Anim Cogn (2006) 9: 110117

DOI 10.1007/s10071-005-0007-2
ORIGINAL ARTICLE
Annika Paukner James R. Anderson Kazuo Fujita
Redundant food searches by capuchin monkeys (Cebus apella):
a failure of metacognition?
Received: 15 March 2005 / Revised: 3 July 2005 / Accepted: 5 July 2005 / Published online: 24 September 2005
C
Springer-Verlag 2005
Abstract This study investigated capuchin monkeys un-
derstanding of their own visual search behavior as a means
to gather information. Five monkeys were presented with
three tubes that could be visually searched to determine
the location of a bait. The baits visibility was experimen-
tally manipulated, and the monkeys spontaneous visual
searches before tube selection were analyzed. In Exper-
iment 1, three monkeys selected the baited tube signi-
cantly above chance; however, the monkeys also searched
transparent tubes. In Experiment 2, a bent tube in which
food was never visible was introduced. When the bent tube
was baited, the monkeys failed to deduce the bait location
and responded randomly. They also continued to look into
the bent tube despite not gaining any pertinent information
fromit. The capuchin monkeys behavior contrasts with the
efcient employment of visual search behavior reported in
humans, apes and macaques. This difference is consistent
with species-related variations in metacognitive abilities,
although other explanations are also possible.
Keywords Metacognition
.
Metacognitive awareness
.
Capuchin monkey
.
Visual search behavior
When humans are aware that they are missing important
information for correctly completing a task, or when they
are unsure about what to do, they often react predictably.
For example, if they realize that a piece of information is
not sufciently known to pass an exam, students may de-
cide to study further (Nelson and Narens 1990). Seeking
information or declining a response in a situation of uncer-
tainty are indicative of metacognitive awareness, which is
A. Paukner () J. R. Anderson
Department of Psychology, University of Stirling,
Stirling, FK9 4LA, UK
e-mail: [email protected]
Tel.: +44-1786-467640
Fax: +44-1786-467641
K. Fujita
Department of Psychology, Faculty/Graduate School of Letters,
Kyoto University,
Yoshida-honmachi, Sakyo, Kyoto 606-8501, Japan
dened as knowing about ones own cognitive states (Smith
et al. 1995). Metacognitive awareness has been investigated
in several non-human species, including dolphins, pigeons
and a fewspecies of primates (e.g. Smith et al. 1995; Inman
and Shettleworth 1999; Hampton 2001; for a review, see
Smith et al. 2003). For example, Hampton (2001) found that
rhesus macaques (Macaca mulatta) may decline a memory
trial that they are unlikely to complete correctly and opt
for a new, easier trial instead. This behavior suggests that
rhesus macaques are capable of accurately monitoring their
own memory contents (Hampton 2001).
Grifn (2003) argued that all animals face uncertainty, for
example, when deciding whether a movement in the veg-
etation is a predator or simply wind, or whether an object
on the ground is edible or not. Decisions regarding these
uncertainties can be crucially important for survival of an
animal. Therefore, a system for accurately monitoring un-
certainties and controlling the animals behavior in terms of
either seeking more information or deciding on the point of
certainty sufcient for making an adaptive response would
appear advantageous. It is therefore conceivable that natu-
ral selection has led to metacognitive awareness evolving
in many species.
Regarding primates, chimpanzees (Pan troglodytes)
and orangutans (Pongo pygmaeus) will efciently seek
additional information. Call and Carpenter (2001) showed
hollow tubes to these apes, one containing a food reward,
and the apes were allowed to look into the tubes before
selecting one. They received the reward only if they
selected the baited tube. Call and Carpenter (2001) found
that both species were more likely to search for the reward
when they had not witnessed the baiting procedure and thus
did not know where the bait was. In addition, searching
the tubes led to more correct tube selections. Apes behave
similarly to 2.5-year-old children in this respect; the latter
also engage in visual searches when they do not know the
correct location of a desirable objects (Call and Carpenter
2001). Recently, Hampton et al. (2004) adapted the tube
paradigm to test rhesus macaques. Like humans and apes,
the macaques searched more when they had not witnessed
the baiting, and selection accuracy increased when they
111
searched. These ndings suggest that metacognitive
awareness might have evolved in the common ancestors of
apes and the Old World monkeys, and could be widespread
amongst primates (Hampton et al. 2004).
In the present study, we used the tube paradigm to in-
vestigate whether tufted capuchin monkeys (Cebus apella)
efciently seek additional information in a situation of
uncertainty. We chose capuchin monkeys as subjects be-
cause they have been ascribed advanced cognitive abilities
such as visible displacements (Dumas and Brunet 1994)
and seriation (McGonigle et al. 2003), but their metacog-
nitive abilities have not yet been assessed. Furthermore,
if similar metacognitive abilities to rhesus macaques and
chimpanzees are found in this New World species, then
metacognitive awareness may have evolved in the common
ancestors of the Old World and New World monkeys. If ca-
puchin monkeys possess similar metacognitive awareness,
they should behave similarly to apes and rhesus macaques.
In particular, we would expect that visual searches are re-
stricted to trials in which they do not know the location of a
reward, and that visual searches result in increased selection
accuracy. In Experiment 1, this hypothesis was tested by
giving capuchin monkeys a tube-based visual search task.
The location of a food reward was either directly apparent,
or could be discovered through searching. The monkeys
search patterns were analyzed with regard to the frequency,
location and success of searches. In Experiment 2, we fur-
ther tested the monkeys awareness of their own search
behavior by presenting them with a bent tube, in which the
bait could never be seen. We reasoned that if the monkeys
were monitoring the outcome of their own search behav-
iors, they might realize that the bent tube never provided
any useful information regarding the location of the bait, so
that if the other (straight) tubes were empty, the bait could
only be within the bent tube. Therefore, they might select
the bent tube without searching it rst.
Experiment 1
Methods
Subjects and housing
The study was carried out at the Faculty of Letters, Kyoto
University, in 2003. Subjects were one adult male (Heiji,
9year-old), three adult female (Kiki, 7year-old; Theta,
7-year-old; Zilla, 9-year-old) and one juvenile male
(Zinnia, 2-year-old) tufted capuchin monkeys (Cebus
apella), all captive born and raised by their biological
mothers. Heiji, Zilla and Zinnia were housed together with
a young adult male; Kiki and Theta were housed as a pair.
Three monkeys (Heiji, Zilla and Theta) had previously
participated in object discrimination tasks, and had been
trained to use a pointing/reaching gesture to indicate their
selections. Zilla was pregnant throughout the experimental
period and gave birth 4 weeks after completion of the
experiment. The monkeys were not deprived of food, but
received part of their daily food rations during testing.
Apparatus
All monkeys were tested individually in a familiar testing
cage made of transparent acrylic board (46 cm 46 cm
52 cm). At the front of the cage was an opening (23.5 cm
3.5 cm), ca. 5 cm above the oor. This allowed the
monkeys to reach out toward the three plastic tubes
(40 cm 5 cm diameter) that were fastened on a platform
and held in position by wooden rails spaced 8 cm apart and
perpendicular to the test cage. During testing, the tubes
were raised ca. 3 cm at the end furthest from the monkey,
so that it was only possible to look through one tube at a
time from any location within the cage. The platform was
situated on a small table (59.5 cm45.5 cm30 cm) that
was in level with the oor of the cage. A white cardboard
screen was placed between the apparatus and the test cage
when required to occlude the baiting process.
Preliminary training
Monkeys were given three to eight preliminary training
sessions to familiarize them with the testing procedure and
to establish a reaching response toward the tubes. During
training, the openings of the tubes facing the monkey were
each blocked with a small piece of white cardboard, so that
the monkey would not learn to look into the tubes. The
experimenter sat behind the apparatus facing the monkey
with the screen occluding both the experimenter and ap-
paratus. In order to minimize unintentional cueing of the
correct response, the experimenter remained passive and
visually focused on the opening of the test cage throughout
the experiment.
Each trial started with the screen being removed, and
then a small piece of apple or sweet potato (the bait) being
visibly put on top of one of the three opaque tubes. If the
monkey reached toward the baited tube, it received the bait
and the trial was ended by replacing the screen. If the mon-
key reached toward a nonbaited tube, no reward was given
and the screen was replaced immediately. Reaching was
simply dened as an extension of the arm toward a tube as
if the monkey was trying to touch the tube. Occasionally,
the monkeys would tap the table in front of a tube, which
was also considered an acceptable selection response. In
the second training phase, several false baiting gestures
were added so that all tubes were touched once in random
order but only one tube was baited. For the nal train-
ing phase, the experimenter placed the bait on top of one
tube before removing the screen, with the platform situated
35 cm in front of the test cage. After removing the screen,
the platform was pushed ca. 5 cm toward but was still out
of reach of the monkey. The rst reach after the platform
came to a halt was taken as the response. Training sessions
were administered once a day and lasted ca. 20 min per
monkey.
Once the monkeys were responding reliably, a more for-
mal evaluation was conducted to make sure they understood
the task. Each test session consisted of 24 trials. In 16 trials,
the baiting was visible, i.e., the screen was removed before
112
baiting so that the monkeys could see the baiting gesture.
The remaining eight trials were unseen, i.e. the bait was
placed on top of one tube before the screen was removed
so that the monkeys did not see the baiting gesture. Each
monkey was judged to have reached the criterion when it
responded correctly to a total of at least 20 trials per session
(over 80% correct). Kiki and Zilla required three prelimi-
nary training sessions to reach the criterion. Heiji and Theta
required four and Zinnia required eight sessions.
Before the actual experiment, the experimenter removed
the cardboard that blocked visual access to the inside of
the tubes and showed the tubes to the monkeys. All mon-
keys showed interest in the tubes, and looked through them
several times.
Preliminary experiment
In a pilot study, we adopted a procedure identical to those
used by Call and Carpenter (2001) and Hampton et al.
(2004). We placed the bait inside a tube so that it was not
directly visible. We then presented the monkeys with two
different types of trials: seen trials, in which the monkeys
saw which tube the experimenter baited, and unseen trials,
in which the monkeys did not observe the baiting. How-
ever, we encountered problems with this procedure. In seen
trials, the monkeys would immediately visually follow the
food reward during baiting and look into the baited tube,
but even when the experimenter had deposited the reward
and, with a clearly empty hand, touched the top of other
(empty) tubes, the monkeys would look into whichever tube
was being touched. It is not clear why the experimenters
hand gestures may have been too salient to the capuchin
monkeys, resulting in failure to visually disengage from
them. Looking into a tube was therefore inuenced by the
experimenters baiting gesture and not an unequivocal in-
dication of metacognitive abilities, and became a confound
to the experimental paradigm.
To provide the monkeys with explicit information about
the bait location without using a baiting gesture, we always
baited a tube while the screen was between the monkey and
apparatus. However, we employed two different types of
tubes. One type (as used during training and the preliminary
experiment) was green-opaque and required the monkey to
crouch down to visually check the contents of the tube.
The second type of tube was transparent which meant that
the monkey could see directly if it was baited. This setup
allowed the monkeys to know the location of the bait when
a transparent tube was baited, and to be uncertain of the bait
location when an opaque tube was baited. We consider this
setup to be functionally equivalent to the paradigm using
seen and unseen baiting gestures in Call and Carpenter
(2001) and Hampton et al. (2004).
Experimental procedure
At the start of each trial, the experimenter sat behind
the apparatus, facing the monkey with the screen in
place occluding the experimenter and apparatus. The
experimenter touched all three tubes in random order to
eliminate any sound or movement cues. The bait was
placed inside one of the tubes, at the end farthest from the
monkey. The screen was then removed and the platform
was pushed toward the monkey after a 5 s delay. The rst
reach after the platform came to a halt was taken as the
response. If the monkey reached toward the baited tube,
the experimenter retrieved the bait, gave it to the monkey
and replaced the screen. If the monkey reached toward
an unbaited tube, the screen was replaced immediately. If
no response occurred after 60 s, the screen was replaced,
the bait was retrieved from the tube and the next trial
started. Trials were separated by a 5 s inter-trial interval. A
video camera placed behind the experimenter recorded all
sessions.
Each monkey received a total of 72 trials in three ses-
sions of 24 trials each, which were conducted once a day
for three consecutive days. This relatively small number
of trials was intended to prevent an improvement in per-
formance based on learned associations between search
behaviors and food rewards. Each monkey received 54 tri-
als in which two opaque tubes and one transparent tube
were used, with the location of the transparent tube and the
bait counterbalanced across trials. When the transparent
tube was baited, the monkeys could directly see the bait,
so that (if aware of this knowledge) no searching was re-
quired before choosing correctly. On the other hand, when
one of the opaque tubes was baited, the monkey could
immediately see that the transparent tube was empty, and
(if aware of this knowledge) could restrict its searches to
the opaque tubes. On the remaining 18 trials, one opaque
and two transparent tubes were used, with bait and tube
locations again counterbalanced. If a transparent tube was
baited in these trials, the monkey again could see the bait
immediately, so no searches were necessary. If the opaque
tube was baited, the monkey could see that the other two
tubes were empty. Therefore, the monkey might deduce
the location of the bait, and choose correctly without any
need for searching. The order of trial types was random-
ized with the only restriction that the bait was not hid-
den at the same location for more than two consecutive
trials.
Data analysis
All sessions were videotaped for later analysis. Baiting ges-
tures were not included in the footage, which allowed blind
scoring of the monkeys behaviors. The number of looks,
the tubes looked into and the order in which the tubes were
looked into were recorded from the moment the screen was
removed until selection occurred. Forty percent of trials
were coded a second time to assess intra-observer relia-
bility; consistency between codings was high (Pearsons
correlations: r=+0.92, p<0.001 for searches; r=+0.96,
p<0.001 for selection). Due to the small number of sub-
jects, results were analyzed separately for each individual
using nonparametric tests.
113
Results
Selection behavior
All monkeys selected a tube on all trials, but only three
monkeys (Zilla, Heiji and Kiki) chose the correct tube
signicantly above chance throughout the experiment
(chance=33%, binomial tests: all p<0.001). The remain-
ing two monkeys showed strong location biases: Zinnia
preferentially selected the middle tube (61 selections out
of 72) whilst Theta avoided the middle tube (7 selections
out of 72). However, although Thetas overall choices ap-
peared random (29 correct out of 72), she chose the correct
transparent tubes signicantly above chance (15 correct out
of 27, binomial test: p=0.011). Since only three monkeys
completed the overall task successfully, only these mon-
keys data are considered in detail.
The visibility of the bait did not inuence the outcome of
a trial. All the three monkeys were equally successful when
the bait was placed inside a transparent tube (100, 88.9 and
100% correct, respectively for Zilla, Heiji and Kiki) and
when it was placed inside an opaque tube (95.6, 95.6 and
97.8% correct, respectively).
Search behavior
Zilla, Heiji and Kiki searched on the rst trial, and contin-
ued to search at least one tube during all subsequent trials.
Typically, the monkeys searched until they saw the bait
and then either made a selection (47.2, 50 and 65.3% of
trials), or they continued to search but returned to search
in the baited tube with their last look before making a
selection (12.5, 18.1 and 16.7% of trials, respectively)
(Fig. 1).
Figure 2 shows the average number of looks per tube
per trial according to bait location and tubes searched. The
monkeys did not make fewer searches when the bait was
directly visible compared to trials when it was not directly
visible (MannWhitney tests for each individual: Heiji:
z=1.578; Kiki: z=0.56; Zilla: z=1.127, all p>0.05).
When comparing the number of looks into the transpar-
ent tubes and the opaque tubes during each trial (taking
into account the unequal number of tubes), no monkey
looked less into the former (68, 61 and 66 looks in to-
tal) than into the latter (80, 70 and 69 looks in total, re-
spectively; Wilcoxon Signed Rank tests: Heiji: z=1.547,
Kiki: z=1.652, Zilla: z=0.520, all p>0.05). Selection ac-
curacy was not associated with the increased number of
looks (correct selections: mean number of looks = 3.0,
2.7 and 2.8 for Heiji, Kiki and Zilla, respectively; incor-
rect selections: mean number of looks = 3.0, 2.0 and 3.6,
respectively; MannWhitney tests: z=0.077, z=0.808
and z=1.842, respectively, all p>0.05). In trials with two
empty transparent tubes and a baited opaque tube, monkeys
never selected the opaque tube without searching in at least
one tube rst.
Fig. 1 Setup of experiment during baiting. E: experimenter, M:
monkey
Fig. 2 Experiment 1average number of looks per tube per trial
according to baiting conditions. TR-VI: looks into transparent tubes
with bait in transparent tube; OP-VI: looks into opaque tubes with bait
in transparent tube; TR-IN: looks into transparent tube with bait in
opaque tube; OP-IN: looks into opaque tubes with bait in opaque tube
Discussion
Although all ve monkeys were successful during training,
only three monkeys selected the correct tube signicantly
above chance during the experiment. Seeing the bait
directly did not reduce the number of searches, and when
the bait location was unknown, more searching did not
lead to greater success. These results contrast with those
of apes (Call and Carpenter 2001) and rhesus macaques
(Hampton et al. 2004). In both these studies, subjects
searched less when they had direct information about the
bait location, and in trials without such information, they
were more successful when they searched.
114
The lack of association between searching and correct se-
lections by the three capuchin monkeys could be attributed
to the selection of the correct tube at ceiling levels, i.e.
there were too few incorrect trials to analyze. The two
monkeys that failed to reliably select the correct tube also
did not search reliably, again making it impossible to match
searching behavior to successful outcomes.
Another behavioral difference between the apes and rhe-
sus macaques on one hand and capuchin monkeys on the
other emerged when direct information about the bait lo-
cation was available. In trials in which an opaque tube was
baited, we expected the monkeys to search in the opaque
tubes. In trials in which a transparent tube was baited, we
still expected the monkeys to search opaque tubes, per-
haps to check if a second bait was available in one of
the tubes. Both these search strategies are consistent with
metacognitive awareness, i.e. the monkeys might be aware
of what information they have and what information they
are lacking at the beginning of a trial. However, it is not
clear why the monkeys would search transparent tubes if
they can directly see the contents of these tubes, and there-
fore should already be aware of the relevant information.
In contrast, both apes and rhesus macaques searched less
in seen trials when direct information about the bait loca-
tion was available. A simple explanation for the capuchin
monkeys unnecessary searches could be that they failed
to notice the bait inside the transparent tubes. However,
one monkey reliably selected the correct tube only if it
was transparent, which makes this explanation unlikely.
Searching in transparent tubes suggests that the capuchin
monkeys failed to effectively monitor the outcome of their
own visual search behavior, which implies a limitation of
metacognitive awareness.
One important aspect of the experimental setup used for
testing capuchin monkeys, however, may have signicantly
impacted on the results of the present experiment. Whereas
rhesus macaques and chimpanzees both encountered visi-
ble baiting gestures, the capuchins received the additional
information of the baits location through the use of trans-
parent tubes. In order to successfully complete the task,
capuchin monkeys therefore needed to understand the con-
cept of transparency, a requirement not posed by the visible
baiting gesture. Even human infants encounter difculties
with understanding transparency and object relations that
involve placement of objects into other objects. For ex-
ample, infants who pass visible displacement tasks may
nonetheless fail invisible displacement tasks that are con-
ducted using transparent cups that clearly show the ob-
ject to the infant (Bower 1982). One possible explanation,
therefore, is that the capuchin monkeys failed to appre-
ciate that the bait inside the tube was the same bait that
was visible when looking through the tube. Even to the
single monkey who in the absence of search behaviors
chose the correct transparent tube, the bait may have been
no more than a discriminative marker without an under-
standing that it was actually the same bait. Therefore, the
excessive search behavior of the capuchins may be caused
by a failure to understand the nature of the transparent
tubes.
Excessive search behavior on every trial led to an absence
of what Call and Carpenter (2001) termed super-efcient
searches, that is, selecting a tube by inference based on in-
direct knowledge gained about the bait location rather than
seeing the bait (either directly or through searching). Apart
from the metacognitive skills (monitoring current knowl-
edge states), this behavior seems to require other advanced
cognitive abilities such as making logical deductions based
on inferential steps (if the bait is not in the rst or sec-
ond tube, it can only be in the third tube). Super-efcient
searches also require an understanding of the limitations
of visual search behavior (for example, that a bait might
be in a tube even though visual access into this tube is
prevented). Call and Carpenter (2001) found that both 2.5-
year-old children and chimpanzees engaged in these super-
efcient searches, although they did so only infrequently
and preferred to locate the bait by searching instead. In
Experiment 2, we devised a new procedure to test whether
capuchin monkeys can infer the bait location, and to fur-
ther investigate their awareness of the consequences of their
own visual search behavior.
Experiment 2
In Experiment 2, one opaque tube with the end bent at 90

was introduced. When this tube was baited, the bait could
not been seen by looking into the tube. Hence unlike the
straight opaque tubes in Experiment 1, this bent tube of-
fered no information about the bait location. In addition, the
monkeys could easily identify this tube before searching.
This setup therefore provided the opportunity to investi-
gate how much capuchin monkeys understood about the
purpose and limitations of their own search behavior. Our
rst question was whether the capuchin monkeys would
infer the bait location within the bent tube and select it if
they could not see the bait in any other tubes. The second
question was whether they could do so without searching
the bent tube rst, thereby showing that they were aware of
the limitations of their own visual search behavior.
Methods
The same ve monkeys from the previous experiment were
tested, and the same basic procedure was used. The appa-
ratus consisted of one 90

horizontally bent opaque tube

(42 cm 5 cm) and two straight opaque tubes. The loca-
tions of the bent tube and the bait were randomized across
trials, with the only restriction that the bait was not in the
same location for more than two consecutive trials. Mon-
keys received one test session of 32 trials. On 12 of those
trials, the bait was hidden in the bent tube. Prior to the
start of the experiment, the experimenter held up the bent
tube in front of the monkeys; all monkeys inspected the
tube and looked into it several times. Data were collected
and analyzed as in Experiment 1. Forty percent of trials
were coded for a second time; intra-observer reliability
between codings was high (Pearsons correlation:
115
r=+0.94, p<0.001 for searches; r=+0.98, p<0.001 for
selection).
Results
Selection behavior
Kiki, Zinnia and Theta selected a tube on all trials. When the
bait was located inside the bent tube, Heiji and Zilla failed
to select a tube on four and six trials respectively despite
searching in all the tubes repeatedly. Looking at only those
trials on which a selection was made, Zilla, Heiji and Kiki
selected the correct tube signicantly above chance (33%),
only when a straight tube was baited (straight tube baited:
Heiji: 20/20, Kiki: 13/20 and Zilla: 18/20 correct, binomial
tests: all p<0.005; bent tube baitedHeiji: 3/8, Kiki: 4/12
and Zilla: 2/6 correct, binomial tests: all p>0.05;). Theta
and Zinnia continued to showstrong location biases against
and for the middle tube (0 and 29 selections out of 32,
respectively). The following analyses are again restricted to
the three successful monkeys, assessed individually using
nonparametric tests.
Search behavior
Zilla, Heiji and Kiki looked into at least one tube on
all trials. Figure 3 shows how the monkeys made signif-
icantly more number of looks on trials when the bait was
in the bent tube compared to trials where the bait was
in a straight tube, regardless of whether the trial ended
with a correct or incorrect response (MannWhitney tests:
Heiji: z=3.569, p<0.001; Kiki: z=1.988, p=0.047;
Zilla: z=3.269, p<0.001). Comparing the number of
looks into the straight tubes and the bent tube on each
trial, the monkeys looked equally frequently into the bent
tube and the straight tubes (adjusted for unequal number
of tubes; Wilcoxon Signed Ranks tests: Heiji: z=1.039,
Fig. 3 Experiment 2average number of looks per tube per trial
according to baiting conditions. ST-VI: looks into straight tubes, bait
in straight tube; BE-VI: looks into bent tube, bait in straight tube;
ST-IN: looks into straight tubes, bait in bent tube; BE-IN: looks into
bent tube, bait in bent tube
Kiki: z=0.744, Zilla: z=1.034, all p>0.05). Finally, no
successful super-efcient searches occurred, although a
total of 36 was possible for all the three monkeys combined.
Discussion
The same three monkeys completed Experiment 2 above
chance, but only when the bait was placed inside a straight
tube and therefore could be located through visual search-
ing. Furthermore, when the bait was inside the bent tube,
the monkeys looked into the tubes more frequently, and
two monkeys failed to select a tube on a total of 10 trials.
It appears that the monkeys failed to understand the visual
affordances of the bent tube. Instead, they treated the bent
tube much as if it was a straight tube, i.e. when no bait
was visible, they continued searching and failed to infer
the baits location within the bent tube.
The monkeys did not avoid searching the bent tube, which
may suggest that they failed to understand that lines of
sight must be straight. In humans, this ability appears to
develop around 5 years of age. Flavell et al. (1991) showed
young children tubes with different degrees of curvature,
and asked themto predict whether they would be able to see
objects placed at the other end of the tubes when looking
through them. Over 70% of 3-year-old children and 53%
of the 5-year-olds believed they would be able to see the
objects through a 90

bent tube, failing to appreciate that

lines of sight must be straight. The children were then
allowed to test their predictions and to look through one
of the tubes. After this feedback, the same tubes as in pre-
feedback were shown to the children, who again were asked
whether they would be able to see an object placed at the
other end of the tubes. This time, 80% of the 5-year-olds
gave the correct answer, while 68% of the 3-year-olds still
failed to answer correctly.
Capuchin monkeys might similarly fail to realize that
they cannot see around corners. However, since the baiting
was done invisibly, they did not know that the bait was
only placed behind and not in front of the bend and there-
fore always hidden from view. Possibly with more trials,
the monkeys would have mastered this aspect of the ex-
perimental procedure. We conducted only a small number
of trials to test the monkeys spontaneous understanding
of their visual search behavior and to exclude improved
performance through simple associations; however future
studies could usefully run more trials with bent tubes to
assess eventual adaptation of the monkeys understanding
of their own search behavior.
General discussion
In two experiments, three out of ve capuchin monkeys
successfully searched for a food reward that was placed
inside one of the three tubes. In Experiment 1, these
monkeys continued searching when they already knew
(or should have known) that the bait was visible inside
transparent tubes. Searching did not increase their success,
116
and no instances of inferring the bait location in an opaque
tube occurred. In Experiment 2, when the bait was placed
within a bent tube so that it could not be seen, the monkeys
failed to infer the correct location and selected tubes
randomly. The monkeys also persistently searched in the
bent tube, failing to show any understanding that they
could not obtain any information about the bait location
from it.
This redundant search behavior by the capuchins con-
trasts with what could be expected in the presence of
metacognitive awareness, which would allow the monkeys
to search only when necessary, i.e. when they do not al-
ready have the relevant information. Instead, it appears that
seeing the bait inside a tube eventually triggered the se-
lection of that tube. This interpretation is supported by the
fact that in Experiment 1, the majority of trials consisted
of the monkeys looking into the baited tube immediately
before making a selection, and the absence of any super-
efcient searches. Furthermore in Experiment 2, when the
bait could not be seen, two otherwise successful monkeys
failed to select a tube on a total of 10 occasions. Execution
of the selection response, after seeing the bait, is consistent
with response competition. In this context, response com-
petition pits searching against selecting a tube (Hampton
et al. 2004); if the tendency to reach for a tube is weak,
seeing the bait will strengthen the response.
If the response competition hypothesis is correct, a possi-
ble explanation for the difference in searching behavior be-
tween our capuchin monkeys and Hampton et al.s (2004)
rhesus macaques might lie in the two species threshold
for the reaching response. Rhesus macaques might have a
lower reaching threshold than capuchin monkeys, the lat-
ter requiring more searching before a selection is made.
The present experimental setup might also have kept the
switchover point from searching to selection high, as mul-
tiple looks into tubes were not penalized and it took little
effort to check the tubes. In addition, capuchin monkeys are
known for their almost excessive curiosity and exploratory
drive (Fragaszy et al. 2005), which may have further in-
creased this species likelihood of searching before making
a selection. Future studies might increase the relative cost
of searching by spacing the tubes further apart, or adjust-
ing the height of tubes to make them more difcult to look
into (Hampton et al. 2004). However, as Hampton et al.
(2004) pointed out, the relative strength of search or choice
responses could also be inuenced by metacognitive abil-
ities, in that awareness of an absence of knowledge about
the bait location could strengthen the search response. This
interpretation would suggest that capuchin monkeys have a
lower appreciation of the consequences of searching tubes
on their knowledge states. However, given our small sam-
ple size and lack of varied age and sex classes, the present
results should be regarded as preliminary and in need of
replication.
It is worth noting that a comparable failure to fully com-
prehend the consequences of actions has been described for
capuchin monkeys tool use. Visalberghi and Limongelli
(1996) reported that a capuchin monkey that used straw to
dip for honey through holes in the side of a box persisted in
trying to dip through the sides when the holes were moved
to the top of the box. Fujita et al. (2003) found that capuchin
monkeys are sensitive to the spatial arrangements of food
and tools, but fail to take into account other environmen-
tal circumstances. Fragaszy et al. (2004) argued that even
though capuchin monkeys are procient tool users, they
fail to understand the underlying causal relations between
tools and objects, and do not anticipate outcomes of their
manipulations. Capuchins may therefore generally fail to
consider possible alternative future outcomes.
It is also conceivable that the capuchin monkeys exces-
sive searching reected an inability to inhibit previously
learned behavioral sequences. Capuchin monkeys, but not
chimpanzees, nd it difcult to inhibit cursor movements
directly toward the goal in mazes presented on a computer
screen (Fragaszy et al. 2003). In spite of the limited number
of trials, it remains possible that our capuchin monkeys ac-
quired an association between searching and food reward,
in which case the tube paradigm would become no more
than a visual discrimination task. Difculties with inhibit-
ing learned behavioral sequences might have caused the
two unsuccessful monkeys in the current study to develop
strong location biases, and the three successful monkeys to
unnecessarily search in the tubes.
One potentially fruitful approach would be to assess du-
ration of visual search behavior in relation to decision pro-
cesses. Using a simple discrimination task with stump-
tailed macaques, Schrier (1983) found that the maximum
frequency of visual scanning coincided with the highest
percentage of correct selections, while the duration of xa-
tions remained stable. Schrier suggested that scanning pat-
terns reect stages of information processing, and dura-
tion of xations the later decision-making processes. If this
is correct and capuchin monkeys possess metacognitive
abilities, then we might expect to nd the longest looks
associated with looking into the baited tube, reecting a
decision-making process, whereas looks into empty tubes
might be comparatively short, perhaps especially just be-
fore making a selection. Unfortunately, the video record-
ings of the present experiments did not give enough detail
to investigate these issues, but future studies might utilize
this paradigm.
Metacognitive theories suppose an intimate link between
understanding ones own mental states and understanding
others mental states. For example, simulation theorists
hold that an individual rst comes to understand its own
mental states through introspection. This insight is then
used to extrapolate from own mental states to those of
others, through simulating their mental activity (Gordon
1986; Goldman 1993; Povinelli and Vonk 2003). If the
present results indicate limitations of capuchin monkeys
understanding of their own mental state of knowing, we
might expect related constraints on their understanding of
others knowledge states and resulting behaviors. However,
the data on this issue are equivocal. Capuchin monkeys can
learn to discriminate between a person knowing and a per-
son guessing the location of a food reward, but it requires
much training and is not readily generalizable (Kuroshima
et al. 2002; 2003).
117
In conclusion, compared to rhesus macaques and apes,
capuchin monkeys show redundant searches on the tube
task, which might reect limited metacognitive abilities,
but other explanations are possible. Should additional evi-
dence eventually weigh in favor of weak or absent metacog-
nitive abilities in capuchin monkeys, then this might in-
dicate an important phylogenetic difference between the
Old World and New World monkeys, namely, the com-
mon ancestors of the Old World monkeys and apes evolved
metacognitive abilities after the split with the New World
monkeys.
Acknowledgements This work was supported by the Japanese
Ministry of Education, Culture, Science, Sports, and Technology
(MEXT) through Grant in Aid for Scientic Research No. 13410026
to K. Fujita, and the 21st Century Center of Excellence Program
(D-10) to Kyoto University. The experiments in this paper complied
with the Guide for the care and use of laboratory primates, Primate
Research Institute, Kyoto University, and the current Japanese laws
on animal experimentation.
References
Bower TGR (1982) Development in infancy, 2nd edn. Freeman, San
Francisco
Call J, Carpenter M (2001) Do apes and children know what they
have seen? Anim Cogn 4:207220
Dumas C, Brunet C(1994) Object permanence in capuchin monkeys:
a study of invisible displacements. Can J Exp Psychol 48:341
358
Flavell JH, Green FL, Herrera C, Flavell ER (1991) Young chil-
drens knowledge about visual perception: lines of sight must be
straight. Brit J Dev Psychol 9:7387
Fragaszy DM, Johnson-Pynn J, Hirsh E, Brakke K (2003) Strate-
gic navigation of two-dimensional alley mazes: comparing
capuchin monkeys and chimpanzees. Anim Cogn 6:149
160
Fragaszy DM, Visalberghi E, Fedigan LM (2004) The complete ca-
puchin. Cambridge University Press, Cambridge
Fujita K, Kuroshima H, Asai S (2003) How do tufted capuchin mon-
keys (Cebus apella) understand causality involved in tools? J
Exp Psychol: Anim Behav Process 29:233242
Goldman AI (1993) The psychology of folk psychology. Behav Brain
Sci 16:1528
Gordon RM (1986) Folk psychology as simulation. Mind and Lang
1:158171
Grifn DR(2003) Signicant uncertainty is common in nature. Behav
Brain Sci 26:346
Hampton RR (2001) Rhesus monkeys know when they remember.
Proc Natl Acad Sci 98:53595362
Hampton RR, Zivin A, Murray EA(2004) Rhesus monkeys (Macaca
mulatta) discriminate between knowing and not knowing and
collect information as needed before acting. Anim Cogn 7:239
246
Inman A, Shettleworth SJ (1999) Detecting metamemory in nonver-
bal subjects: a test with pigeons. J Exp Psychol: Anim Behav
Process 25:389395
Kuroshima H, Fujita K, Fuyuki A, Masuda T(2002) Understanding of
the relationship between seeing and knowing by tufted capuchin
monkeys (Cebus apella). Anim Cogn 5:4148
Kuroshima H, Fujita K, Adachi I, Iwata K, Fuyuki A (2003) A
capuchin monkey (Cebus apella) recognizes when people do
and do not know the location of food. Anim Cogn 6:283291
McGonigle B, Chalmers M, Dickinson A (2003) Concurrent disjoint
and reciprocal classication by Cebus apella in seriation tasks:
evidence for hierarchical organization. Anim Cogn 6:185197
Nelson TO, Narens L (1990) Metamemory: a theoretical framework
and new ndings. In: Bower G (ed) The psychology of learning
and motivation. Academic Press, New York, pp. 125173
Povinelli DJ, Vonk J (2003) Chimpanzee minds: suspiciously human?
Trends Cogn Sci 7:157160
Schrier AM (1983) Eye movements of stumptailed monkeys during
discriminative performance: an overview. Int J Primatol 4:373
381
Smith JD, Schull J, Strote J, McGee K, Egnor R, Erb L (1995)
The uncertain response in the bottlenosed dolphin (Tursiops
truncatus). J Exp Psychol: Anim Behav Process 124: 391408
Smith JD, Shields WE, Washburn DA (2003) The comparative psy-
chology of uncertainty monitoring and metacognition. Behav
Brain Sci 26:317373
Visalberghi E, Limongelli L (1994) Lack of comprehension of cause-
effect relations in tool-using capuchin monkeys (Cebus apella).
J Comp Psychol 108:1522