Behavioural Processes, 7 (1982) 63-72 Elsevier Scientific Publishing Company,
63 Amsterdam
-
Printed
in The Netherlands
CAPACITY OF WORKING MEMORY IN RATS AS DETERMINED PERFORMANCE ON A RADIAL MAZE
0. BURESOVA
and J. BURES
Institute of Physiology, 4-K+ (Czechoslovakia) (Accepted
1 July
BY
Czechoslovak
Academy
of Sciences,
Videhska’
1083
142-20
Prague
1981)
ABSTRACT BureXova, 0. and BureH, J., 1982. Capacity of working memory performance on a radial maze. Behav. Processes, 7: 63-72.
in rats as determined
by
Capacity of the working memory was tested in 12 rats highly overtrained in the 12and 24-arm radial mazes. Asymptotic performance levels were characterized by 1.01 and 2.78 errors/trial in the 12- and 24-arm mazes, respectively. The incidence of errors increased from 31% on the last choice in the 12-arm maze to 51% on choices 23 and 24 in the 24-arm maze, but remained significantly below the expected error probability of about 85%. Linear extrapolation of the above trend to mazes with more arms suggests working memory capacity of 40 to 50 items. When two trials in a 12-arm maze were repeated in immediate succession, error incidence increased from 1.17 in the first trial to 2.13 in the second trial. The tendency to avoid choice repetition could be observed in any string of 12 continuous choices, but was weakest in segments divided by trial boundary (2.48 errors in choices 7 to 18). With a different trial separation (choices l-6 and 19-24 in maze A, choices 7-18 in an adjacent maze B) errors dropped to 1.09 in B but increased to 2.30 in A. It is concluded that radial maze performance reflects avoidance of choice repetition which is improved by recognition of trial boundaries and is adversely influenced by forgetting and interference.
INTRODUCTION
The spectacular performance of rats in the radial maze (Olton and Samuelson, 1976; Olton et al., 1979) has raised the question what are the limits of the rat’s working memory in this task. Olton et al. (1977) attempted to estimate its capacity by extrapolating the decreasing probability of correct responding in the 17-arm maze. They suggested that choice accuracy would decrease to chance levels at 25 to 30 choices, provided that the performance decay is a linear function of the choice number. Roberts (1979) used a radial maze with eight primary arms, each giving access to three secondary arms, baited with food. The rats had to locate 24 food pellets, but the branching geometry of the maze precludes direct comparison of their performance with the radial maze results. The present study compares the choice behavior of rats in the 12- and 24-arm radial mazes. 0376-6357/82/000~0000/$
02.75
o 1982
Elsevier
Scientific
Publishing
Company
64 METHOD
The apparatus was a tubular radial maze of the type described by Magni et al. (1979) and Buresova (1980). The choice platform (40 or 80 cm in diameter, respectively) had a central opening (8 cm in diameter) and was surrounded by 40 cm high plexiglass walls. One-way swing-doors (6 X 6 cm) provided access into the 35 cm long alleys with perforated ceiling and a recessed food cup at the far end. One-way exit door connected the channel with the outside area separated from the rest of the room by a 50 cm high plexiglass wall. The maze rested on 5 cm high supporting legs and could be entered through the space between the floor and the choice platform. The animals were 12 male hooded rats 6 months old at the time of the experiment. They were maintained on a 24 h food deprivation schedule with food available only in the maze and for 30 min after the daily trial. In the standard experiment, the rat placed on the floor outside the maze ran under the choice platform, climbed onto it through the central opening, chose one of the maze channels, ate the reward, descended to the floor and made the next choice. The animals were allowed 12 or 24 choices, respectively, and were removed from the apparatus after completing the last choice. The channels were numbered clockwise (l-12 or P-24) and the sequence of choices was recorded. Entering an already visited channel was considered an error. RESULTS
Performance
in the 1Barm
maze
The rats were trained in the 12-arm radial maze from the age of 3 months. With one daily trial, performance became asymptotic after 4 to 6 weeks. The animals had a tendency to enter maze arms opposite rather than adjacent to the channel just left (Magni et al., 1979), but no systematic strategies were displayed. At the time of the experiment the rats were highly overtrained. In the last three trials before introducing the 24-arm maze they made 1.01 f 0.12 errors on the average (n = 36), 70% of which occurred in the last three choices. Even on the last choice they performed, however, highly above chance (31% errors instead of the predicted 86%). Their behavior is illustrated by Fig. 1 showing the average incidence of the observed errors (pO) in choices 1 to 12 in this experiment (curve a). The observed values are-compared with the error incidence pm predicted by the random choice model (curve b). The error probability pm during the j-th choice is given by the equation 1
Pmj=
i
j-l
C i=l
(1 -Pmi)
(1)
65
Fig. 1. The incidence of errors (ordinate) on choices l-12 (abscissa) in the 12-arm radial maze. Asymptotic performance based on three trials in 12 rats. a, observed error predicted incidence pO ; b, modelled random choice behavior pm ; c, error probability from the observed errors in preceding choices p,.. For details see text.
where n is the number of the radial maze channels and i the number of choices already made. Curve c represents the probability of an error p,. on the j-th choice calculated on the basis of the observed error incidence on choices 1 to j - 1 according to the equation Prj=
1
n
j-l
c i= 1
(2)
(1 - Poi)
In other words, the error probability on the j-th choice corresponds to the sum of different channels entered on the preceding j - 1 choices divided by the number of maze channels. The observed error probability p. is considerably below the predicted one (p,) throughout the trial. Performance
in the 24-arm
maze
The animal’s performance during the first three trials in the 24-arm radial maze is shown in Fig. 2. The average number of errors per trial was 3.89 f 0.20 (n = 35), that is significantly less than would correspond to random choice behavior (8.64 errors per trial). The observed error incidence bo) was well below the predicted one @) during the first 18 choices but approached the latter in choices 23 and 24. This indicates that the choice behavior was not significantly influenced by the contents of the working memory after 22 choices. With further training in the 24-arm maze the performance somewhat improved and reached the asymptotic level of 2.78 * 0.37 errors per trial illustrated in Fig. 3. The observed errors stayed below
66 % 100
I
b0
-
40
-
10
-
b
.‘,’ ::’ .i’
6
18
12
Fig. 2. The incidence of errors (ordinate) trials in the 24-arm radial maze. Average. Fig. 1.
Fig. 3. The incidence of errors maze. Asymptotic performance Fig. 1.
IU
ch
on choices l-24 (abscissa) in the first three data from 12 rats. .Other description as in
(ordinate) on choices l-24 (abscissa) in the 24-arm radial based on three trials in 12 rats. Other description as in
the predicted ones even on choices 23 and 24. The regression line calculated from p. values on choices 14 to 24 had a steeper slope (k = 1.25) than the regression line based on choices 7 to 12 in the 12-arm maze (k = 0.67). The difference in performance of the 12- and 24-arm maze tasks can be assessed by dividing the average number of observed errors per trial by the number of errors predicted by the random choice model. The index is 0.23 and 0.32 for the 12- and 24-arm mazes, respectively. In the last 25% of
67
choices (i.e. choices 10 to 12 and 19 to 24) the average incidence of errors was 24% and 37%, respectively, in the last 8.33% of choices (i.e. choices 12 and 23, 24) 31% and 51.1%, respectively. Twenty-four
choices
in the same 12-arm
maze
The increased difficulty of the 24-arm maze task can be due not only to the necessity to keep more items in the working memory but also to more difficult discrimination between the items which are separated by only 15” angles in the 24-arm maze. To test this factor, 24 choices were given in two immediately successive trials in the same 12-arm maze in which the feeders were rebaited when the animal was making the twelfth and thirteenth choice. Rats made 2.74 f 0.32 (n = 23) errors defined as re-entering the same channel more than twice in the 24 choices. The performance was better in the first (1.17 f 0.21) than in t.he second trial (2.13 ? 0.17) (paired comparison, t (22) = 3.55, P < 0.003). The good performance in choices 13 to 24 can be due to resetting of the working memory (Olton, 1978) after the first 12 choices. Indeed, some rats hesitated after completion of the first trial to make the next choice, which may indicate recognition of the trial boundary. An alternative explanation is that rats tend to avoid choice repetition in any continuous string of 12 choices. This would be most efficiently accomplished by repeating the same sequence of choices on all trials. Although this was not the case, analysis of errors (defined as repetition of the same choice within a continuous series of 12 choices - Fig. 4 above) showed that their average incidence increased rather abruptly between blocks 1 to 12 and 4 to 15 (Fig. 4, middle). The incidence of errors in blocks 7--18 and 8--19 was significantly higher than in blocks 11-22 and 12-23 (paired comparison, t (22) = 3.8, P < 0.003). This finding supports the reset explanation. On the other hand, average distribution of repeated choices in the thirteen 12-choice sequences 1 to 12, 2 to 13, . . . 12 to 23 and 13 to 24 shows that rats do not re-enter the same channel in 3 to 6 subsequent choices (Fig. 4, below). It seems that both resetting and avoidance of choice repetition contribute to the final result. Twenty-four
choices
in two different
1Barm
mazes
The successive use of the 12-arm maze is actually an interference test. Ideal performance requires that the rat enters each channel twice, once in the first and once in the second trial. Choice repetition which is an error in choices 1 to 12 becomes a correct response in choices 13 to 24. To distinguish between the two trials becomes particularly difficult when they follow in immediate succession. The separation of the two 12-arm maze trials can be made easier by transferring the animal between two radial mazes. In another series of experi-
68
%
100 -
SO.
60.
40.
20.
0-
Fig. 4. Performance of rats in 2 x 12 choices in the 12-arm maze. Average data from two trials in 12 rats. Above: incidence of repeated choices (ordinatepin la-choice segments l-12, 7-18, and 13-24 (abscissa). Middle: Average cumulative incidence of errors (ordinate) in thirteen la-choice segments l-12 to 13-24 (abscissa). The vertical bars indicate SEM values. Below: average distribution of repeated choices in the above 12choice segments. Ordinate: Probability of choice repetition. Abscissa: Ordinal choice number within the segment.
69
ments the rats were allowed six trials in maze A, were then removed from it and placed into a separate maze B in an adjacent compartment of the experimental room. After 12 choices in maze B the rats were returned into maze A and allowed to make the remaining six choices there. The results are illustrated in Fig. 5, comparing the incidence of errors in choices 1 to 6 A, 1 to 12 B, and 7 to 12 A with choice repetition in similar segments of Experiment 3. The performance in maze B was unaffected by the preceding si choices in maze A and reached the usual 1.09 f 0.22 errors per trial (n = 24). Performance in the last six choices in maze A was considerably impaired, however, leading to the average 2.31 f 0.19 errors in maze A. Hence, the overall error incidence in the 24 choices was 3.40 f 0.20 and was thus close to the value found in the 24-arm maze or in two successive trials in the same 1Zarm maze. Choice repetition in comparable segments of the choice field (7-18 A vs. l-12 B, l--6 and 19-24 A vs. l--6 and 7-12 A) was higher in the A-A-A than in the A-B-A sequence (Fig. 5). This indicates that change of trial boundaries was accompanied by a substantial modification of the rat’s strategy for solving the task. Performance decrement in the last six choices of the A-B-A sequence could be due to the time elapsed between choices 1 to 6 and 7 to 12 in maze A. In a control experiment the rats were removed
T
t
-
19-24 A
l-6 A
i’ 1
l-12 B
7-12 A
-
l-6 A
5 min
7-12 A
ch
Fig. 5. Error incidence in comparable segments of the 12 x 12 choices performed in the same radial maze A or in two separate mazes A and B. Data from two trials in 12 rats. A 5-min interval was substituted for choices l-12 B or 7-M A in the control experiment.
after six choices from maze A and placed into a waiting cage for the time corresponding to running the maze B (5 min). After this interval they were returned to maze A and allowed to complete the trial. The 5 min interruption had only a negligible effect on performance: there were 1.37 ? 0.26 errors per trial (n = 24). This indicates that the deterioration of performance in the last six choices of the A-B-A experiment is not due to the time elapsed but to the interfering influence of the B choices. DISCUSSION
Olton et al. (1977) and Olton et al. (1979) estimated the capacity of the rat’s working memory by extrapolating the observed and expected choice probabilities in the 17-arm maze to mazes with more arms. Such extrapolation is unwarranted, however, since curves obtained in an n-arm maze cannot be projected to an (n + m)-arm maze. This is clearly illustrated by comparison of performance in the 12-arm and 24-arm maze. Linear extrapolation of the observed error probabilities on choices 7-12 in Fig. 1 (curve a) would intercept the extrapolated curve c around choice 24. Asymptotic performance in the 24-arm maze (Fig. 3) shows, however, that the rats do much better and that error incidence is far below the predicted one even on choice 24. Extrapolative assessment of memory capacity must compare performance in comparable segments (e.g. last twelfth) of the choices. The increment of error incidence in the last choice in the 12-arm maze and in the last two choices in the 24-arm maze is surprisingly small (from 31.0 to 51.2%) and linear extrapolation leads to projected working memory capacity of about 40 to 50 items. A similar estimate has recently been proposed by Nadel (1979). Experimental testing of this extrapolation is impractical, however, since the increasing running time (about 20 to 30 s per choice) enhances forgetting of the initial choices. The performance deterioration would be due to a combined effect of maze complexity and forgetting. The assumption that memory capacity is reached when the observed error incidence equals predicted error probability should be qualified, however. The latter value is obtained by dividing the number of already entered channels by the number of maze arms (see eq. 2). Avoidance of choice repetition effectively diminishes the number of choices available to the animal and decreases thus both the denominator and numerator of the pr ratio. Assuming that the rat can avoid entering the last visited 10 channels, the pr value for choice 24 in Fig. 3 should be corrected from 20.7/24 = 0.86 to 10.7/14 = 0.76. The relative significance of this correction decreases with increasing number of maze channels and does not substantially influence the estimation of memory capacity. Comparison of Fig. 2 and Fig. 3 indicates that orienting in the 12-arm maze does not completely transfer to the 24arm maze. The initial performance deficit may reflect poor cognitive representation of the 24-arm maze, the arms of which are not yet perceived as
71
elements of a single configuration. Such configurational interpretation, originally proposed for rats in the divergent arm maze (Lachman and Brown, 1957), may also account for the resetting effect seen in Experiment 3. The comparison of performance in a 24-arm maze and in two successive or overlapping tests in a 12-arm maze yielded very similar quantitative results which are, however, due to different mechanisms. Decrease of performance during eight successive tests in an eight-arm maze was reported by Olton (1978). Although in his experiments the trial separation was stressed by 1-min interval and confinement between the eight-choice blocks, his results can also be due to the tendency of rats to separate identical choices by the maximum number of other choices. Anyway, in the first 3 X 8 choices Olton’s rats made only 0.917 errors which is substantially less than in the 2 X 12 or 1 X 24 choices (3.30 and 2.66 errors, respectively) in the present experiment. This difference can be due to procedural variables (elevated versus tubular maze), but it more probably reflects improved avoidance of choice repetition with short choice sequences. As follows from the analysis of two successive trials in the 12-arm maze (Fig. 4), probability of choice repetition in eight consecutive choices is 0.47, which would lead to 1.41 errors in 3 X 8 choices. This value was further reduced by the more prominent role of resetting. Experiment 3 can be evaluated in a more global way when error is defined as entering the same channel more than twice. Such global performance measure can considerably differ from the performance measures based on 12-choice segments of the same data. With an errorless global performance, segment errors can vary from zero (when the sequence of the first 12 choices is exactly repeated in the second trial) to 12 (when each choice is immediately repeated, e.g. 1, 1, 5, 5, 10, 10, etc). The global error incidence in Experiment 3 was 2.74 and did not significantly differ from the 3.30 errors obtained by addition of errors in the first and second 12choice segments. On the other hand, division of the A-A-A choice field in the way corresponding to the A-B-A sequence of Experiment 4 yielded a high error score (5.52), indicating artificiality of such segmentation. The channels visited in the first six choices were revisited in the last six choices of the A-A-A sequence with the same probability as the unentere.d ones. This independence of choices l-6 and 19924 was obviously due to absence of cues connecting them into a trial. When such cues were provided in Experiment 4, performance in the last six choices indicated significant memory for choices l-6. The positive role of the resetting mechanism has its limits, however. Even when choice lists are made easily recognizable, their overlapping causes interference effects, probably due to competition for a limited storage space. The above result contrasts with a formally similar experiment reported by Maki et al. (1979). These authors removed rats from an eight-arm elevated maze after choice 4 and let them perform another spatial task in a four-arm radial maze. When the rats were returned immediately afterwards
72
to the eight-arm maze, their performance in the remaining four choices (95% correct) was not significantly different from control trials (98% correct). It seems that absence of significant interference in this experiment was due to a relatively low memory load (total of 12 choices as compared with the total of 24 choices in the present A-B-A experiment) and to various procedural differences: elevated maze versus tubular maze, different maze configurations (eight arms and four arms) versus same maze configuration (12 arms, 12 arms) during the basic trial and the interference treatment.
REFERENCES
BureHovl,O., 1930.
Spatial memory and instrumental conditioning. Acta Neurobiol. Exp., 40: 51-65. Lachman, S.J. and Brown, C.R., 1957. Behavior in a free choice multiple path elimination problem. J. Psychol., 43: 27-40. Magni, S., Krekule, I. and Bure& J., 1979. Radial maze type as a determinant of the choice behavior of rats. J. Neurosci. Meth., 1: 343-352. Maki, W.S., Brokofsky, S. and Berg, B., 1979. Spatial memory in rats: resistence to retroactive interference. Anim. Learning Behav., 7: 25-30. Nadel, L., 1979. Working memory won’t work. Behav. Brain Sci., 2: 338-339. Olton, D.S., 1978. Characteristics of spatial memory. In: S.H. Hulse, H. Fowler and W.K. Honig (Eds.), Cognitive Aspects of Animal Behavior. Lawrence Erlbaum, Hilsdalle, N.J., pp. 341-373. Olton, D.S. and Samuelson, R.J., 1976. Remembrance of places passed: spatial memory in rats. J. Exp. Psychol., 2: 97-116. Olton, D.S., Collison, C. and Werz, M.A., 1977. Spatial memory and radial arm maze performance of rats. Learn. Motiv., 8: 289-314. Olton, D.S., Becker, J.T. and Hendelman, G.E., 1979. Hippocampus, space, and memory. Behav. Brain. Sci., 2: 313-322. Roberts, W.A., 1979. Spatial memory in the rat on a hierarchical maze. Learn. Motiv., 10: 117-140.
63 Amsterdam
-
Printed
in The Netherlands
CAPACITY OF WORKING MEMORY IN RATS AS DETERMINED PERFORMANCE ON A RADIAL MAZE
0. BURESOVA
and J. BURES
Institute of Physiology, 4-K+ (Czechoslovakia) (Accepted
1 July
BY
Czechoslovak
Academy
of Sciences,
Videhska’
1083
142-20
Prague
1981)
ABSTRACT BureXova, 0. and BureH, J., 1982. Capacity of working memory performance on a radial maze. Behav. Processes, 7: 63-72.
in rats as determined
by
Capacity of the working memory was tested in 12 rats highly overtrained in the 12and 24-arm radial mazes. Asymptotic performance levels were characterized by 1.01 and 2.78 errors/trial in the 12- and 24-arm mazes, respectively. The incidence of errors increased from 31% on the last choice in the 12-arm maze to 51% on choices 23 and 24 in the 24-arm maze, but remained significantly below the expected error probability of about 85%. Linear extrapolation of the above trend to mazes with more arms suggests working memory capacity of 40 to 50 items. When two trials in a 12-arm maze were repeated in immediate succession, error incidence increased from 1.17 in the first trial to 2.13 in the second trial. The tendency to avoid choice repetition could be observed in any string of 12 continuous choices, but was weakest in segments divided by trial boundary (2.48 errors in choices 7 to 18). With a different trial separation (choices l-6 and 19-24 in maze A, choices 7-18 in an adjacent maze B) errors dropped to 1.09 in B but increased to 2.30 in A. It is concluded that radial maze performance reflects avoidance of choice repetition which is improved by recognition of trial boundaries and is adversely influenced by forgetting and interference.
INTRODUCTION
The spectacular performance of rats in the radial maze (Olton and Samuelson, 1976; Olton et al., 1979) has raised the question what are the limits of the rat’s working memory in this task. Olton et al. (1977) attempted to estimate its capacity by extrapolating the decreasing probability of correct responding in the 17-arm maze. They suggested that choice accuracy would decrease to chance levels at 25 to 30 choices, provided that the performance decay is a linear function of the choice number. Roberts (1979) used a radial maze with eight primary arms, each giving access to three secondary arms, baited with food. The rats had to locate 24 food pellets, but the branching geometry of the maze precludes direct comparison of their performance with the radial maze results. The present study compares the choice behavior of rats in the 12- and 24-arm radial mazes. 0376-6357/82/000~0000/$
02.75
o 1982
Elsevier
Scientific
Publishing
Company
64 METHOD
The apparatus was a tubular radial maze of the type described by Magni et al. (1979) and Buresova (1980). The choice platform (40 or 80 cm in diameter, respectively) had a central opening (8 cm in diameter) and was surrounded by 40 cm high plexiglass walls. One-way swing-doors (6 X 6 cm) provided access into the 35 cm long alleys with perforated ceiling and a recessed food cup at the far end. One-way exit door connected the channel with the outside area separated from the rest of the room by a 50 cm high plexiglass wall. The maze rested on 5 cm high supporting legs and could be entered through the space between the floor and the choice platform. The animals were 12 male hooded rats 6 months old at the time of the experiment. They were maintained on a 24 h food deprivation schedule with food available only in the maze and for 30 min after the daily trial. In the standard experiment, the rat placed on the floor outside the maze ran under the choice platform, climbed onto it through the central opening, chose one of the maze channels, ate the reward, descended to the floor and made the next choice. The animals were allowed 12 or 24 choices, respectively, and were removed from the apparatus after completing the last choice. The channels were numbered clockwise (l-12 or P-24) and the sequence of choices was recorded. Entering an already visited channel was considered an error. RESULTS
Performance
in the 1Barm
maze
The rats were trained in the 12-arm radial maze from the age of 3 months. With one daily trial, performance became asymptotic after 4 to 6 weeks. The animals had a tendency to enter maze arms opposite rather than adjacent to the channel just left (Magni et al., 1979), but no systematic strategies were displayed. At the time of the experiment the rats were highly overtrained. In the last three trials before introducing the 24-arm maze they made 1.01 f 0.12 errors on the average (n = 36), 70% of which occurred in the last three choices. Even on the last choice they performed, however, highly above chance (31% errors instead of the predicted 86%). Their behavior is illustrated by Fig. 1 showing the average incidence of the observed errors (pO) in choices 1 to 12 in this experiment (curve a). The observed values are-compared with the error incidence pm predicted by the random choice model (curve b). The error probability pm during the j-th choice is given by the equation 1
Pmj=
i
j-l
C i=l
(1 -Pmi)
(1)
65
Fig. 1. The incidence of errors (ordinate) on choices l-12 (abscissa) in the 12-arm radial maze. Asymptotic performance based on three trials in 12 rats. a, observed error predicted incidence pO ; b, modelled random choice behavior pm ; c, error probability from the observed errors in preceding choices p,.. For details see text.
where n is the number of the radial maze channels and i the number of choices already made. Curve c represents the probability of an error p,. on the j-th choice calculated on the basis of the observed error incidence on choices 1 to j - 1 according to the equation Prj=
1
n
j-l
c i= 1
(2)
(1 - Poi)
In other words, the error probability on the j-th choice corresponds to the sum of different channels entered on the preceding j - 1 choices divided by the number of maze channels. The observed error probability p. is considerably below the predicted one (p,) throughout the trial. Performance
in the 24-arm
maze
The animal’s performance during the first three trials in the 24-arm radial maze is shown in Fig. 2. The average number of errors per trial was 3.89 f 0.20 (n = 35), that is significantly less than would correspond to random choice behavior (8.64 errors per trial). The observed error incidence bo) was well below the predicted one @) during the first 18 choices but approached the latter in choices 23 and 24. This indicates that the choice behavior was not significantly influenced by the contents of the working memory after 22 choices. With further training in the 24-arm maze the performance somewhat improved and reached the asymptotic level of 2.78 * 0.37 errors per trial illustrated in Fig. 3. The observed errors stayed below
66 % 100
I
b0
-
40
-
10
-
b
.‘,’ ::’ .i’
6
18
12
Fig. 2. The incidence of errors (ordinate) trials in the 24-arm radial maze. Average. Fig. 1.
Fig. 3. The incidence of errors maze. Asymptotic performance Fig. 1.
IU
ch
on choices l-24 (abscissa) in the first three data from 12 rats. .Other description as in
(ordinate) on choices l-24 (abscissa) in the 24-arm radial based on three trials in 12 rats. Other description as in
the predicted ones even on choices 23 and 24. The regression line calculated from p. values on choices 14 to 24 had a steeper slope (k = 1.25) than the regression line based on choices 7 to 12 in the 12-arm maze (k = 0.67). The difference in performance of the 12- and 24-arm maze tasks can be assessed by dividing the average number of observed errors per trial by the number of errors predicted by the random choice model. The index is 0.23 and 0.32 for the 12- and 24-arm mazes, respectively. In the last 25% of
67
choices (i.e. choices 10 to 12 and 19 to 24) the average incidence of errors was 24% and 37%, respectively, in the last 8.33% of choices (i.e. choices 12 and 23, 24) 31% and 51.1%, respectively. Twenty-four
choices
in the same 12-arm
maze
The increased difficulty of the 24-arm maze task can be due not only to the necessity to keep more items in the working memory but also to more difficult discrimination between the items which are separated by only 15” angles in the 24-arm maze. To test this factor, 24 choices were given in two immediately successive trials in the same 12-arm maze in which the feeders were rebaited when the animal was making the twelfth and thirteenth choice. Rats made 2.74 f 0.32 (n = 23) errors defined as re-entering the same channel more than twice in the 24 choices. The performance was better in the first (1.17 f 0.21) than in t.he second trial (2.13 ? 0.17) (paired comparison, t (22) = 3.55, P < 0.003). The good performance in choices 13 to 24 can be due to resetting of the working memory (Olton, 1978) after the first 12 choices. Indeed, some rats hesitated after completion of the first trial to make the next choice, which may indicate recognition of the trial boundary. An alternative explanation is that rats tend to avoid choice repetition in any continuous string of 12 choices. This would be most efficiently accomplished by repeating the same sequence of choices on all trials. Although this was not the case, analysis of errors (defined as repetition of the same choice within a continuous series of 12 choices - Fig. 4 above) showed that their average incidence increased rather abruptly between blocks 1 to 12 and 4 to 15 (Fig. 4, middle). The incidence of errors in blocks 7--18 and 8--19 was significantly higher than in blocks 11-22 and 12-23 (paired comparison, t (22) = 3.8, P < 0.003). This finding supports the reset explanation. On the other hand, average distribution of repeated choices in the thirteen 12-choice sequences 1 to 12, 2 to 13, . . . 12 to 23 and 13 to 24 shows that rats do not re-enter the same channel in 3 to 6 subsequent choices (Fig. 4, below). It seems that both resetting and avoidance of choice repetition contribute to the final result. Twenty-four
choices
in two different
1Barm
mazes
The successive use of the 12-arm maze is actually an interference test. Ideal performance requires that the rat enters each channel twice, once in the first and once in the second trial. Choice repetition which is an error in choices 1 to 12 becomes a correct response in choices 13 to 24. To distinguish between the two trials becomes particularly difficult when they follow in immediate succession. The separation of the two 12-arm maze trials can be made easier by transferring the animal between two radial mazes. In another series of experi-
68
%
100 -
SO.
60.
40.
20.
0-
Fig. 4. Performance of rats in 2 x 12 choices in the 12-arm maze. Average data from two trials in 12 rats. Above: incidence of repeated choices (ordinatepin la-choice segments l-12, 7-18, and 13-24 (abscissa). Middle: Average cumulative incidence of errors (ordinate) in thirteen la-choice segments l-12 to 13-24 (abscissa). The vertical bars indicate SEM values. Below: average distribution of repeated choices in the above 12choice segments. Ordinate: Probability of choice repetition. Abscissa: Ordinal choice number within the segment.
69
ments the rats were allowed six trials in maze A, were then removed from it and placed into a separate maze B in an adjacent compartment of the experimental room. After 12 choices in maze B the rats were returned into maze A and allowed to make the remaining six choices there. The results are illustrated in Fig. 5, comparing the incidence of errors in choices 1 to 6 A, 1 to 12 B, and 7 to 12 A with choice repetition in similar segments of Experiment 3. The performance in maze B was unaffected by the preceding si choices in maze A and reached the usual 1.09 f 0.22 errors per trial (n = 24). Performance in the last six choices in maze A was considerably impaired, however, leading to the average 2.31 f 0.19 errors in maze A. Hence, the overall error incidence in the 24 choices was 3.40 f 0.20 and was thus close to the value found in the 24-arm maze or in two successive trials in the same 1Zarm maze. Choice repetition in comparable segments of the choice field (7-18 A vs. l-12 B, l--6 and 19-24 A vs. l--6 and 7-12 A) was higher in the A-A-A than in the A-B-A sequence (Fig. 5). This indicates that change of trial boundaries was accompanied by a substantial modification of the rat’s strategy for solving the task. Performance decrement in the last six choices of the A-B-A sequence could be due to the time elapsed between choices 1 to 6 and 7 to 12 in maze A. In a control experiment the rats were removed
T
t
-
19-24 A
l-6 A
i’ 1
l-12 B
7-12 A
-
l-6 A
5 min
7-12 A
ch
Fig. 5. Error incidence in comparable segments of the 12 x 12 choices performed in the same radial maze A or in two separate mazes A and B. Data from two trials in 12 rats. A 5-min interval was substituted for choices l-12 B or 7-M A in the control experiment.
after six choices from maze A and placed into a waiting cage for the time corresponding to running the maze B (5 min). After this interval they were returned to maze A and allowed to complete the trial. The 5 min interruption had only a negligible effect on performance: there were 1.37 ? 0.26 errors per trial (n = 24). This indicates that the deterioration of performance in the last six choices of the A-B-A experiment is not due to the time elapsed but to the interfering influence of the B choices. DISCUSSION
Olton et al. (1977) and Olton et al. (1979) estimated the capacity of the rat’s working memory by extrapolating the observed and expected choice probabilities in the 17-arm maze to mazes with more arms. Such extrapolation is unwarranted, however, since curves obtained in an n-arm maze cannot be projected to an (n + m)-arm maze. This is clearly illustrated by comparison of performance in the 12-arm and 24-arm maze. Linear extrapolation of the observed error probabilities on choices 7-12 in Fig. 1 (curve a) would intercept the extrapolated curve c around choice 24. Asymptotic performance in the 24-arm maze (Fig. 3) shows, however, that the rats do much better and that error incidence is far below the predicted one even on choice 24. Extrapolative assessment of memory capacity must compare performance in comparable segments (e.g. last twelfth) of the choices. The increment of error incidence in the last choice in the 12-arm maze and in the last two choices in the 24-arm maze is surprisingly small (from 31.0 to 51.2%) and linear extrapolation leads to projected working memory capacity of about 40 to 50 items. A similar estimate has recently been proposed by Nadel (1979). Experimental testing of this extrapolation is impractical, however, since the increasing running time (about 20 to 30 s per choice) enhances forgetting of the initial choices. The performance deterioration would be due to a combined effect of maze complexity and forgetting. The assumption that memory capacity is reached when the observed error incidence equals predicted error probability should be qualified, however. The latter value is obtained by dividing the number of already entered channels by the number of maze arms (see eq. 2). Avoidance of choice repetition effectively diminishes the number of choices available to the animal and decreases thus both the denominator and numerator of the pr ratio. Assuming that the rat can avoid entering the last visited 10 channels, the pr value for choice 24 in Fig. 3 should be corrected from 20.7/24 = 0.86 to 10.7/14 = 0.76. The relative significance of this correction decreases with increasing number of maze channels and does not substantially influence the estimation of memory capacity. Comparison of Fig. 2 and Fig. 3 indicates that orienting in the 12-arm maze does not completely transfer to the 24arm maze. The initial performance deficit may reflect poor cognitive representation of the 24-arm maze, the arms of which are not yet perceived as
71
elements of a single configuration. Such configurational interpretation, originally proposed for rats in the divergent arm maze (Lachman and Brown, 1957), may also account for the resetting effect seen in Experiment 3. The comparison of performance in a 24-arm maze and in two successive or overlapping tests in a 12-arm maze yielded very similar quantitative results which are, however, due to different mechanisms. Decrease of performance during eight successive tests in an eight-arm maze was reported by Olton (1978). Although in his experiments the trial separation was stressed by 1-min interval and confinement between the eight-choice blocks, his results can also be due to the tendency of rats to separate identical choices by the maximum number of other choices. Anyway, in the first 3 X 8 choices Olton’s rats made only 0.917 errors which is substantially less than in the 2 X 12 or 1 X 24 choices (3.30 and 2.66 errors, respectively) in the present experiment. This difference can be due to procedural variables (elevated versus tubular maze), but it more probably reflects improved avoidance of choice repetition with short choice sequences. As follows from the analysis of two successive trials in the 12-arm maze (Fig. 4), probability of choice repetition in eight consecutive choices is 0.47, which would lead to 1.41 errors in 3 X 8 choices. This value was further reduced by the more prominent role of resetting. Experiment 3 can be evaluated in a more global way when error is defined as entering the same channel more than twice. Such global performance measure can considerably differ from the performance measures based on 12-choice segments of the same data. With an errorless global performance, segment errors can vary from zero (when the sequence of the first 12 choices is exactly repeated in the second trial) to 12 (when each choice is immediately repeated, e.g. 1, 1, 5, 5, 10, 10, etc). The global error incidence in Experiment 3 was 2.74 and did not significantly differ from the 3.30 errors obtained by addition of errors in the first and second 12choice segments. On the other hand, division of the A-A-A choice field in the way corresponding to the A-B-A sequence of Experiment 4 yielded a high error score (5.52), indicating artificiality of such segmentation. The channels visited in the first six choices were revisited in the last six choices of the A-A-A sequence with the same probability as the unentere.d ones. This independence of choices l-6 and 19924 was obviously due to absence of cues connecting them into a trial. When such cues were provided in Experiment 4, performance in the last six choices indicated significant memory for choices l-6. The positive role of the resetting mechanism has its limits, however. Even when choice lists are made easily recognizable, their overlapping causes interference effects, probably due to competition for a limited storage space. The above result contrasts with a formally similar experiment reported by Maki et al. (1979). These authors removed rats from an eight-arm elevated maze after choice 4 and let them perform another spatial task in a four-arm radial maze. When the rats were returned immediately afterwards
72
to the eight-arm maze, their performance in the remaining four choices (95% correct) was not significantly different from control trials (98% correct). It seems that absence of significant interference in this experiment was due to a relatively low memory load (total of 12 choices as compared with the total of 24 choices in the present A-B-A experiment) and to various procedural differences: elevated maze versus tubular maze, different maze configurations (eight arms and four arms) versus same maze configuration (12 arms, 12 arms) during the basic trial and the interference treatment.
REFERENCES
BureHovl,O., 1930.
Spatial memory and instrumental conditioning. Acta Neurobiol. Exp., 40: 51-65. Lachman, S.J. and Brown, C.R., 1957. Behavior in a free choice multiple path elimination problem. J. Psychol., 43: 27-40. Magni, S., Krekule, I. and Bure& J., 1979. Radial maze type as a determinant of the choice behavior of rats. J. Neurosci. Meth., 1: 343-352. Maki, W.S., Brokofsky, S. and Berg, B., 1979. Spatial memory in rats: resistence to retroactive interference. Anim. Learning Behav., 7: 25-30. Nadel, L., 1979. Working memory won’t work. Behav. Brain Sci., 2: 338-339. Olton, D.S., 1978. Characteristics of spatial memory. In: S.H. Hulse, H. Fowler and W.K. Honig (Eds.), Cognitive Aspects of Animal Behavior. Lawrence Erlbaum, Hilsdalle, N.J., pp. 341-373. Olton, D.S. and Samuelson, R.J., 1976. Remembrance of places passed: spatial memory in rats. J. Exp. Psychol., 2: 97-116. Olton, D.S., Collison, C. and Werz, M.A., 1977. Spatial memory and radial arm maze performance of rats. Learn. Motiv., 8: 289-314. Olton, D.S., Becker, J.T. and Hendelman, G.E., 1979. Hippocampus, space, and memory. Behav. Brain. Sci., 2: 313-322. Roberts, W.A., 1979. Spatial memory in the rat on a hierarchical maze. Learn. Motiv., 10: 117-140.
