Animal Learning & Behavior

1994, 22 (4), 409–420

Performance of goldfish trained in allocentric

Goldfish were trained to obtain food in a four-arm maze placed in a room with relevant spatial
cues. Four experimental conditions were run: allocentric, egocentric, egocentric + allocentric, and
control. Relative to controls, all groups were able to solve the different tasks with high accuracy after
1 week of training. Subsequent transfer tests revealed place and response strategies for allocentric
and egocentric groups, respectively, and both types of strategies for the ego-allocentric group. More-
over, the allocentric group showed the capacity to choose the appropriate trajectory toward the goal,
even from novel starting points, presumably by using the distal cues as a whole. The results suggest
that, in addition to using egocentric strategies, goldfish are able to solve spatial tasks on the basis
of allocentric frames of reference and to build complex spatial cognitive representations of their
environment.

Fishes travel across a wide range of distances with sur- taneously a system of learned, route-specific landmarks
prising efficiency, whether in intercontinental migrations in conjunction with solar compass orientation and per-
or on excursions within their habitual living areas. These haps a cognitive map (Reese, 1989). In summary, natu-
travels indicate remarkable spatial abilities of fishes in ralistic fish studies describe in detail the various aspects
navigating, orienting themselves, piloting, and recogniz- of fishes’ travel but are rather imprecise about the spa-
ing their environment. Research in this field has been fo- tial strategies that may be employed in travel.
cused mainly on the innate fixed patterns of behavior and Spatial learning and memory capabilities of fishes
on sensory, ecological, and zoological factors. What seems have been examined more closely in laboratory experi-
to be underestimated in this research, though, is the pos- ments. For example, the performance of Siamese fight-
sibility that spatial behavior is a flexible process that in- ing fish (Betta splendens) in the eight-arm maze proce-
volves learning and memory mechanisms and cognitive dure shows a strong algorithmic component in the
phenomena (for a review, see Dodson, 1988). Such mech- determination of the sequential choices, although some
anisms are indicated by a number of naturalistic and ex- small amount of spatial memory is involved in the task
perimental studies, such as the pioneer work of Aronson (Roitblat, Tham, & Golub, 1982). Goldfish (Carassius
(1951, 1971) that showed that the gobiid fish Bathygob- auratus) have been shown to remember the spatial posi-
ius soporator uses learned information about the spatial tion of food patches in a tank (Pitcher & Magurran, 1983),
relationships of tide pools, or topographical memories to use landmarks as indirect spatial reference points
acquired during exploration, to orient itself accurately. In (Warburton, 1990), and also to swim in a constant direc-
fact, complex spatial learning and memory capabilities tion relative to visual cues, even if they approach the
in fishes can be inferred from recent naturalistic studies goal from opposite directions. This ability could reflect
(Hallacher, 1984; Helfman, Meyer, & McFarland, 1982; a capacity for discriminating spatial relationships in the
Helfman & Schultz, 1984; Markevich, 1988). Further- environment independently of a body-centered reference
more, it has been suggested that some fishes use simul- system (Ingle & Sahagian, 1973). Fishes also show an
organized pattern of exploration when they are intro-
duced into a novel environment, indicating some degree
of spatial knowledge and memory (Kleerekoper, Matis,
We wish to thank J. Bruce Overmier and Catherine Thinus-Blanc for Gensler, & Maynard, 1974), and they can detect envi-
their helpful suggestions and critical reading of a previous version of
this manuscript. We also greatly appreciated the comments of Peter J. ronmental modifications, reacting with increased ex-
Urcuioli and two anonymous reviewers, which led to improvements in ploratory activity (Welker & Welker, 1958). Recently,
this paper. This work has been supported by grants from CICYT, the Teyke (1989) reported that following release in unfa-
Junta de Andalucía, and the FISSS. Requests for reprints should be ad- miliar surroundings, or after alteration of a familiar en-
dressed to F. Rodríguez, Laboratorio de Psicobiología, Grupo Neuro-
biología de Vertebrados, Departamento de Fisiología y Biología Ani-
vironment, the blind cave fish (Anoptichthys jordani)
mal, Facultad de Psicología, Avda. San Francisco Javier s/n, 41005 swim around boundary features and increase their swim-
Sevilla, Spain. ming velocity, possibly to optimize lateral line organ

409 Copyright 1994 Psychonomic Society, Inc.

stimulation. He proposed that fishes develop an internal they mutually exclusive? (4) Does prolonged training
map of their environment that allows them to swim at ve- differentially affect these strategies, and might they be
locities below the optimal range in familiar environments revealed by the reversal of the task?
without observable deficits in avoiding obstacles.
The preceding studies suggest the presence of com- METHOD
plex spatial learning and memory capabilities in fishes
and the participation of cognitive mechanisms in the so- Subjects
Goldfish (Carassius auratus) obtained from a local supplier
lution of spatial problems. The concept of a cognitive were maintained, for 2 months prior to the experiments, in small
map, first suggested by Tolman (1948), has become a groups in glass aquaria with aerated filtered water at 20ºC
2º.
major theoretical construct since the formulation of the They were kept on a 14:10-h light:dark cycle and given food ad li-
spatial learning and memory theory of O`Keefe and Nadel bitum during this time.
(1978). This theory postulates the presence of two dif- Thirty-two animals, 10–14 cm in length, were randomly se-
ferent spatial behavior systems: the cognitive mapping lected from stock tanks and assigned to four groups. They were
and taxon systems. They appear to act independently, but, housed in similar aquaria and deprived of any food for 2 days be-
fore the experiment. The animals were recognized by phenotypi-
under the appropriate conditions, they may cooperate or cal characters. Throughout the experiment, the fishes received only
conflict with each other (O`Keefe & Nadel, 1978; the five food sticks they obtained every day in the experimental
Schenk & Morris, 1985; Whishaw, 1989; Whishaw & session. Each stick was 20
0.8 mg (mean
SD) of dry food for
Mittleman, 1986). The cognitive mapping system is based pond fish (Tetra pond).
on allocentric strategies, it involves an internal map-like
representation of the objective spatial relationships among Apparatus
The apparatus was an elevated four-arm maze made out of trans-
different landmarks, and it is independent of any partic- parent glass; the floor was covered with opaque white Perspex.
ular view of the surrounding (Mazmanian & Roberts, Only three of the arms were used on any one conditioning trial.
1983; Morris, 1981; O’Keefe & Conway, 1978; Suzuki, Each arm was 80 cm long, 17 cm wide, and 20 cm high, with a 17-
Augerinos, & Black, 1980). The internal spatial repre- cm-square central platform. For each trial, the startbox was de-
sentations of this system provide animals with greater limited by a 20-cm-high opaque Perspex guillotine door placed
flexibility in their spatial behavior than that offered by 16 cm from the end of the arm. The door was controlled by a hand
operated device. Before every session, the maze was filled with
taxon strategies. Hence, animals that rely on such a cog- water at 20ºC
1º to a depth of 15 cm and aerated at the end of
nitive, topographical representation can locate a place each arm. At the end of each of the two non–start arms, a food
from different directions and adopt novel routes from holder was attached by suction to the floor of the arm. The food
points previously unvisited, even in the absence of any holder was composed of a 90º curved glass bar (5 mm in diame-
specific cue associated with the goal. By contrast, the ter). The upper segment of the bar was horizontal, elevated 5 cm
taxon systems are based on purely egocentric strategies above the maze floor. At one end of the bar was inserted a dark
and include guidance and orientation. The guidance strate- latex tube (1.5 cm long), which could enclose one stick of food.
This food was not visible, and the fish took it out by suction. The
gies involve approaching or avoiding particular cues,
whereas orientation indicates a body-centered turn in re-
sponse to a cue. Properties of these two systems have
been described elsewhere (O`Keefe & Nadel, 1978) and
have been thoroughly studied in mammals (Nadel, 1991;
O’Keefe & Nadel, 1978) and birds (Bingman, 1990;
Bingman, Bagnoli, Ioalé, & Casini, 1989). To date, how-
ever, the relevance of such systems is not well known in
fishes. Comparative research with fish could reveal in-
teresting insights about the evolutionary and adaptive
importance of these strategies in vertebrates.
The general aim of this work was to determine whether
fishes are able to solve spatial tasks by using allocentric
frames of reference in addition to the purely body-centered
ones. Furthermore, we investigated whether fishes could
build and use internal maps of their environment, and
whether such hypothetical internal maps show some of
the properties described by O`Keefe and Nadel (1978).
With this purpose, the performances of goldfish trained
in allocentric versus egocentric tasks in a four-arm maze
were compared. In particular, the following questions
were addressed: (1) Are differences in learning evident
in allocentric versus egocentric tasks? (2) What kind of
Figure 1. Plan to scale of the experimental room and maze. The
cues is used to solve allocentric tasks? For example, do maze location is shown in its normal position (solid line) and in its dis-
goldfish use distal visual cues to locate a place? (3) Can placed positions for Type 2 transfer tests (dashed lines). Rhombuses
allocentric and egocentric strategies act in concert or are indicate the anchorage points for the curtains.
 ALLOCENTRIC AND EGOCENTRIC SPATIAL LEARNING IN GOLDFISH 411

goal arm was the one containing the baited food holder. The maze Allocentric procedure. In this group, two opposite start points
was placed in the center of a room 4.504.642.60 m, which was were randomly used (50% each), and the fish were always re-
illuminated by four overhead lights. The plan of the room, the var- warded at the extreme of the goal arm situated in the same place
ious extramaze cues, and the location of the maze during training of the room (Figure 2A). This procedure was used to determine
are shown in Figure 1. whether or not the fish could learn to go to a rewarded site solely
The experimenter, located in an adjoining room, controlled the on the basis of extramaze cues. No fixed-turn direction was rele-
door of the maze and, through a small opening, observed the be- vant to task solution, because, depending on the location of the
havior of the animal. When called for, the maze could be surrounded startbox, subjects were required to make a left- or right-hand turn.
by thick brown curtains, which hung from ceiling to floor, and Egocentric procedure. For this group, two opposite start arms
which formed a 3-m-square enclosure. These curtains excluded vir- were randomly used (50% each), and for all trials, the goal arm was
tually all extramaze visual cues. The maze was randomly rotated determined by a fixed-turn response (e.g., always left) relative to
between experimental sessions in order to preclude the use of any the start arm (Figure 2B). This procedure determined whether or not
possible intramaze cues. the fish could choose the correct arm on the basis of a specific turn
response with extramazes cues irrelevant to task solution.
Procedure Ego-allocentric procedure. The fish in this group left a startbox
Pretraining. To get the fish to obtain food from the food holder, that was always situated in the same place of the room, and they
they were placed for 2 consecutive days in a little aquarium pro- were rewarded exclusively in a goal arm located in another constant
vided with a food holder, which was rebaited until each fish ate place (Figure 2C). This task thus allowed selection of the correct
five times. This aquarium was located outside of the experimen- arm on the basis of a specific turn direction and/or extramaze cues.
tal room. On the same days, two 2-h sessions were also run in Control procedure. The fish in the control group left from a
which the subject was placed in the experimental room and al- startbox that was always situated in a fixed place of the room.
lowed to explore freely throughout the maze. During these ses- However, their training involved two possible correct goal arms,
sions, the food holders were removed from the apparatus. Follow- which were randomly assigned across trials (Figure 2D). This
ing this exploration period, the animals were deprived of any food group controlled for the possibility that subjects might find the
for 2 days prior to the experiment proper. reward by attending to odor or other uncontrolled variables.
Training. Animals were assigned to four different groups (n = 8) To control for possible preferences for a specific place and/or
for training in one of the following procedures: allocentric, ego- for a particular turn, half of the animals in each experimental
centric, ego-allocentric, and control (see Figure 2). group were trained to choose a specific turn or place, and the other

Figure 2. Schematic diagrams of the four conditioning procedures, showing also Type 1 and 2 transfer tests for
each group. Trajectory examples are given for one half of each experimental group. For the other half, the proce-
dure was identical, except that the goal arm was the opposite for allocentric and ego-allocentric groups and the
turn conditioned was the opposite for the egocentric group. In the conditioning trials, arrows mark the most ap-
propriate trajectories from the start to the goal (asterisks) for each group. The numbers represent the percentages
of trials on which a specific combination of startbox and goal arm were used. In Type 1 and 2 transfer tests, the
arrows denote the start arms used.
412 RODRIGUEZ, DURAN, VARGAS, TORRES, AND SALAS

half were trained to choose the opposite. For example, 4 animals ures 1 and 2). The locations of the startbox for these trials are
in the egocentric group were always rewarded in the arm situated shown in Figures 2A–2D. For both types of trials, the first choice
to the left of the start arm, and the other 4 were rewarded in the was recorded and then the fish was removed from the maze. Ten
arm located to the right. transfer trials (5 of each type) were randomly intercalated among
The fish in all groups were individually trained in daily sessions the trials of the overtraining period—but never more than 1 per
of five consecutive trials. To begin a trial, the fish was gently placed session, with a minimum of 5 conditioning trials between them.
in the startbox with a small net and confined there for 20 sec. Then According to the arm chosen, the choice was designated as ego-
the guillotine door was raised and lowered after the fish left the centric, allocentric, or other. Egocentric choices were scored
startbox, which allowed it to swim freely in the accessible arms of when the animal made the same turn that was rewarded during the
the maze. A choice was recorded when the tail of the fish crossed training trials, irrespective of the start place within the room. Al-
the entrance of one arm. The fishes were left in the maze until they locentric choices were scored when the animal reached the same
obtained reward (correction procedure) or until 2 min were spent. place in the room at which it had been rewarded in the training tri-
Even though the trial was considered successful only when the ini- als, irrespective of the start place and turn direction. All choices
tial choice was correct, all errors were recorded as well as the time that could not be classified into the two latter categories were con-
to get food. Thus, it was possible for an animal to make several er- sidered other.
rors on a given trial. All errors after the first one were scored as Probe tests. To test the relevance of the distal visual cues in the
error perseverations. Once the trial was finished, the animal was solution of the different tasks, two types of probe trials were also
removed from the arm and placed into a container during a 2-min run. For Type 1 probes, one of the most salient cues in the exper-
intertrial interval. For each new trial, the fish was re-placed into imental room (e.g., the poster) was removed or hidden with brown
the startbox, and, in a random sequence, the experimenter rebaited curtains on each trial. For Type 2 probes, a brown curtain was
the corresponding food holder and simulated the rebaiting of the placed around the entire maze to exclude all the cues. Ten of these
other food holder to control for unintentional cuing. probe trials (5 of Type 1 and 5 of Type 2) were randomly interca-
The training phase lasted 54 sessions (270 trials). An acquisi- lated during overtraining trials, with a minimum of 5 conditioning
tion criterion of 13 correct trials out of 15 (a mean of 86.67% cor- trials between them. The location of the maze and the procedures
rect over three consecutive sessions) was established. When the used for each group for the probe tests were the same as those dur-
experimental groups reached criterion, the next session was con- ing the training trials, except that the food holders were removed
sidered to be the beginning of overtraining, during which the trans- from the maze (i.e., animals were not reinforced). The first arm
fer- and probe-trial tests described below were also run. choice was recorded and considered as correct if it coincided with
Transfer tests. The transfer trials were aimed at elucidating those reinforced during training.
whether the animals of the different groups solved their respective Reversal. Following Session 54, the location of food was re-
tasks on the basis of turn or place responses. All four arms of the versed. For ego-allocentric and allocentric groups, the reversal
maze were opened during these transfer trials, and all food holders consisted of a 180º shift in the rewarded arm location. For the ego-
were removed (i.e., the animals were not reinforced). The transfer centric group, the turn opposite to that previously conditioned was
trials were of two types. For Type 1 trials, the maze remained in rewarded. For the control group, the baited goal arm was randomly
its usual position in the room, but the animals were released from assigned during this phase. All other spatial characteristics of the
a startbox located in a novel place. For Type 2 trials, the maze was experimental situation remained unchanged, as did all other pro-
displaced in the room in such a way that the end of one arm was cedural details used during the training period. The reversal phase
located in the same place in the room where the fish was rewarded lasted for 15 sessions (75 trials).
during training trials, but the startbox was situated in a place in the Data analysis. To evaluate the different characteristics of the
room different from where it was situated in training trials (cf. Fig- spatial learning processes involved in the solution of each task,

Figure 3. Percentage of correct choices for each group during the experiment. Symbols represent means of
three sessions.
 ALLOCENTRIC AND EGOCENTRIC SPATIAL LEARNING IN GOLDFISH 413

several measures of performance were obtained: percentage of ini- of the experiment (Mann-Whitney Us > 8, all ps > .19),
tial correct choices, days to reach criterion, errors to reach crite- so these are collapsed in the group averages.
rion, time spent to get food, and error perseverations. The group The percentage of correct choices remained at or near
means for each measure were compared in nonparametric tests.
The following statistical tests were used to analyze between- and
chance level in the first few sessions of the experiment,
within-group differences: Kruskal-Wallis one-way analysis of and no significant between-group differences [Kruskal-
variance (ANOVA), Mann-Whitney U, Friedman two-way ANOVA, Wallis 2(3) = 4.23, p > .23] were found for the first
chi-square, and Wilcoxon matched-pairs signed-ranks test. three sessions. On subsequent sessions, the control group
remained at chance, whereas the experimental groups
RESULTS progressively increased their accuracy and showed sig-
nificant differences relative to the control group from
Acquisition, Overtraining, and Reversal Sessions 4 to 20 (Mann-Whitney: allo vs. control U =
The accuracy of each group during acquisition, over- 1.0, p < .01; ego vs. control U  3.0, p < .01; ego-allo
training, and the later reversal is shown in Figure 3. vs. control U  1.0, p < .01; allo = allocentric group,
There were no significant differences between the coun- ego = egocentric group, and ego-allo = ego-allocentric
terbalanced conditions within each group in any phase group). In addition, there were significant differences

Figure 4. Mean time to get food (A) and mean number of error perseverations (B) for each group during the
experiment.
414 RODRIGUEZ, DURAN, VARGAS, TORRES, AND SALAS

between experimental groups. Animals trained in the Following the overtraining sessions and the transfer/
egocentric procedure required more sessions of training probe trials (discussed below), the location of food was
(ego = 20; allo = 16; ego-allo = 14) and made signifi- reversed in each group. This change produced a dra-
cantly more errors (ego = 33.57
8.8; allo = 23.43
6.9; matic decrease in the level of correct choices for all ex-
ego-allo = 19.6
4.4) to reach criterion relative to the perimental groups, whereas performance in the control
other experimental groups (Mann-Whitney: ego vs. allo group was (of course) unmodified. Over the course of
U  12.8, p < .05; ego vs. ego-allo U  8.0, p < .05). reversal training, differences among the experimental
The difference between the ego-allocentric group and groups emerged as well (see Figure 3).
the allocentric group was not significant. At the beginning of the reversal phase, accuracy in the
During the overtraining phase, the experimental groups experimental groups dropped from a level of 83%–
maintained a steady and high level of accuracy in the so- 100% during overtraining to 4%–21%. Of course, these
lution of their tasks (allo, 87.8%
7%; ego, 91%
6.1%; differences were significant for the allocentric, egocen-
ego-allo, 96%
3.2%). No significant difference was tric, and ego-allocentric groups (Wilcoxon Zs > 2.36, all
found between groups during this phase [Kruskal-Wallis ps < .05). By contrast, the overtraining versus reversal
 2 (2) = 2.62, p > .27]. These three groups continued to difference in the control group was not significant
show significantly better performance than did controls (Wilcoxon Z = 1.06, p > .68). It was also noteworthy that
(Mann-Whitney Us < 6.0, all ps < .01), which remained the percentage of correct choices in each experimental
close to chance level (51.03%
4.5%). The control data group was significantly lower than that for the control
show that chemosensory cues, direct visual location of group (Mann-Whitney Us < 3.2, all ps < .01).
the food, or other uncontrolled variables were not re- The level of accuracy for the allocentric and ego-
sponsible for the accuracy level observed in the experi- allocentric groups increased quickly during reversal,
mental groups. whereas it remained very low for the egocentric group

Figure 5. Schematic representation of the trajectories chosen during the transfer tests for each group. Conditioning-
trial performance during overtraining is shown in the left part of the figure. Arrows indicate trajectories chosen from the
starting place; their relative thicknesses and numbers denote the percentages of times that the choice was made. Solid
lines represent the position of the maze for conditioning and transfer trials; dashed lines denote the normal position of
the maze before its displacement for Type 2 tests. The asterisks mark the theoretical goal if the animals of allocentric and
egocentric groups were using a place or orientation response, respectively, and if both responses were present for the ego-
allocentric group.
 ALLOCENTRIC AND EGOCENTRIC SPATIAL LEARNING IN GOLDFISH 415

Figure 6. Mean percentage of choices in Type 1 and Type 2 transfer tests and their average (total) for each group. Each
bar corresponds to one of three types of choices as described in the text.

throughout this period (Figure 3). After 1 week, statisti- and 1.41, respectively, ps > .17) and from the control
cally significant differences were found between the al- group (both Mann-Whitney Us > 24.4, p > .25).
locentric and ego-allocentric groups versus the egocen- Error perseverations diminished during acquisition and
tric group (Mann-Whitney Us = 2.2 and 3.1, respectively, overtraining in every group, reaching values close to
ps < .01). At the end of the reversal phase, the former zero (see Figure 4B). At the onset of the reversal, the al-
two experimental groups performed significantly more locentric and ego-allocentric groups greatly increased
accurately than the control group (Mann-Whitney Us = their error perseverations relative to those in the over-
9.07 and 6.9, respectively, ps < .05 and .01). Interest- training phase (both Wilcoxon Zs = 2.20, p < .05) and
ingly, the egocentric group did not show any significant relative to those in control and egocentric groups (Mann-
increase in accuracy relative to the level shown at the be- Whitney Us < 11.7, ps < .05). However, these error per-
ginning of this period (Wilcoxon Z  .89, p > .30). severations rapidly decreased to their prereversal levels,
During acquistion, all groups showed a decrease in eventually dropping to levels similar to those in the ego-
the time to get food relative to initial levels, and signif- centric and control groups [Kruskal-Wallis  2(3) = 4.17,
icant differences were established between this phase p > .24]. Error perseveration levels in the egocentric and
and subsequent overtraining (Wilcoxon Zs > 2.03, all control groups were similar during overtraining and re-
ps < .05; see Figure 4A). At the onset of reversal, a large versal periods (Wilcoxon Zs = .16 and .13, ps > .8).
increase in the time to get food was observed in allo-
centric and ego-allocentric groups in comparison with Transfer Trials
their performance in the previous phase (both Wilcoxon Figures 5 and 6 summarize the data from the two types
Zs = 2.36, p < .05), with the egocentric group (Mann- of transfer-test trials. In the control group, the choice
Whitney Us = 11.3 and 5.62, respectively, ps < .05 and frequency of the three arms was close to chance on both
.01) and with the control group (Mann-Whitney Us = Type 1 and Type 2 trials [both 2s(2) = .35, p > .83], in-
12.01 and 9.54, respectively, ps < .05). The time to get dicating a lack of a predominant preference for an arm
food in the allocentric and ego-allocentric groups de- or place (see Figures 5D and 6D). By contrast, in the
creased quickly, however, eliminating the significant egocentric and allocentric groups, the choice frequency
differences from prereversal levels (Wilcoxon Zs = 1.12 of the three arms was significantly different from that
416 RODRIGUEZ, DURAN, VARGAS, TORRES, AND SALAS

Figure 7. Percentage of correct choices in probe tests for each group when one salient cue was hidden or re-
moved from the room, and when all distal cues were removed by curtains surrounding the maze. The percent-
age of correct choices during conditioning is also shown for comparison.

expected by chance on both types of trials [ego, 2s(2) = in the same place as that rewarded during training (allo-
31.4 and 24.35, ps < .001, for Type 1 and Type 2, re- centric choice) on 45.7%
23.1% of the trials. In the
spectively; allo,  2s(2) = 48.05 and 24.35, ps < .001, for Type 2 test, the egocentric (65.71%
19.02%) and allo-
Type 1 and Type 2, respectively]. For the egocentric group, centric (31.95%
18.1%) choices again occurred more
the arm most often chosen was that coinciding with its often than the remaining choices. No significant differ-
learned turn independently of the location of the start- ence was found between egocentric versus allocentric
box in the experimental room (see Figures 5B and 6B). choices in the Type 1 test [ 2 (1) = .105, p > .74], whereas
In the Type 1 transfer test, the fish in this group showed there were more egocentric than allocentric choices in
a significant preference for the arm corresponding with the Type 2 test [ 2 (1) = 4.33, p < .05].
an egocentric response [76.5%
15.7%;  2 (1) = 10, p <
.01]. In the Type 2 transfer test, in which the maze was Probe Trials
displaced in the room and the fish started from an unfa- The probe-trial data are shown in Figure 7. When each
miliar place, the fish also showed a strong and signifi- of the salient cues in the room was individually removed
cant preference for the arm corresponding with the or hidden, performance was not significantly affected in
learned turn [73.45%
23%;  2 (1) = 6.4, p < .05]. By any group [Friedman  2s(5) = 4.5, 2.29, 3.11, and 6.73,
contrast, animals in the allocentric group consistently for allocentric, egocentric, ego-allocentric, and control
chose the arm whose extreme was at the same place of groups, respectively, all ps > .35]. However, when all
the room where the fish was rewarded during condi- cues were deleted by surrounding the maze with thick
tioning trials (85.7%
13% and 72%
19.6% for Type 1 curtains, the mean percentage of correct responses in
and Type 2 tests, respectively; see Figures 5A and 6A). the allocentric group significantly differed from perfor-
This tendency was significant in both tests [ 2 (1) = 19.6 mance during conditioning, decreasing to close to chance
for Type 1, p < .001;  2 (1) = 6.4 for Type 2, p < .05]. level (Wilcoxon Z  2.31, p < .05), whereas perfor-
During the transfer trials for the ego-allocentric group, mance in the other three groups remained unchanged
the positions of the maze and the startbox in the room (Wilcoxon Zs < 1.07, all ps > .30).
were changed so that a turn response was incompatible
with a place response. The choice frequency in this DISCUSSION
group was also significantly different from that expected
by chance on both types of trials [ 2 s(2) = 14.6 and The present experiment focused on whether goldfish
23.45 for Type 1 and Type 2, respectively, ps < .001]. In could learn to solve spatial tasks by employing allocen-
the Type 1 test (see Figures 5C and 6C), fish chose the tric frames of reference and, more generally, inquired
arm corresponding with the same turn they had made on whether parallel spatial learning and memory systems
rewarded training trials (an egocentric choice) on with the properties described by the cognitive mapping
51.4%
22.6% of these trials, and chose the arm situated theory of O’Keefe and Nadel (1978) might be present in
 ALLOCENTRIC AND EGOCENTRIC SPATIAL LEARNING IN GOLDFISH 417

fishes. These issues will be first discussed in view of the Therefore, it seems that the allocentric group indeed
transfer results. Afterward, other questions such as the used allocentric frames of reference to solve their task.
relative difficulty of each task, the use of distal visual In conjunction with previous data showing, for example,
cues, and the effect of reversal will be addressed. the ability of goldfish to maintain a constant environmen-
tal direction independently of the entrance point (Ingle
Use of Allocentric and Egocentric Strategies & Sahagian, 1973) and their use of landmarks as indi-
As Revealed by Transfer Tests rect reference points (Warburton, 1990), the data from
The most noteworthy results of this experiment come the present study indicate a cognitive mapping system
from the accurate performance of fishes trained in the like that proposed by O`Keefe and Nadel, for a lower
allocentric procedure and their behavior in the transfer vertebrate group (cf. Aronson, 1951, 1971; Reese, 1989;
trials. These fishes were able to reach the goal even Teyke, 1989). This cognitive mechanism could be rele-
when released from novel start arms (transfer test, Type 1) vant for fish for traveling in their natural habitat, espe-
and from unvisited locations of the room (transfer test, cially for species that are relatively site attached, such as
Type 2). These results suggest that animals in this group goldfish and common carp (Reynolds, 1983), or for those
made place responses by using allocentric frames of ref- that perform repeated migrations to different areas of their
erence. Their ability to spontaneously choose the ap- habitat (Dodson, 1988; Reese, 1989). Certainly, such a
propriate trajectory to the goal place from novel start lo- cognitive system would permit flexible and adaptive
cations and to use new routes without a history of previous spatial behaviors, such as those observed here.
training (see trajectories in Figure 5A) indicates the ca- However, it should be mentioned that there is a seri-
pacity to discriminate and represent spatial relationships ous controversy about the nature of spatial representa-
in the environment independently of a body-centered tion supporting place responses. In addition to the hy-
reference system. pothesis of allocentric spatial representations (Biegler &
By contrast, other strategies such as the use of guid- Morris, 1993; Nadel, 1991; Worden, 1992), several au-
ance (i.e., approaching or avoiding a particular cue; see thors propose that place responses might be explained
O’Keefe & Nadel, 1978) cannot explain the results ob- by mechanisms involving egocentrically referenced rep-
tained for the allocentric group, particularly those from resentations of landmarks. For example, it has been sug-
the transfer test, Type 2. If these animals used a guidance gested that animals can be guided by stored information
strategy, they should have moved toward a single distal about the retinal size of each of several landmarks in the
cue that served as a reference object. Instead, in the Type 2 room, as viewed from the goal (Leonard & McNaughton,
transfer test, they navigated preferentially in a reverse 1990; McNaughton, 1987). Another interesting hypoth-
direction to the place where they were reinforced during esis (Collet, Cartwright, & Smith, 1986) suggests a goal-
training. Of course, the possibility of a guidance strat- centered spatial memory, consisting of a set of vectors
egy was minimized procedurally by avoiding the use of describing the distance and direction from the goal to
salient cues associated with the extremes of the goal each landmark.
arms. The success of this procedure was illustrated by The performance of fishes in the egocentric group,
the results of the probe trials in which particular cues like that of those in the allocentric group, also showed a
were removed. On probe trials, the fish still chose cor- high level of accuracy despite a very different strategy.
rectly in the absence of any single salient distal cue. As is shown in Figure 5B, animals in this group mainly
On the other hand, several species of fish may use dif- choose the arm corresponding to the 90º turn made in
ferent compass senses to orient themselves (Quinn, training independently of the start point and of maze lo-
1980; Quinn & Brannon, 1982; Quinn, Merrill, & Bran- cation in the room. Thus, it seems that the choices of
non, 1981; Walker, 1984). Even so, the transfer data ob- these animals were purely egocentric, and that environ-
tained here indicate that the fishes were not solving the mental information was not taken into account for the
tasks by simply employing a “direction sense” (e.g., a arm selection. Following the terminology of O’Keefe
geomagnetic sense), because they navigated toward the and Nadel (1978), these animals employed a taxon sys-
goal from different directions. These results are not sur- tem and, more specifically, an orientation strategy.
prising, given that deliberate attempts to condition gold- Thus, the data from the two groups described above
fish to respond to a magnetic field have been unsuc- show that goldfish can use either egocentric or allocen-
cessful (Walker & Bitterman, 1986). tric strategies to solve spatial tasks. Moreover, the results
Finally, the accurate performance of the allocentric of the ego-allocentric group indicate that fish can also
group during training could have been realized through use both strategies simultaneously. During transfer, the
the use of two conditional response strategies: (1) when animals in the latter group chose the two arms corre-
beginning from one start arm, to turn to the right, and sponding with either egocentric or allocentric choices.
(2) when beginning from the other, to turn to the left This pattern was not the result of random processes, be-
(Blodgett & McCutchan, 1947; Ingle & Sahagian, 1973; cause the third arm corresponding to other strategies
Thinus-Blanc & Ingle, 1985). Again, however, the trans- was scarcely chosen. The use of both egocentric and al-
fer tests rule out this possibility as being necessary to locentric strategies could explain the tendency of this
performance (see Olton, 1979), because the fishes went group to perform more accurately and steadily than the
to the goal irrespective of their starting point. remaining groups. The cooperative use of different spa-
418 RODRIGUEZ, DURAN, VARGAS, TORRES, AND SALAS

tial strategies has been suggested for fishes (Roitblat, with extensive training, as reported previously for mam-
Tham, & Golub, 1982) and mammals (Schenk & Mor- mals (Hicks, 1964; Mackintosh, 1965; Restle, 1957),
ris, 1985; Whishaw, 1989; Whishaw & Mittleman, 1986) and could correspond to a pure orientation strategy in
and has been proposed for animals in their natural envi- which the animals solve the task by performing a par-
ronment (Able, 1991; Reese, 1989). Usually, nature does ticular turn within a body-centered reference system
not provide conflicting spatial information, so one might (O’Keefe & Nadel, 1978).
expect that the use of multiple behavioral systems would For the ego-allocentric group, the probe-trial data
increase navigational efficiency in an environment with again suggested the use of both a cognitive mapping and
redundant cues (Able, 1991). Finally, the combined use a taxon strategy. Animals in this group continued to
of egocentric and allocentric strategies agrees with the choose correctly, despite the absence of salient distal
proposal that “cognitive and noncognitive systems are cues. Apparently, with visual cues absent, these animals
not mutually exclusive, since they can act in concert” relied on the taxon system. Others have reported that for
(O’Keefe & Nadel, 1978, p. 520). subjects using simultaneous strategies to solve a spatial
As mentioned before, the experimental groups could task, if one strategy is blocked, the animal still performs
use egocentric and/or allocentric strategies to solve spa- correctly on the basis of the remaining one (O’Keefe &
tial tasks. The differences in number of errors and days Nadel, 1978; Quinn, 1980; Quinn & Brannon, 1982;
to criterion in acquisition indicate that the task requiring Schenk & Morris, 1985; R. J. Sutherland & Rudy, 1989;
an egocentric solution was more difficult to learn. Sim- Whishaw, 1989).
ilar results have been described in mammals. For exam-
ple, rodents trained to turn in one direction in a maze in Response to Reversal for Each Experimental Group
which the environmental and intramaze cues were irrel- The reacquisition data following reversal revealed other
evant were slower to learn the task than those trained in between-group differences related to the spatial strategy
a task in which a particular place was rewarded (Hill & in each group. For example, it is consistent with the test
Thune, 1952; Scharlock, 1955; Tolman, Ritchie, & Kalish, data suggesting that the egocentric group used an orien-
1946). These results have been explained by the dis- tation strategy that fishes in this group persistently chose
turbing effect of extramaze cues in the response group the arm coinciding with the same turn after which they
(Blodgett & McCutchan, 1947) or by assuming that rats were previously rewarded.
use a place strategy during initial learning and only shift The lack of flexibility shown by fish using egocentric
to a response strategy with continued training (Hicks, strategies notably contrasts with the capacity of the al-
1964; Means & Douglas, 1970; O’Keefe & Nadel, 1978, locentric groups to quickly modify their behavior during
1979; N. S. Sutherland & Mackintosh, 1971). This ac- reversal. Thus, whereas the fishes in the egocentric group
count could also explain why the ego-allocentric and al- only reached the new goal after visiting the unrewarded
locentric groups learned most rapidly in the present study. arm (formerly the rewarded arm), the fishes in the allo-
centric and ego-allocentric groups quickly learned to
Relevance of Visual Cues to Solving move toward the new goal by using the shortest trajec-
Allocentric Procedure tory. The reversal performances of the latter two groups
Probe trials in which the most salient visual cues were reveal a capacity for rapidly detecting environmental
removed or hidden also provided interesting evidence changes (Poucet, Chapuis, Durup, & Thinus-Blanc, 1986;
about the characteristics of the spatial strategies. No ef- Thinus-Blanc et al., 1987; Welker & Welker, 1958).
fects were observed when any single cue was taken Warburton (1990) reported similar reversal learning dif-
away, but when all cues were removed, the performance ferences in goldfish trained in an arena tank to find re-
of the allocentric group deteriorated, becoming as poor ward in either a directly or an indirectly cued location.
as that of the control group. These results suggest that
the behavior of the allocentric group was controlled by CONCLUDING REMARKS
a combination of many environmental cues, in agree-
ment with the cognitive mapping system described by The present findings indicate that goldfish are able to
O’Keefe and Nadel’s (1978) theory. According to this establish novel routes toward a goal even from unfamil-
theory, such a system defines a place by its spatial rela- iar start points, to use distal visual cues as a whole to
tionships to a number of landmarks, none of them being navigate accurately, and to reorganize their spatial strate-
essential by itself. Similar evidence consistent with this gies during reversal. These data suggest that goldfish
cognitive mapping view has been shown by altering the build complex spatial cognitive representations of their
extramaze stimuli (Mazmanian & Roberts, 1983; Morris, environment, which permit flexibility in their spatial be-
1981; O’Keefe & Conway, 1978; Suzuki et al., 1980). havior. The data also suggest that an allocentric cogni-
On the other hand, choices of the egocentric group tive system works in parallel with, and can cooperate
were not affected by removing any one cue or all of them, with, egocentric systems in such a way that goldfish can
suggesting that the behavior of these animals was use the most profitable strategy in each situation. These
scarcely controlled by environmental stimuli. Rather, the findings are in agreement with the current notion of
animals apparently performed on the basis of behavioral multiple parallel learning systems and support the gen-
stereotypes. Such stereotypes were probably acquired eralization of the spatial learning and memory theory of
 ALLOCENTRIC AND EGOCENTRIC SPATIAL LEARNING IN GOLDFISH 419

O’Keefe and Nadel to apply to a lower vertebrate group. McNaughton, B. L. (1987). Neural associations of movements and
The possible presence of a cognitive mapping system in space: Preliminary steps toward a non-cartographic theory of spa-
tial representation and learning. Neuroscience Letters, 29, S143-
fishes encourages the study of its comparative peculiar- S144.
ities and neural substratum and encourages the notion Means, L. W., & Douglas, R. J. (1970). Effects of hippocampal le-
that complex spatial cognitive systems appeared very sions on cue utilization in spatial discrimination in rats. Journal of
early in the evolutionary history of vertebrates. Comparative & Physiological Psychology, 73, 254-260.
Morris, R. G. M. (1981). Spatial localization does not require the pres-
ence of local cues. Learning & Motivation, 12, 239-260.
REFERENCES Nadel, L. (1991). The hippocampus and space revisited. Hippo-
campus, 1, 221-229.
Able, K. P. (1991). Common themes and variations in animal orien- O’Keefe, J., & Conway, D. H. (1978). Hippocampal place units in the
tation systems. American Zoologist, 31, 157-167. freely moving rat: Why they fire where they fire. Experimental Brain
Aronson, L. R. (1951, January 17). Orientation and jumping behav- Research, 31, 573-590.
ior in the gobiid fish Bathygobius soporator. American Museum O’Keefe, J., & Nadel, L. (1978). The hippocampus as a cognitive
Novitates (No. 1486), pp. 1-22. map. Oxford: Oxford University Press, Clarendon Press.
Aronson, L. R. (1971). Further studies on orientation and jumping be- O’Keefe, J., & Nadel, L. (1979). Précis of O’Keefe & Nadel’s The
havior in the gobiid fish Bathygobius soporator. In H. E. Adler hippocampus as a cognitive map. Brain & Behavioral Sciences, 2,
(Ed.), Orientation: Sensory bases (Annals of the New York Acad- 487-533.
emy of Sciences, Vol. 188, pp. 378-392). New York: New York Olton, D. S. (1979). Mazes, maps and memory. American Psycholo-
Academy of Sciences. gist, 34, 583-596.
Biegler, R., & Morris, R. G. M. (1993). Landmark stability is a pre- Pitcher, T. J., & Magurran, A. E. (1983). Shoal size, patch prof-
requisite for spatial but not discrimination learning. Nature, 361, itability and information exchange in foraging goldfish. Animal Be-
631-633. haviour, 31, 546-555.
Bingman, V. P. (1990). Spatial navigation in birds. In R. Kesner, & Poucet, B., Chapuis, N., Durup, M., & Thinus-Blanc, C. (1986). A
D. S. Olton (Eds), Neurobiology of comparative cognition (pp. 423- study of exploratory behavior as an index of spatial knowledge in
477). Hillsdale, NJ: Erlbaum. hamsters. Animal Learning & Behavior, 14, 93-100.
Bingman, V. P., Bagnoli, P., Ioalé, P., & Casini, G. (1989). Behav- Quinn, T. P. (1980). Evidence for celestial and magnetic compass ori-
ioral and anatomical studies of the avian hippocampus. In V. Chan- entation in lake migrating sockeye salmon fry. Journal of Compar-
Palay & C. Köhler (Eds), The hippocampus: New vistas (pp. 379- ative Physiology A, 137, 243-248.
394). New York: Alan R. Liss. Quinn, T. P., & Brannon, E. L. (1982). The use of celestial and mag-
Blodgett, H. C., & McCutchan, K. (1947). Place versus response netic cues by orienting sockeye salmon smolts. Journal of Com-
learning in the simple T-maze. Journal of Experimental Psychology, parative Physiology A, 147, 547-552.
37, 412-422. Quinn, T. P., Merrill, R. T., & Brannon, E. L. (1981). Magnetic field
Collet, T. S., Cartwright, B. A., & Smith, B. A. (1986). Landmark detection in sockeye salmon. Journal of Experimental Zoology, 217,
learning and visuo-spatial memories in gerbils. Journal of Com- 137-142.
parative Physiology A, 158, 835-851. Reese, E. S. (1989). Orientation behavior of butterflyfishes (family
Dodson, J. J. (1988). The nature and role of learning in the orienta- Chaetodontidae) on coral reefs: Spatial learning of route specific
tion and migratory behavior of fishes. Environmental Biology of landmarks and cognitive maps. Environmental Biology of Fishes,
Fishes, 23, 161-182. 25, 79-86.
Hallacher, L. E. (1984). Relocation of original territories by dis- Restle, F. (1957). Discrimination of cues in mazes: A resolution of
placed black-and-yellow rockfish, Sebastes chrysomelas, from the “place-vs-response” question. Psychological Review, 64, 217-
Carmel Bay, California. Californian Fish & Game, 7, 158-162. 228.
Helfman, G. S., Meyer, J. L., & McFarland, W. N. (1982). The on- Reynolds, L. F. (1983). Migrations patterns of five fish species in the
togeny of twilight migration patterns in grunts (pisces: Haemuli- Murray-Darling river system. Australian Journal of Marine &
dae). Animal Behaviour, 30, 317-326. Freshwater Research, 34, 857-871.
Helfman, G. S., & Schultz, E. T. (1984). Social transmission of be- Roitblat, H. L., Tham, W., & Golub, L. (1982). Performance of Betta
havioral traditions in a coral reef fish. Animal Behaviour, 32, 379- splendens in a radial arm maze. Animal Learning & Behavior, 10,
384. 108-114.
Hicks, L. H. (1964). Effects of overtraining on acquisition and reversal Scharlock, D. P. (1955). The role of extramaze cues in place and re-
of place and response learning. Psychological Reports, 15, 459-462. sponse learning. Journal of Experimental Psychology, 50, 249-254.
Hill, C. W., & Thune, L. E. (1952). Place and response learning in Schenk, F., & Morris, R. G. M. (1985). Dissociation between com-
the white rat under simplified and mutually isolated conditions. ponents of spatial memory in rats after recovery from the effects of the
Journal of Experimental Psychology, 43, 289-297. retrohippocampal lesions. Experimental Brain Research, 58, 11-28.
Ingle, D., & Sahagian, D. (1973). Solution of a spatial constancy Sutherland, N. S., & Mackintosh, N. J. (1971). Mechanisms of an-
problem by goldfish. Physiological Psychology, 1, 83-84. imal discrimination learning. New York: Academic Press.
Kleerekoper, H., Matis, J., Gensler, P., & Maynard, P. (1974). Ex- Sutherland, R. J., & Rudy, J. W. (1989). Configural association the-
ploratory behaviour of goldfish Carassius auratus. Animal Behav- ory: The role of the hippocampal formation in learning, memory,
iour, 22, 124-132. and amnesia. Psychobiology, 17, 129-144.
Leonard, B. J., & McNaughton, B. L. (1990). Spatial representation Suzuki, S., Augerinos, G., & Black, A. H. (1980). Stimulus control
in the rat: Conceptual, behavioral and neurophysiological perspec- of spatial behavior on the eight-arm maze in rats. Learning & Mo-
tives. In R. P. Kesner & D. S. Olton (Eds), Neurobiology of com- tivation, 11, 1-18.
parative cognition (pp. 363-422). Hillsdale, NJ: Erlbaum. Teyke, T. (1989). Learning and remembering the environment in the
Mackintosh, N. J. (1965). Overtraining, transfer to proprioceptive blind cave fish Anoptichthys jordani. Journal of Comparative Phys-
control, and position reversal. Quarterly Journal of Experimental iology A, 164, 655-662.
Psychology, 17, 26-36. Thinus-Blanc, C., Bouzouba, L., Chaix, K., Chapuis, N.,
Markevich, A. I. (1988). Nature of territories and homing in the east- Durup, M., & Poucet, B. (1987). A study of spatial parameters en-
ern sea-perch Sebastes taczanowski. Journal of Ichthyology, 28, coded during exploration in hamsters. Journal of Experimental Psy-
161-163. chology: Animal Behavior Processes, 13, 418-427.
Mazmanian, D. S., & Roberts, W. A. (1983). Spatial memory in rats Thinus-Blanc, C., & Ingle, D. (1985). Spatial behavior in gerbils
under restricted viewing conditions. Learning & Motivation, 14, (Meriones unguiculatus). Journal of Comparative Psychology, 99,
123-139. 311-315.
420 RODRIGUEZ, DURAN, VARGAS, TORRES, AND SALAS

Tolman, E. C. (1948). Cognitive maps in rats and men. Psychological gula) to environmental changes. Ecology, 39, 283-288.
Review, 55, 189-208. Whishaw, I. Q. (1989). Dissociating performance and learning defi-
Tolman, E. C., Ritchie, B. F., & Kalish, D. (1946). Studies in spatial cits on spatial navigation tasks in rats subjected to cholinergic mus-
learning: II. Place learning versus response learning. Journal of Ex- carinic blockade. Brain Research Bulletin, 23, 347-358.
perimental Psychology, 3, 221-229. Whishaw, I. Q., & Mittleman, G. (1986). Visits to starts, routes and
Walker, M. M. (1984). Learned magnetic field discrimination in yel- places by rats (Rattus norvegicus) in swimming pool navigation
lowfin tuna, Thunnus albacares. Journal of Comparative Physiol- tasks. Journal of Comparative Psychology, 100, 422-431.
ogy A, 155, 673-679. Worden, R. (1992). Navigation by fragment fitting: A theory of
Walker, M. M., & Bitterman, M. E. (1986). Attempts to train goldfish hippocampal function. Hippocampus, 2, 165-188.
to respond to magnetic field stimuli. Naturwissenschaften, 73, 12-16.
Warburton, K. (1990). The use of local landmarks by foraging gold-
fish. Animal Behaviour, 40, 500-505.
Welker, W. I., & Welker, J. (1958). Reaction of fish (Eucinostomus