Neuropsychologia,1975,Vol. 13,pp. 111to 113.PergamonPress.Printedin England.
NOTE VISUAL, DISCRIMINATION
IN THE MONKEY:
THE INCORRECT
RESPONSE
ET~INGER
RIDLEY
G.
and R. M.
DISTINGUISHING
Institute of Psychiatry, De Crespigny Park, London, SE5 8AF. England (Received 14 March 1974) Abstract-Monkeys were trained to learn 2 visual discrimination tasks and the reversals of these tasks under 2 conditions: (1) with the response panel (or lid of food-box) locked after incorrect response; and (2) with the panel (or lid) unlocked after incorrect response. Criterion was achieved more rapidly during original learning with the panel locked, but no significant difference was observed during reversal learning. It is suggested that improved performance resulted when the panel remained locked because either the outcome was more salient or because the incorrect response was more frustrating. TNTRODUCTION SMALL DIFFERENCES in procedure
are recognised to influence the outcome of training, sometimes to an unexpected extent. LASHLEY [l] introduced a novel way of training rats: the animal jumped from a stand at one of 2 doors which consisted of painted cards. If the choice of doors was correct, the card fell backwards to allow the rat access to the reward; if incorrect, the card was locked and the rat fell into a net. The “Lashley jumping stand” yielded significantly better performance scores than earlier methods of training. The standard method of training monkeys was introduced by HARLOW [2]. The monkey is rewarded for displacing the correct stimulus object in a Wisconsin General Test Apparatus (WGTA), but obtains no reward for displacing the incorrect stimulus. In the present experiment, the monkeys were prevented from displacing the incorrect stimulus, and overall this resulted in improved rates of learning. METHOD Subjects Twelve immature Macaca mulatta monkeys were used, 8 with automatic and 4 with manual training, All were previously untrained, and weighed 2-3 kg. Automatic training Animals were trained in apparatus previously described @ITLINGER 131). In brief, a Lint-8 computer controlled the presentation of stimuli which were back-projected onto 2 vertical panels, and of rewards which were peanuts delivered to cups behind the correct panel. The precise sequence on correct response was: monkey presses correct panel; the panel is unlocked so that its lower edge can be pushed through 45 deg. about hinge along top edge; stimuli are switched off but light behind correct panel is switched on; peanut delivered to cup behind correct panel; when the monkey releases the panel it is locked and the light behind the panel is switched off. The sequence on incorrect response was either (“door locked”): monkey presses incorrect panel; stimuli are switched off (i.e. door does not unlock); or (“door unlocked”): monkey presses incorrect panel; the panel is unlocked; stimuli are switched off (i.e. door unlocks but there is no reward and no light). Forty trials were given daily, 5 days/week; in all tasks training continued until there were only 10 errors in 100 consecutive trials. After standard adapting procedure, all animals were trained on 4 tasks in the same sequence: A, B, reversal of A, reversal of B. As indicated in Table I,4 animals were trained on the 4 tasks with doors locked, then unlocked, again unlocked and finally locked; and 4 in the opposite order. Task A was the pair of pattern stimuli shown in the bottom row of Fig. 3 of ETTLINGER and GAIJTRIN141.Both patterns are crosses but with gaps in the horizontal or vertical arms. Task B was a similar pair made of diagonal lines. Manual training Animals were trained to push the lids of food boxes located on the shelf of a WGTA. Stimuli were cardboard plaques attached to the lids. On correct response: monkey pushes the correct lid; the lid slides off box; the monkey finds a peanut in a well beneath the lid. On incorrect response the sequence was either 111
112
NOTE
(“lid locked”) : monkey pushes the incorrect lid which is mechanically locked; or (“lid unlocked”) : monkey pushes the incorrect lid; the lid is displaced off box; the monkey finds no peanut in food well. Training procedure and the design were comparable to automatic training. Task C was a discrimination between two levels of reflectivity, gray and white; Task D a discrimination between two shapes M and W. These 2 tasks have been described for stages V and VI of training by GAUTRINet al. [5].
RESULTS Automatic training The performance scores of 8 monkeys are shown in Table 1, where sequence of training is indicated by successive columns from left to right. Using Wilcoxon’s Test for matched pairs, two separate analyses were made: (a) task A and B were learnt more rapidly with doors locked on incorrect response than with doors unlocked (PC 0.05 one-tailed); (b) the reversals of tasks A and B were not learnt more rapidly with doors locked on incorrect response (P >0.05 one-tailed). In the order given, tasks A and B, as also the reversals of tasks A and B, were individually of approximately equal difficulty. However, original learning was a little slower (mean=468 trials) than reversal learning (mean=360 trials) when tasks A and B were combined. In Table 1 underlining of scores indicates a reversal of the trend for more rapid learning with doors locked Table 1. Trials required to learn two tasks and their reversals when the door was locked or unlocked on incorrect response
Animals
Task A Door Locked Unlocked
1 2 3 4 5 6 7 8
371 648 472 241
Mean
433
Task B Door Locked Unlocked
Task A reversed Door Locked Unlocked
748 469 lOOOF 289 441 523 374 560
187 621 ZTS 316
475
335
145 581 151 -699 627
394
Task B reversed
370 474 107 -
Locked 284 347 290 -
317
307
Door Unlocked
293 833 220 -332 420
Figures are trials required to reach a standard level of performance (10 errors in 100 trials), excluding the final 100 trials. F indicates that the animal failed to reach the standard level of performance within the stated number of trials. - indicates that training was discontinued. Underlining indicates a reversal of the general trend for faster learning with door locked. on incorrect response. In 15 pairs of observations such reversals occurred only four times, equally distributed between original and reversal learning. No animal consistently learnt more rapidly with the door unlocked on incorrect response. However, 3 animals consistently learnt more rapidly with the door locked. Manual training Unexpectedly, naive animals required about 4 times as many trials to learn task D as to learn task C, and about 3 times as many to learn the reversal of task D as the reversal of task C. (These observations differ from those of GAUTRINet al. [5], and of NIE et al. [q, who used the same tasks but after training on another difficult pattern task.) The observed large difference in level of task difficulty totally obscures any expected difference related to conditions of door locked/unlocked on incorrect response.
DISCUSSION No comparable study with monkeys is known to us. When all the scores during automatic training in the present study are combined, our monkeys required on the average 460 trials to learn with doors unlocked on incorrect response, but only 367 with doors locked. This represents a savings of better than one tifrh of trials under conditions of locked doors. Nonetheless, the differences on reversal learning were not significant. However, this failure is largely attributable to animal 8 which showed a very large difference in a direction opposite to that shown by 5 of the remaining 6 animals.
NOTE
113
The overall good performance on reversal learning during automatic training was unexpected. Serial reversals with the same apparatus conform to the expected profile of high scores on early reversals (N. Hester and G. Ettlinger, unpublished observations). Therefore, the low scores for reversal of task A and B probably reflect the interference effects of performance on another task between learning and reversal (i.e. of learning task B before reversal of task A; of reversal on task A before reversal on task B). Can the exceptionally fast learning of standard shape discriminations with the present apparatus, as described by ETTLINGER and GAUTRIN[4], be explained by reference to the doors being locked on incorrect responses? Other factors appear also to contribute. Most important perhaps is the total filling of the response panel by the stimuli, so that in making a response the monkey’s hand is certain to touch illuminated regions of the panel. Next, the present apparatus delivers reward to a cup immediately behind the correct response panel. (A feature of the original apparatus, namely continued presentation of both stimuli during retrieval of the peanut after correct response, has now been discontinued because it did not accelerate learning.) Why should locking the doors on incorrect response accelerate learning? No definitive answer is possible. If the door unlocks on correct response but remains locked on incorrect response the monkey is given additional and more rapid differential information than is provided merely by the presence or absence of a peanut behind a door which always unlocks. Since the response panels are spring-loaded to close rapidly after release, a response requires effort, and if incorrect, this effort is immediately counteracted by a door that remains locked. Lastly, such vain effort may, in itself, prove frustrating to the animal, or more frustrating than mere absence of reward. Acknowledgements-Miss R. M. RIDLEY was supported by a grant from The Wellcome Trust. The expenses of this work were provided by a grant to G.E. from the Maudsley Hospital Research Fund. We are grateful to both, and to DR. R. E. PASSINGHAM for helpful criticism of the typescript.
REFERENCES 1. LASHLEY,K. S. The mechanism of vision: I. A method for rapid analysis of pattern-vision in the rat. J. gen. Psychol. 37,453460, 1930. 2. HARLOW,H. F. The formation of learning sets. Psychol. Rev. 56,51-65, 1949. 3. ETTLMGER,G. Training monkeys by Lint-8 computer. Cortex 6,41&416,1970. 4. ETIZMGER, G. and GAUTRIN,D. Visual discrimination performance in the monkey: the effect of unilateral removals of temporal cortex. Cortex 7,317-331,197l. 5. GAUTRIN, D., FENTON, G. and ET~LINGER,G. Aluminium hydroxide implants on the infero-temporal cortex of the monkey: their mode of influencing visual discrimination performance. Exp. Neural. 33, 459-474, 1971. 6. Nra, V., UPTON, A. and E~INGER, G. Behavioural impairment in the monkey following implantation of aluminium hydroxide on the temporal cortex: the role of cortical destruction. Exp. Neural. 40,632-651, 1973. Resume-On a entrain6 des singes a deux taches de discrimination visuelle ainsi que le renversement de ces epreuves sous deux conditions: (1) avec le parmeau de reponse (ou le couvercle de la bolte de nourriture) ferme apres une reponse incorrecte et, (2) avec le panneau (avec le couvercle) ouvert apres une reponse incorrecte. Le critbre Ctait atteint plus rapidement durant l’apprentissage d’origine avec le panneau ferrne mais on n’observait aucune difference significative pendant l’apprentissage de renversement. On suggere que l’amelioration de la performance provient, lorsque le panneau reste ferme, soit que I’effet est plus saillant, soit parce que la reponse incorrecte entraine plus de frustation. Zusammenfasstmg-AITen wurden auf die Unterscheidungsfiihigkeit zweier optischer Diskriminationen trainiert und deren Umkehrungen. Das Training erfolgte unter zwei Bedingungen : 1. mit verschlossenem Fach oder Deckel einer Dose mit ESwaren nach unrichtiger Antwort, und 2. mit unverschlossenem Fach oder Deckel nach unrichtiger Antwort. Eine Unterscheidung 1ieS sich dabei rascher erreichen, wenn das ursprtingliche Lemen mit dem verschlossenen Fach vonstatten ging. Dagegen wurde keine signifikante Differenz wahrend der Phase des umgekehrten Lernens beobachtet. Es ist zu vermuten, dab die bessere Leistung dann zustande kam, wenn das Fach verschlossen blieb, weil das Ergebnis mehr hervorstach oder weil die inkorrekt Antwort mehr frustrierte.
NOTE VISUAL, DISCRIMINATION
IN THE MONKEY:
THE INCORRECT
RESPONSE
ET~INGER
RIDLEY
G.
and R. M.
DISTINGUISHING
Institute of Psychiatry, De Crespigny Park, London, SE5 8AF. England (Received 14 March 1974) Abstract-Monkeys were trained to learn 2 visual discrimination tasks and the reversals of these tasks under 2 conditions: (1) with the response panel (or lid of food-box) locked after incorrect response; and (2) with the panel (or lid) unlocked after incorrect response. Criterion was achieved more rapidly during original learning with the panel locked, but no significant difference was observed during reversal learning. It is suggested that improved performance resulted when the panel remained locked because either the outcome was more salient or because the incorrect response was more frustrating. TNTRODUCTION SMALL DIFFERENCES in procedure
are recognised to influence the outcome of training, sometimes to an unexpected extent. LASHLEY [l] introduced a novel way of training rats: the animal jumped from a stand at one of 2 doors which consisted of painted cards. If the choice of doors was correct, the card fell backwards to allow the rat access to the reward; if incorrect, the card was locked and the rat fell into a net. The “Lashley jumping stand” yielded significantly better performance scores than earlier methods of training. The standard method of training monkeys was introduced by HARLOW [2]. The monkey is rewarded for displacing the correct stimulus object in a Wisconsin General Test Apparatus (WGTA), but obtains no reward for displacing the incorrect stimulus. In the present experiment, the monkeys were prevented from displacing the incorrect stimulus, and overall this resulted in improved rates of learning. METHOD Subjects Twelve immature Macaca mulatta monkeys were used, 8 with automatic and 4 with manual training, All were previously untrained, and weighed 2-3 kg. Automatic training Animals were trained in apparatus previously described @ITLINGER 131). In brief, a Lint-8 computer controlled the presentation of stimuli which were back-projected onto 2 vertical panels, and of rewards which were peanuts delivered to cups behind the correct panel. The precise sequence on correct response was: monkey presses correct panel; the panel is unlocked so that its lower edge can be pushed through 45 deg. about hinge along top edge; stimuli are switched off but light behind correct panel is switched on; peanut delivered to cup behind correct panel; when the monkey releases the panel it is locked and the light behind the panel is switched off. The sequence on incorrect response was either (“door locked”): monkey presses incorrect panel; stimuli are switched off (i.e. door does not unlock); or (“door unlocked”): monkey presses incorrect panel; the panel is unlocked; stimuli are switched off (i.e. door unlocks but there is no reward and no light). Forty trials were given daily, 5 days/week; in all tasks training continued until there were only 10 errors in 100 consecutive trials. After standard adapting procedure, all animals were trained on 4 tasks in the same sequence: A, B, reversal of A, reversal of B. As indicated in Table I,4 animals were trained on the 4 tasks with doors locked, then unlocked, again unlocked and finally locked; and 4 in the opposite order. Task A was the pair of pattern stimuli shown in the bottom row of Fig. 3 of ETTLINGER and GAIJTRIN141.Both patterns are crosses but with gaps in the horizontal or vertical arms. Task B was a similar pair made of diagonal lines. Manual training Animals were trained to push the lids of food boxes located on the shelf of a WGTA. Stimuli were cardboard plaques attached to the lids. On correct response: monkey pushes the correct lid; the lid slides off box; the monkey finds a peanut in a well beneath the lid. On incorrect response the sequence was either 111
112
NOTE
(“lid locked”) : monkey pushes the incorrect lid which is mechanically locked; or (“lid unlocked”) : monkey pushes the incorrect lid; the lid is displaced off box; the monkey finds no peanut in food well. Training procedure and the design were comparable to automatic training. Task C was a discrimination between two levels of reflectivity, gray and white; Task D a discrimination between two shapes M and W. These 2 tasks have been described for stages V and VI of training by GAUTRINet al. [5].
RESULTS Automatic training The performance scores of 8 monkeys are shown in Table 1, where sequence of training is indicated by successive columns from left to right. Using Wilcoxon’s Test for matched pairs, two separate analyses were made: (a) task A and B were learnt more rapidly with doors locked on incorrect response than with doors unlocked (PC 0.05 one-tailed); (b) the reversals of tasks A and B were not learnt more rapidly with doors locked on incorrect response (P >0.05 one-tailed). In the order given, tasks A and B, as also the reversals of tasks A and B, were individually of approximately equal difficulty. However, original learning was a little slower (mean=468 trials) than reversal learning (mean=360 trials) when tasks A and B were combined. In Table 1 underlining of scores indicates a reversal of the trend for more rapid learning with doors locked Table 1. Trials required to learn two tasks and their reversals when the door was locked or unlocked on incorrect response
Animals
Task A Door Locked Unlocked
1 2 3 4 5 6 7 8
371 648 472 241
Mean
433
Task B Door Locked Unlocked
Task A reversed Door Locked Unlocked
748 469 lOOOF 289 441 523 374 560
187 621 ZTS 316
475
335
145 581 151 -699 627
394
Task B reversed
370 474 107 -
Locked 284 347 290 -
317
307
Door Unlocked
293 833 220 -332 420
Figures are trials required to reach a standard level of performance (10 errors in 100 trials), excluding the final 100 trials. F indicates that the animal failed to reach the standard level of performance within the stated number of trials. - indicates that training was discontinued. Underlining indicates a reversal of the general trend for faster learning with door locked. on incorrect response. In 15 pairs of observations such reversals occurred only four times, equally distributed between original and reversal learning. No animal consistently learnt more rapidly with the door unlocked on incorrect response. However, 3 animals consistently learnt more rapidly with the door locked. Manual training Unexpectedly, naive animals required about 4 times as many trials to learn task D as to learn task C, and about 3 times as many to learn the reversal of task D as the reversal of task C. (These observations differ from those of GAUTRINet al. [5], and of NIE et al. [q, who used the same tasks but after training on another difficult pattern task.) The observed large difference in level of task difficulty totally obscures any expected difference related to conditions of door locked/unlocked on incorrect response.
DISCUSSION No comparable study with monkeys is known to us. When all the scores during automatic training in the present study are combined, our monkeys required on the average 460 trials to learn with doors unlocked on incorrect response, but only 367 with doors locked. This represents a savings of better than one tifrh of trials under conditions of locked doors. Nonetheless, the differences on reversal learning were not significant. However, this failure is largely attributable to animal 8 which showed a very large difference in a direction opposite to that shown by 5 of the remaining 6 animals.
NOTE
113
The overall good performance on reversal learning during automatic training was unexpected. Serial reversals with the same apparatus conform to the expected profile of high scores on early reversals (N. Hester and G. Ettlinger, unpublished observations). Therefore, the low scores for reversal of task A and B probably reflect the interference effects of performance on another task between learning and reversal (i.e. of learning task B before reversal of task A; of reversal on task A before reversal on task B). Can the exceptionally fast learning of standard shape discriminations with the present apparatus, as described by ETTLINGER and GAUTRIN[4], be explained by reference to the doors being locked on incorrect responses? Other factors appear also to contribute. Most important perhaps is the total filling of the response panel by the stimuli, so that in making a response the monkey’s hand is certain to touch illuminated regions of the panel. Next, the present apparatus delivers reward to a cup immediately behind the correct response panel. (A feature of the original apparatus, namely continued presentation of both stimuli during retrieval of the peanut after correct response, has now been discontinued because it did not accelerate learning.) Why should locking the doors on incorrect response accelerate learning? No definitive answer is possible. If the door unlocks on correct response but remains locked on incorrect response the monkey is given additional and more rapid differential information than is provided merely by the presence or absence of a peanut behind a door which always unlocks. Since the response panels are spring-loaded to close rapidly after release, a response requires effort, and if incorrect, this effort is immediately counteracted by a door that remains locked. Lastly, such vain effort may, in itself, prove frustrating to the animal, or more frustrating than mere absence of reward. Acknowledgements-Miss R. M. RIDLEY was supported by a grant from The Wellcome Trust. The expenses of this work were provided by a grant to G.E. from the Maudsley Hospital Research Fund. We are grateful to both, and to DR. R. E. PASSINGHAM for helpful criticism of the typescript.
REFERENCES 1. LASHLEY,K. S. The mechanism of vision: I. A method for rapid analysis of pattern-vision in the rat. J. gen. Psychol. 37,453460, 1930. 2. HARLOW,H. F. The formation of learning sets. Psychol. Rev. 56,51-65, 1949. 3. ETTLMGER,G. Training monkeys by Lint-8 computer. Cortex 6,41&416,1970. 4. ETIZMGER, G. and GAUTRIN,D. Visual discrimination performance in the monkey: the effect of unilateral removals of temporal cortex. Cortex 7,317-331,197l. 5. GAUTRIN, D., FENTON, G. and ET~LINGER,G. Aluminium hydroxide implants on the infero-temporal cortex of the monkey: their mode of influencing visual discrimination performance. Exp. Neural. 33, 459-474, 1971. 6. Nra, V., UPTON, A. and E~INGER, G. Behavioural impairment in the monkey following implantation of aluminium hydroxide on the temporal cortex: the role of cortical destruction. Exp. Neural. 40,632-651, 1973. Resume-On a entrain6 des singes a deux taches de discrimination visuelle ainsi que le renversement de ces epreuves sous deux conditions: (1) avec le parmeau de reponse (ou le couvercle de la bolte de nourriture) ferme apres une reponse incorrecte et, (2) avec le panneau (avec le couvercle) ouvert apres une reponse incorrecte. Le critbre Ctait atteint plus rapidement durant l’apprentissage d’origine avec le panneau ferrne mais on n’observait aucune difference significative pendant l’apprentissage de renversement. On suggere que l’amelioration de la performance provient, lorsque le panneau reste ferme, soit que I’effet est plus saillant, soit parce que la reponse incorrecte entraine plus de frustation. Zusammenfasstmg-AITen wurden auf die Unterscheidungsfiihigkeit zweier optischer Diskriminationen trainiert und deren Umkehrungen. Das Training erfolgte unter zwei Bedingungen : 1. mit verschlossenem Fach oder Deckel einer Dose mit ESwaren nach unrichtiger Antwort, und 2. mit unverschlossenem Fach oder Deckel nach unrichtiger Antwort. Eine Unterscheidung 1ieS sich dabei rascher erreichen, wenn das ursprtingliche Lemen mit dem verschlossenen Fach vonstatten ging. Dagegen wurde keine signifikante Differenz wahrend der Phase des umgekehrten Lernens beobachtet. Es ist zu vermuten, dab die bessere Leistung dann zustande kam, wenn das Fach verschlossen blieb, weil das Ergebnis mehr hervorstach oder weil die inkorrekt Antwort mehr frustrierte.
