responses to the positive object generally became more rapid and certain, a change in behaviour that is obvious enough to an observer, but difficult to quantify. This matter has to be borne in mind when considering the results of tactile experiments, which sometimes appear to show that all learning is concerned with responses to the negative object. For most of the experiments there were 40 trials per day in two groups of 20 not less than 6 hr. apart. In each group, individual trials were at intervals of 5 min., arranged thus: + — + — + + — — +- — + — — + + — + — + In a relatively small number of experiments, different training techniques and trial sequences were used. These are specified in the text.

In various experiments between 1957 and 1963, 35 octopuses were trained to discriminate between Pi and P4 after removal of the vertical and/or median superior frontal lobes. In parallel with these animals 42 blinded but otherwise unoperated controls were trained. The results are compared in Fig. 2. The operated animals on average required more trials to learn the discrimination than the controls (mean 4-2 groups of 20 trials to reach 75 % correct responses, cf. mean of 2-3 by controls ; Mann-Whitney P = < 0·001). Within the operated group there was no significant difference between animals having the vertical lobe alone removed, and animals having both the vertical and median superior frontal lobes removed. The four animals with the median superior frontal alone removed also fell within the same range as the others (Fig. 2a).

Clearly, lesions in the median superior frontal/vertical lobe region interfere with touch learning. There is no obvious correlation between the nature of the lesion and the number of errors made (Table 1). In considering further whether the operated animals were, as a group, more variable than controls, it is important to take into account that they required, on average, more training trials to attain a given standard than controls, an increase that, of itself, would have tended to create a greater arithmetic variability in the experimental group. The variability of controls and experimentáis cannot be compared with the data in their crude form, because of the skewness imposed by the barrier at zero groups. This tends to reduce the variation in both series, but particularly in the controls, which averaged only 2-3 groups of trials to reach 75 % correct responses. In Fig. 2 b, the data have been transformed to a log scale after which the score distributions of controls and experimentáis are both approximately normal, and can legitimately be compared by tests such as ‘t ‘tests which presuppose a normal distribution. After the transformation, a ‘t’ test for the significance of the difference in the mean scores of controls and experimentáis gives a probability value of the same order as the Mann-Whitney non-parametric test on the crude data P = <0·001, (see Fig. 2). It is now also possible to test for a difference in variance, and when this is done (Fig. 2), it becomes evident that the variability of the experimental animals was indeed somewhat greater than that of the controls—in other words that the appearance of much greater experimental variability in the raw data is not entirely an artifact produced by the skewness of the data added to the greater arithmetic mean score of the experimental animals. The difference in variance between controls and experimentáis is significant at about the 1 % level (Snedecor F test, see Fig. 2).

It is important to know whether the effect of lesions to the vertical and superior frontal lobes is specifically related to positive or negative learning. In touch experiments, typical animals start by taking all of the objects presented to them. This is not surprising, since it is how the animals normally feed. As training continues, there is a steady decrease in the number of negative errors per group of 20 trials. Compared with the speedy reduction in the number of errors due to taking the negative object, positive errors show little change. There may even be a slight increase in positive errors in the early phases of training (individuals vary somewhat in this respect), and in general the rate of decline of positive errors is less than that of negative errors. As a result, the ratio of negative to positive errors varies somewhat throughout training, tending to be highest when the animals are making most mistakes.*

To find out whether experimental animals make an unusually high proportion of positive or negative errors it is therefore necessary to compare their pattern of erring with controls at times when the two types of individual are making the same total number of mistakes. This is done in Table 2, which considers the proportion of negative errors made by 42 controls and 35 experimentals in 242 groups of trials where the total number of errors was the same. There would appear to be no difference between controls and experimentals, whether with the vertical or the vertical and superior frontal lobes removed. Both erred mainly by taking the negative objects, and in both the ratio of negative to positive errors remained approximately the same (70-80 % negative errors) whatever the absolute number of errors made in a group. The only obvious exception to this is in the 10-mistakes category, where the errors were practically all negative. Such groups occur almost only at the beginning of training, when the animals tend to accept everything that is presented to them (see above). At this stage training has had no detectable effect, and as soon as discrimination begins at all the proportion of negative errors drops towards the 70-80 % level and stays there.

The trial sequence, used, + — + — + + — — +—, + —,— + + — + — + —, has the advantage that the first and second halves of the series each contain 5 positive and 5 negative trials. Performance in the first half and the second half of each group of 20 can be compared directly. This is done in Table 3. It can be seen from this that during the period of training up to and including the first group in which 75 % or more correct responses were made: (1) the animals made fewer mistakes in the second half of each 20-trial group, i.e. they learned within groups, the controls more than the experimental animals; and (2) the pattern of errors changed within each group, with a fall in the negative/positive ratio in the second half; that is, the animals became more cautious about taking objects presented to them, having received electric shocks for incorrect responses to the negative in the first half of the group. Since they erred in any case mainly by taking objects that they should have left alone, this depression of the tendency to respond positively improves performance by reducing the overall probability of a positive response towards 50%, at which level any tendency to discriminate is most likely to become apparent.

Comparing controls with the two types of experimental animal there are again no significant differences in the pattern of errors (Table 3). Whatever the nature of the effect of vertical lobe removal (or vertical plus superior frontal lobe removal) it is evidently not exclusively concerned with either positive or negative learning. The experimental animals make more mistakes than controls, and they do so in all possible ways.

A measure of the persistence of memories between successive groups of trials can be obtained by comparing the performance in the last half of a group with that in the first half of the next group. If the effect of training persisted without decay, one would expect the half-group immediately following the break to contain, on average, fewer errors than the half-group preceding it. In Table 4 a comparison of controls and experimentals is made in terms of whether the animals made more or fewer errors when training was resumed following breaks in training. The results are as might be expected. During the period of initial rapid improvement in performance, controls regularly made fewer errors in the 10 training trials following a break than in the 10 before (Table 4 a). It would be surprising if this were not so, since the animals reached a 75 % correct criterion of correct responses within two or three groups of trials. Experimental animals also improved steadily, but the effect was less well marked because they took a greater number of trials to learn.

It is in the period after the attainment of 75 % correct responses that the performance of controls and experimentals begins to differ in these experiments (Table 4 b).

The experimental animals are at this stage approaching the maximum accuracy of which they seem to be capable, and in consequence cease to show regular improvement. They appear to lose the capacity to discriminate during the breaks, and rapidly relearn during each group of 20 trials. The effect is particularly marked following long (overnight) breaks in training. There are two possible explanations of this. One is that in the operated animals the capacity to distinguish between the objects fades. The other is that the animals overwhelmingly drift towards positive responses during breaks in training, so that although the capacity to discriminate remains, it is masked. The 29 animals with which training was continued after reaching 75 % correct responses made too few errors for a conclusive analysis of this to be made. The data, such as it is, is presented in Table 5, in. the same form as in Table 2, and shows that on occasions when the experimental animals made the same number of errors as controls they did not make an unusually high proportion of negative errors.

The question of the importance of the swing towards positive responses during breaks in training has been considered elsewhere in relation to the performance of octopuses in retention tests (Wells & Wells, 1958b). Briefly, the situation is that octopuses trained to distinguish between P1 and P4 continue to discriminate between these in unrewarded retention tests up to at least 10 days after the end of training, and would probably do so for much longer. This applies both to controls and to animals with the vertical (or vertical and median superior frontal) lobes removed. In the tests each animal was trained until it reached a standard of 75 % or 85 % correct responses in a 20-trial group. It was then overtrained for a number of trials equal to the number required to reach this standard. Training ceased, and the animal had no further experience of the test objects until 5 or 10 days later, when a series was presented to it, exactly as in training, but without rewards or punishments; in. some cases the animals were again tested after a further 10 days. The results are summarized in Table 6. There was little difference in performance between animals with brain lesions and controls.

The tests were, however, not wholly satisfactory in that the overtraining procedure, intended to ensure an equal accuracy of performance by the end of the training period, failed in fact to do this. The controls, in general, continued to improve after the attainment of 75 % or 85 % correct responses, while the experimental animals did not. In comparing the results of retention tests it has therefore to be remembered that the two types of individual commonly start from somewhat different base-lines—a direct comparison of the decline in performance in the two cases can be criticized on the grounds that an increase from (for example) a mean of 2·9 to 7·4 errors (experimentals —difference 4·5) indicates a greater loss of memory than an increase from 2·3 to 6·8 (controls—difference again 4-5, see Table 6). Fig. 3 is an attempt to overcome this difficulty by comparing individuals starting from the same standards. It looks as if the experimental animals show slightly more decrement than controls in the 10-day retention tests, but there are clearly not enough data to be sure of this.

What retention tests do show quite clearly is that a capacity to distinguish between objects touched can very readily be masked by an overall drift towards positive responses. In several instances, among both controls and experimental animals, an initially low score due to taking nearly all of the objects presented was improved simply by repeating the tests. In other cases a similar improvement was brought about more rapidly by giving an electric shock wherever an object (of whatever type) was presented. Some examples are given in Fig. 4. There was also some evidence that low scores due to indiscriminate refusal of objects presented could be raised by feeding half an hour or so before a series of tests (Wells & Wells, 19586). These phenomena were observed both in controls and in experimental animals ; there was no indication of an accelerated rise in the probability of positive responses in the experimental animals.

The responses of animals trained to distinguish between P1 and P4, taking one and rejecting the other, can be reversed by further training. In some early experiments (Wells & Wells, 1957 a) it was found that 5 octopuses with the vertical lobes removed actually reversed faster, following a 36 hr. break in training, than 5 unoperated controls. At the time this result was attributed to more rapid forgetting of the initial training by the experimental animals, an explanation since shown to be unlikely by the results of retention tests (see above). In 1961 and 1963 a further 16 controls and 19 animals with the vertical and/or superior frontal lobes removed were trained.

In this larger group the experimental animals learned and reversed their training more slowly than controls (Fig. 5). The apparently anomalous result obtained in 1957 has two possible explanations. One is simply that in the small samples used the controls happened by chance to be an unusually poor lot (they took longer to learn and reverse than in 1961-63) and the experimentals happened to be unusually good (they learned and reversed more swiftly than in 1961-63)— the difference between the two is in any case not significant at any acceptable level. The other possible explanation is in relation to a detail of the training technique used. In 1957 rejected positive objects were re-presented to the animal together with the reward. In 1961-63 they were not; if the animal rejected the positive it simply failed to get fed and no further action was taken. It is arguable that the unusually large number of trials required to teach the controls to reverse their responses in 1957 is attributable to a technique that, in effect, taught the animals to reject all objects presented. By doing so they avoided shocks, and continued to get fed anyway. In initial training, where the octopuses generally began by taking everything, this possibility was unlikely to alter performance. In reversal, where the octopuses commonly at first rejected a high proportion of the positive (ex-negative) objects, it could have had a selectively adverse effect on the more rapidly-learning controls.

In 1961, some of the animals, 6 controls and 3 experimentals, were used in experiments involving more than one reversal. As before, training was stopped whenever an animal made 75 % or more correct responses in a group of 20 trials. Retraining, with the rewards and punishments switched, was begun 36 hr. afterwards. The results are shown in Table 8. They show only that experimentals and controls can be caused to reverse their responses more than once.

Octopuses from which the vertical and/or superior frontal lobes have been removed after training to discriminate between P1 and P4 do not entirely lose their capacity to separate the objects, although their performance is considerably worsened as a result of the operation and under some conditions may even be so bad as to conceal all traces of the capacity to discriminate (Wells & Wells, 1958 a). Persistence of the effects of pre-operational training has been shown, both from the savings when the animals are retrained and from the performance of pretrained animals in post-operational retention tests.

Fig. 6 summarizes the situation with regard to retraining after operation. Animals trained before the operation required fewer trials to achieve a standard of 75 % or more correct responses than octopuses without pre-operational training.

Fig. 7 summarizes the results of 20-trial retention tests carried out 36 hr. after operation following the end of training. Some of the animals were trained at 8 trials per day (4 + 4 —), others at 40 trials per day (in two groups of 20, each 10 +, 10 —). In fig. 7 the scores in the last 16 or 20 trials are compared with the scores in retention tests. The scores of the operated animals are considerably worse than those of the controls. This is not simply a result of anaesthesia or opening the cranium ; controls had dummy operations. It is not, moreover simply due to the experimental animals indiscriminately taking a higher proportion of the objects presented, for this would lead to a rise in the proportion of negative errors, which is not found. Thus in 72 preoperational training trials the 4 controls totalled 15 errors, 5 of these negative errors. In a corresponding 204 trials the 11 animals destined to be operated upon made 20 errors, 10 of these negative errors. Post-operationally, the 4 controls made 4 errors in 80 retention-test trials, 3 of these negative errors, while the 11 experimentals, in 220 trials, made 62 mistakes 36 of these due to taking the negative object. The numbers are small, but there is clearly no great difference between controls and experimentals in the pattern of errors.

The poor post-operational performance of the operated animals is not accountable simply on the grounds that they forget more rapidly; this has already been shown in (5) above—once controls and experimentals have been trained to a given standard of accuracy of response the decay of the memory is about equally rapid in the two series. The operations, in short, produce an immediate decline in performance that cannot be attributed either to more rapid forgetting or to any qualitative change in responses to the test objects.

If a small inedible object is presented by dropping it in front of an unblinded octopus, the animal will reach out and grasp it. The object is passed in under the interbrachial web, and the octopus tries to eat it; soft objects have pieces bitten out when rejected. After a minute or two the object is passed out from under the web and thrust away with one of the arms. If the same object is repeatedly presented at short (2 min.) intervals, the octopus quickly (2-3 trials) ceases to pass the object in under the web and spends progressively less time examining it with the arms before rejection. Eventually it ceases to respond altogether, but will again reach out and take the object after an interval of an hour or more. A series of such experiments has been reported in Wells & Wells (19576). The main difference between the 5 controls and the 6 experimentals (5 with the vertical lobe and one with the superior frontal lobe removed) was that the experimental octopuses took about six times as long as controls to examine and reject the object at its first presentation. The control animals rejected the object after an average of 2 min. 23 sec. (range 30 sec.—4 min. 40 sec.) ; the experimentals averaged 14 min. 19 sec. (range 3 min. 35 sec.-35 min. 50 sec.—the difference between the two groups is significant: Mann-Whitney, P = < 0·01). Two experimental animals and 4 controls were tested for retention and retrained in a series of tests with breaks of from 2 to 74 hr. The experimental animals required rather more retraining than the controls.

In addition to the above repeated-presentation experiments a number of others were made in which the animals were given 6 V. a.c. shocks if they took the test object. Most of these experiments were performed on blinded control animals in the course of a study of arm-to-arm transfer of experience (Wells, 1959 a). Others were concerned in experiments on the inferior frontal/subfrontal tactile learning system (Wells, 1959b). Controls, in general, learned to reject an object that they were punished for taking within 4 or 5 trials. Seven out of 10 octopuses with vertical lobe lesions (4 of them having much more extensive lesions, including large parts of the basal lobes) learned within 5 trials, and 5 of them made only 1 or 2 mistakes. On the basis of these experiments, involving a very simple problem and trials at short intervals, there are no grounds for separating the control and the experimental animals.

The experiments summarized show that:

1. Removal of the vertical and/or median superior frontal lobes slows learning in tactile discrimination experiments. The experimental animals take longer to learn and are somewhat more variable than controls.

2. Animals trained before operation perform less accurately when tested immediately afterwards and may appear to have lost the capacity to discriminate altogether. With further training, however, such animals relearn more rapidly than octopuses without pre-operational training.

3. After training, controls and experimental animals both revert in time to entirely positive responses. This tends to mask a capacity to discriminate that can be revealed by any means that lowers the threshold for negative responses towards objects touched.

4. The relatively poor performance of the animals with brain lesions is not due to more rapid forgetting between trials or groups of trials. Once trained to a similar standard, controls and experimentals forget with equal rapidity. Experimental animals, moreover, behave differently from controls even at the first trial; when an unfamiliar object is presented they take about six times as long to examine and reject it.

5. In training to discriminate there is no difference in the pattern of errors made by operated and control animals. After the operations octopuses make the same sorts of errors as before, in the same proportions. Both err mainly by taking the negative objects in the first half of their 20-trial training groups. There is no indication that removal of the vertical (or median superior frontal, or both) lobes has a specific effect on positive or negative learning.

These experiments show that removal of the vertical and/or superior frontal lobes has quantitative, but apparently no qualitative effects on touch learning.

The ability to learn to make tactile discriminations is known to depend on the integrity of the inferior frontal and subfrontal lobes (Wells, 1959b,1961). The anatomy of these lobes has much in common with that of the superior frontal and vertical lobe pair lying above and behind them (Fig. 1). In both cases there is a lobe (the inferior or superior frontal) consisting largely of tracts leading into a region characterized by an enormous number of very small cells. The vertical lobe contains upwards of 25 × 10⁶ cells, the subfrontal 5·2 × 10⁶—between them these two include more than half of the neurones in the central supraoesophageal part of the octopus brain (Young, 1963 a). A number of experiments have been made to discover the function of the inferior frontal and subfrontal lobes. From these it seems that the median inferior frontal is a region of connexions ensuring that information arriving along the nerves of one arm is distributed to centres controlling the activities of the others. Octopuses can be trained, provided that experience is limited to one arm at a time, even after removal of the whole of this region. The subfrontal is in contrast essential to touch learning. Octopuses cannot be taught to discriminate between P1 and P4 if less than about 250,000 subfrontal cells remain. A few thousand will suffice for learning to reject one of these objects in repeated-presentation experiments ; below this number touch learning seems altogether impossible (Wells, 19596).

Somewhat similar results have been obtained in experiments on the vertical lobe, where also partial lesions have an effect on performance in proportion to their extent, and seem particularly to limit the capacity to learn to make difficult discriminations (Young, 1958 a).

These experiments, and the marked anatomical similarities of the superior frontal-vertical and inferior frontal-subfrontal lobe pairs, together suggest that the vertical lobe may affect touch learning by adding further computing and storage units to those already present in the subfrontal. This is by no means a new idea. A similar explanation has been used in the past to describe the relationship of the vertical lobe to the visual system. ‘Most but perhaps not all of the results of removing vertical lobe tissue can be understood if we assume that the cells of this lobe serve to add further units to the memory stores established in the optic or inferior frontal and subfrontal lobes. The memories are thus made more effective in situations where they otherwise fail’ (Young, 1958 a). The evidence at present available from tactile experiments does not, of itself, at present appear to warrant a revision of this description of vertical lobe function.

This is not, however, a description of vertical lobe function currently proposed in explanation of some more recent visual experiments. Thus Muntz, Sutherland & Young (1962), summarizing the conditions under which animals with the vertical lobe removed behave most like controls in visual experiments, conclude that the effects of vertical lobe removal are not so much due to a reduction in the storage capacity of the memory system as to ‘motivational ‘difficulties—a failure to adjust levels of response. The conditions under which animals with vertical lobe lesions perform best i.e. most like controls—are: (1) when the trials are closely spaced; (2) when rewards and punishments are discontinued after training; (3) in extinction experiments, with the same object repeatedly presented without reward; (4) in transfer tests where no rewards are given; (5) in discrimination training where both situations are presented simultaneously. As Muntz et al. point out, these are all situations in which the effect of fluctuations in the level of attacks will be minimal, either because there is little or no time for it to alter, or because food and shocks, which might cause fluctuations, are absent.

Their explanation of these results in terms of a failure to adjust levels of response depends, however, on the supposition that vertical lobe removal is followed by pro-longed fluctuations in the level of an attack after rewards or shocks and, in particular, by a tendency to attack more than normal animals in experiments involving rewards. Where these effects have been specifically investigated they have been shown not to exist. The effect of individual shocks or rewards in fact dies away more rapidly in operated than in control animals (Young, i960). Octopuses with the vertical lobes removed do not all attack more readily than before; instead their pre-operational tendencies seem to be changed by the operation, those that attacked often before in general attacking less afterwards, and vice versa (Young, 19586). These findings argue against a motivational hypothesis, but are compatible with the ‘added units ‘type of explanation proposed on a basis of the tactile experiments. Thus the conditions under which animals lacking the vertical lobes perform best are precisely those under which a memory system, rendered less effective than usual by a reduction in the number of its outputs, would be best able to control overt behaviour. The more rapid failure of the immediate effects of individual experiences, and the tendency to reverse the sign of responses—presumably learned responses, if the animal attacks very much more or less than average—are both compatible with the view that vertical lobe removal has quantitative effects because it reduces the number of units in the memory store. They do not appear to be compatible with a theory that it operates qualitatively by damping down fluctuations in the overall level of response.

In the evolution of theories about the function of the superior frontal and vertical lobes it was at first assumed on structural grounds that the median superior frontal is a relay station, since, compared with the vertical lobe, it contains few cells (1·7 × 10⁶ cf 25 × 10⁶—Young, 1963 a) and many tracts, all of which appear to go into the vertical lobe. Observations that its removal (or severance of its output to the vertical lobe) produced the same effects in discrimination experiments as vertical lobe removal were attributed to removal of the input to the vertical lobe (Boycott & Young, 1955). Recently, however, it has been shown that removal of the median superior frontal does not always have the same effect as removal of the vertical lobe. Thus, with animals selected as particularly ‘good attackers removal of the vertical lobes had little effect on the tendency to make unrewarded attacks on crabs. Removal of the median superior frontal, on the other hand, led to an immediate decline in the proportion of attacks, which was restored as soon as the vertical lobe was removed as well. These results are interpreted as showing that it is the normal function of the vertical lobe to raise thresholds for positive responses in the absence of rewards (or when punished), and the function of the median superior frontal to lower them when a reward is obtained ; a balance between these paired centres determines biologically appropriate responses (Young, 19636).

Anatomically the ‘paired centres’ hypothesis is well supported. The connexions between the lobes that are now known (Fig. 8) certainly show pairs of complementary circuits, each involving two anatomically distinct centres. Whether these paired centres have the functions that Young (1963b) suggests, is, however, another matter. If the vertical lobe is a suppressor, raising the threshold for positive responses to modify or cancel the output from a median superior frontal concerned in lowering the same thresholds, then it should be possible to show differences in the pattern of errors made after the two operations. This has not so far been shown. In a recent summary of a considerable series of experiments made to test the ‘paired centres’ hypothesis, Young (1964) has shown: (1) That there is no clear difference in the tendency to attack crabs after superior frontal or vertical lobe removal. After both sorts of operation, some individuals attack more, others less ; in general those that attacked often before the operation attack less afterwards and vice versa. An exception is the case, already cited, where the animals were selected as particularly ‘good attackers’ and tested immediately after operation. (2) Both operations slow learning to attack a horizontal rectangle and learning to discriminate between horizontal and vertical rectangles. In 50 trials with each figure, octopuses with the median superior frontal lobes removed did better than those lacking the vertical, mainly due to improvement within training sessions of 10 massed trials. (3) In experiments on the extinction of unrewarded responses to a black disc, and in learning not to attack crabs, animals lacking the vertical lobes learned more slowly than controls or animals with the median superior frontal lobe removed. These results show that removal of the superior frontal and vertical lobes have different effects and in general removal of the superior frontal alters performance less than removal of the vertical lobe. They do not show the marked differences in pattern of errors follówing the two operations that one might expect on a basis of the ‘paired centres ‘hypothesis.

Maldonado (1963 a-c), using automatic techniques to record attacks on crabs, has shown that vertical lobe removal slows positive learning as well as having much more obvious effects on learning not to attack. With trained animals the immediate effect of the operation is to return performance to a condition resembling the early phases of preoperational training. The animals relearn, but never come to respond as rapidly and accurately as unoperated octopuses. Maldonado interprets these results as showing that the vertical lobe has a ‘general amplifying’ effect, which would be compatible, for example, with the opposite effects of its removal on animals found to attack often or little before operation. He has developed an information-flow model of the organization of the brain of Octopus, in which at least some of the components can be identified with structures in the supraeosophageal lobes. In this model the vertical lobe is seen as an amplifier, but not as the site of storage elements in the learning system (Maldonado, 1963 c).

In conclusion, it would seem that although some of the visual evidence is suggestive, it does not at the present time justify concluding that the presence of the vertical lobe does more for touch learning than to enhance capacities already present in the inferior frontal and subfrontal system. Whether it does so by adding units to, or amplifying the effect of, learning systems located elsewhere in the octopus brain is not known. Indeed, it is difficult to imagine a test that would allow one to decide between the two hypotheses, which produce similar predictions about the effect of vertical lobe removal on the behaviour of the animals.

With this state of affairs, the results of subfrontal removal become an important consideration. The subfrontal, as pointed out above, stands in a similar relationship to the tactile as the vertical lobe to the visual system. Yet a minimal number of subfrontal cells must remain if an octopus is to learn at all; below this number learning ceases. The lesion, cannot, as with vertical lobe removal, be compensated for by further training. If the subfrontal has only a general amplifying function, as would be supposed by analogy with the vertical lobe following Maldonado’s hypothesis, it is difficult to see why not. On the grounds of economy of hypothesis and the anatomical similarity of the two, it would seem illogical to favour the ‘general amplification’ hypothesis of vertical lobe function while the evidence is against this explanation in the case of the subfrontal.

The firm statements that can be made about vertical lobe function remain:

1. The presence of the vertical lobe increases the effects of experience in situations where the octopus has to learn.

2. It does not significantly slow the rate at which learned responses are forgotten.

3. It does not appear to alter the pattern of errors made other than by decreasing their number ; its effects on performance are quantitative, not qualitative.

4. It is not functionally subdivided; partial lesions anywhere in it have effects proportional to their extent.

5. Its removal leads to an immediate increase in the proportion of errors made by trained animals.

6. In view of (2) and (3) above, this last must mean either that it forms part of the memory store or, less probably, that it is a mechanism for amplifying the effects of memory stores located elsewhere.

1. In this paper the effects of vertical and superior frontal lobe removal are considered in relation to touch learning and discussed in the light of findings from visual experiments.

2. Removal of the vertical and/or the median superior frontal lobe slows touch learning. The operated animals take nearly twice as many trials as controls to learn a simple tactile discrimination. Their performance is more variable than that of controls.

3. Removal of these parts increases the number, but does not alter the sign of errors made. Removal of the median superior frontal does not appear to add to the effect of removing the vertical lobe.

4. The pattern of errors made by controls and experimental animals is similar within groups of massed trials given in the course of training.

5. Once trained to a given standard of accuracy (75 or 85 % correct responses) the operated animals remember for as long as controls. Retention tests show little difference in performance 5 and 10 days after the end of training; the difference between controls and experimentals cannot be attributed to more rapid forgetting by the latter.

6. In experiments involving reversal of learned discriminations controls learn in fewer trials than experimentals.

7. Octopuses operated after training show an immediate decline in performance that can, to a considerable extent, be compensated by further training. The operations do not entirely remove the effects of past experience. Again, the effects are quantitative, not qualitative.

8. Octopuses with the vertical lobes removed take about six times as long as controls to examine and reject an unfamiliar object.

9. These results are considered in relation to (a) the effects of lesions to the inferior frontal and subfrontal lobes, which are anatomically similar to the superior frontal and vertical, and (b) current theories of the function of the vertical lobe in visual experiments.

10. It is concluded that the vertical lobe produces quantitative increases in the effects of experience on subsequent actions. It remains unestablished whether it does this by amplifying the effects of events taking place elsewhere in the central nervous system or by adding further units to a pool of neurones alterable by experience. The latter seems more probable in view of experiments on the anatomically similar subfrontal lobe, where a minimal number of cells must be present if the animals are to learn at all.

The author wishes to thank Dr K. Machin, of the Zoology Department, Cambridge, for advice on the treatment and for the statistical analysis of the data summarized in Fig. 2. He would also like to thank Professor J. Z. Young, F.R.S., Dr Vernon Barber, both of the Anatomy Department, University College, London, and Dr C. H. F. Rowell of the Zoology Department, Makerere, Kampala, Uganda, all of whom have read and commented on this work in preparation. The facilities for the experimental work were provided by the Stazione Zoologica di Napoli, and the money to support the research by the Rockefeller foundation. Professor Young also kindly provided the pictures for Figs. 1 and 8.