social learning in captive african elephants (loxodonta africana africana)
TRANSCRIPT
ORIGINAL PAPER
Social learning in captive African elephants(Loxodonta africana africana)
Brian J. Greco • Tracey K. Brown •
Jeff R. M. Andrews • Ronald R. Swaisgood •
Nancy G. Caine
Received: 1 May 2012 / Revised: 29 November 2012 / Accepted: 30 November 2012 / Published online: 14 December 2012
� Springer-Verlag Berlin Heidelberg 2012
Abstract Social learning is a more efficient method of
information acquisition and application than trial and error
learning and is prevalent across a variety of animal taxa.
Social learning is assumed to be important for elephants,
but evidence in support of that claim is mostly anecdotal.
Using a herd of six adult female African bush elephants
(Loxodonta africana africana) at the San Diego Zoo’s
Safari Park, we evaluated whether viewing a conspecific’s
interactions facilitated learning of a novel task. The tasks
used feeding apparatus that could be solved in one of two
distinct ways. Contrary to our hypothesis, the method the
demonstrating animal used did not predict the method used
by the observer. However, we did find evidence of social
learning: After watching the model, subjects spent a greater
percentage of their time interacting with the apparatus than
they did in unmodeled trials. These results suggest that the
demonstrations of a model may increase the motivation of
elephants to explore novel foraging tasks.
Keywords Elephants � Loxodonta � Social learning �Imitation � Animal cognition
Introduction
Social learning occurs when an animal acquires informa-
tion from another individual’s behavior and then applies
that information in a similar situation at a later time to
achieve a similar objective (Coussi-Korbel and Fragaszy
1995). Psychologists typically categorize social learning
into a variety of imitative and non-imitative forms. For
example, in true imitative social learning the observer
copies some portions of a behavioral motor sequence
(Dawson and Foss 1965; Whiten et al. 1996; Zentall et al.
1996; Zentall 2004). The various forms of non-imitative
social learning can take place when the model’s behavior
enhances or gives new value to novel stimuli, objects,
locations, or events (Heyes 1993; Call and Carpenter 2002;
Zentall 2006). For example, social facilitation occurs when
a conspecific’s behavior motivates the observer to engage
in a similar behavior (e.g., foraging-specific behaviors)
(Fragaszy and Visalberghi 1990; Zentall 2006). Stimulus
enhancement plays a role in non-imitative social learning
when the actions of a model draw the attention of an
observing animal toward a specific stimulus or part of a
stimulus (e.g., an object or location) (Call and Carpenter
2002; Whiten et al. 2004; Zentall 2006). Observational
conditioning allows an observing animal to learn some-
thing about the relationship between two stimuli, as occurs
when an observer learns from watching a model that a food
All experimental procedures were approved by the San Diego Zoo’s
IACUC #09-014.
Electronic supplementary material The online version of thisarticle (doi:10.1007/s10071-012-0586-7) contains supplementarymaterial, which is available to authorized users.
Present Address:B. J. Greco (&)
Department of Animal Science, One Shields
Avenue, Davis, CA 95616, USA
e-mail: [email protected]
B. J. Greco � J. R. M. Andrews � R. R. Swaisgood
San Diego Zoo Global, San Diego, CA, USA
T. K. Brown
Department of Biological Sciences, California State University
San Marcos, San Marcos, CA, USA
N. G. Caine
Department of Psychology, California State University San
Marcos, San Marcos, CA, USA
123
Anim Cogn (2013) 16:459–469
DOI 10.1007/s10071-012-0586-7
hopper delivers (i.e., is associated with) food (Call and
Carpenter 2002; Zentall 2006). Imitative and non-imitative
learning can and often do co-occur, and it can be difficult to
determine which is operating at a given time (Zentall
2006).
Many social learning experiments rely on an observer–
demonstrator paradigm, which compares the performance
of a non-exposed control group to that of subjects who are
first allowed to observe a conspecific demonstrate a novel
task (Voelkl and Huber 2000; Horner et al. 2006; Dindo
et al. 2008). Following a predetermined number of suc-
cessful demonstrations, the demonstrating animal is
removed from the enclosure, and the observer is allowed to
participate in the experimental task by itself (Voelkl and
Huber 2000; Horner et al. 2006; Dindo et al. 2008). If the
performance of the observing subjects is significantly dif-
ferent from that of the non-exposed subjects, inferences can
be made about the presence of social learning (Voelkl and
Huber 2000). For example, should the observers reliably
show a bias toward the motor sequence and solution
method demonstrated by the model (i.e., copying fidelity),
imitation may be responsible (Voelkl and Huber 2000;
Caldwell and Whiten 2004; Horner et al. 2006; Dindo et al.
2008). Additionally, changes in performance speed, accu-
racy, and/or the ways in which observing subjects manip-
ulate an apparatus may be interpreted as indicators of
non-imitative social learning (Fragaszy and Visalberghi
1990; Caldwell and Whiten 2004). For example, in a
number of studies of social learning in capuchins (Cebus
apella), Fragaszy and Visalberghi (1990) found that the
behavior of models who solved novel foraging tasks
influenced the actions of group members who had not yet
learned the task on their own. The authors concluded that
increased apparatus exploration time, elevated apparatus
manipulation, and faster solution times by observers were
likely attributable to stimulus enhancement and/or social
facilitation. Evidence of learning by imitation has been
claimed for certain species of birds (Dawson and Foss
1965; Zentall 2004), primates (Whiten et al. 2004), and
cetaceans (Rendell and Whitehead 2001; Krutzen et al.
2005). Non-imitative social learning has been demon-
strated in a larger number of animal taxa, including fish
(Schuster et al. 2006), rats (Ray and Heyes 2002), bats
(Page and Ryan 2006), gray squirrels (Hopewell et al.
2010), ravens (Stowe et al. 2006), horses (Krueger and
Heinze 2008; Krueger et al. 2011), swine (Oostindjer et al.
2011), parrots (Huber and Gajdon 2006), and capuchins
(Dindo et al. 2009), to name just a few.
Female elephants (Loxodonta and Elephas) live in
hierarchical matrilineal family groups composed of 5–10
related adult females and their immature offspring (Moss
1988; Sukumar 2003; Byrne et al. 2009). In contrast,
mature male elephants spend most of their time living
solitarily, only occasionally associating with female herds
or bachelor herds (Moss 1988; Sukumar 2003). In African
bush elephants (Loxodonta africana africana), family
groups are also known to aggregate into large bond groups,
in which the elephants travel, forage, and communally
raise/protect their young (Moss 1988; Sukumar 2003; By-
rne et al. 2009). Their cooperative behaviors are mediated
by acoustic, chemical, seismic, tactile, and visual com-
munication signals (Langbauer 2000).
Given the social cohesiveness of elephant herds and the
strength of social bonds among elephants, there seems to be
ample opportunity for social learning to occur. Surpris-
ingly, however, evidence of social learning in elephants is
limited primarily to anecdotes (Byrne et al. 2009). The
reasons for this curious absence are probably related to the
challenges elephants pose as experimental subjects. Not
only does their extraordinary size make them difficult and
potentially dangerous to manipulate, but constructing
experimental apparatus that can withstand an elephant’s
strength and curiosity is formidable. Furthermore, it is
impossible to maintain the amount of experimental control
that can be achieved with smaller, more tractable animals
such as birds [e.g., Campbell et al. (1999)]. Nonetheless, to
better understand the ways in which elephants adapt and
respond to ecological and social challenges, we need to find
ways to investigate their learning styles and capabilities.
Here we report the results of a controlled study of social
learning in a group of African bush elephants (Loxodonta
africana africana) housed in captivity. We expected that
adult female elephants would use the same technique for
solving a foraging task as a conspecific model. We also
developed two metrics, focus score and initial interest, to
investigate other forms of social learning as well.
Methods
Subjects
The subjects for this study were six female African bush
elephants (Loxodonta africana africana), ranging in age
from 18 to 21 years and living at San Diego Zoo’s Safari
Park. These females are believed to be related, but their
complete histories are not known. The elephants were
wild-born in South Africa and brought to the Safari Park
in 2003 after it was learned that they were to be culled for
the purpose of population control (Andrews et al. 2005).
At the time of our study, all but one of the females had at
least one calf that was born while living at the Safari
Park. In addition to these six subjects, two adult male
African elephants were used to test the safety, function-
ality, and difficulty of the experimental apparatus in pilot
trials.
460 Anim Cogn (2013) 16:459–469
123
The six female subjects and 11 other herd members (two
adult bulls and nine calves) lived in a 2.23-ha outdoor
enclosure. Experimental trials were conducted in two
adjacent, 0.20-ha enclosures separated by a steel-cable
barrier fence through which the elephants could see each
other. During experimental trials, subjects were removed
from the main group for no more than 60 min at a time.
Water was provided during each trial.
Experimental apparatus
The six apparatus used in this study were individually
designed and constructed to be unique and novel to the
subjects. All apparatus delivered food reward (alfalfa-
based dry herbivore pellet or pressed alfalfa) when solved,
but no two apparatus shared the same mechanisms. One
apparatus could only be solved in one way, but the
remaining five apparatus had two distinct solutions that
involved different behavioral actions. Additionally, many
of the apparatus were constructed with either noticeably
movable or breakaway components on the apparatus. Thus,
a subject could cue on the model’s movements, apparatus
movements, the location of apparatus components, and/or
breakaway apparatus components when observing the
model (see Fig. 1 and the supplementary material for
drawings and photographs).
The Kerplunk (K) was a roughly cylindrical apparatus
composed of six open-ended pipes attached to a central
shaft. Four of the six outer pipes were stuffed with pressed
alfalfa bundled in paper (to ensure the alfalfa slid out when
the K was solved). When hung vertically, four bamboo
shafts transected the four stuffed pipes, preventing the
alfalfa from sliding out and preventing the subjects from
pulling the alfalfa out of the apparatus. The K was the only
one of the six apparatus that did not include two different
solution options. In order for a subject to solve the appa-
ratus, she needed to remove two bamboo shafts.
The Counterweighted Feeder (CW) was composed of a
long rope slung over a tree. One end of the rope was free
and accessible to the subjects, while the other end was
weighted at ground level. A bundle of pressed alfalfa was
tied to the free end of the rope out of the subjects’ reach.
When given access to the remaining length of free rope, a
subject could draw the alfalfa bundle toward herself by
pulling. If the rope was not secured, the weight would tug
the alfalfa back out of reach. However, one of two
behavioral motor sequences—(a) securing the slack with
the mouth or (b) securing the slack with a forefoot—could
be used to ‘‘choke-up’’ on the rope and draw the alfalfa
bundle into trunk’s reach.
The Pop and Roll (PR) apparatus was a tube with an
internal trolley-mounted feeder ball. When presented to the
elephants, the PR was hung horizontally, just below the
subjects’ eye level. The elephants were prevented from
freely accessing the feeder ball by two end caps. One end
cap was mounted to the trolley, and the other was a free-
floating plug. To solve the PR, the elephants needed to
either remove the free-floating plug (using an attached
handle) or push on the trolley-mounted cap (which would
force the free-floating plug out of the opposite end of the
tube). From the perspective of an observing subject, the
model solving the apparatus by end cap removal would
present a yanking behavior with her trunk from the left side
of the PR, followed by the emergence of her trunk
clutching the end plug. If the model chose to push the
trolley-mounted end cap, she would present a trunk-
Fig. 1 Pop and Roll a subject solving the apparatus by pulling the
end plug. Notice the outward arch of the trunk as the subject pulls the
end plug free. b Subject solving the apparatus by pushing the feeder
ball to force the end plug out. Notice the trunk is forcefully inserted
into the right side of the apparatus, thus popping the free-floating end
plug out of the left side. For detailed descriptions and photographs of
all apparatus, please see the supplementary material section on
experimental apparatus
Anim Cogn (2013) 16:459–469 461
123
thrusting motion from the right side of the PR and the cap
would fall out of the left side. In both scenarios, the model
would retrieve a reward from the left side of the PR. If the
model pushed the trolley-mounted end cap, the subject
might additionally observe the movement of externalized
trolley components.
The Push Pop (PP) was an apparatus with an internal-
ized plunger. When presented to the subjects, the PP was
hung vertically with the plunger arm protruding from the
bottom. Pressed alfalfa was loaded into a cavity on top of
the plunger pad. Subjects could manipulate the plunger by
grasping and thrusting the plunger arm or by striking the
base of the arm with the dorsal surface of the trunk. Both
thrusting and striking caused the plunger to rise, forcing the
alfalfa reward out of the cylinder.
The Boxall (BX) was constructed by combining a plastic
enrichment box and a plastic enrichment ball. The ball was
suspended in the box with one hemisphere protruding from
one of the box’s faces. Two holes were cut into the inner
hemisphere of the ball, and the outer hemisphere was left
intact. On an adjacent face of the cube, a hinged flap was
attached to cover a trunk-sized hole. The apparatus was
presented to the subjects with the ball facing outward and
the flap on the top face of the box. Subjects could solve the
apparatus by rotating the ball, so that there was one hole in
each hemisphere, inner and outer, or by opening the hinged
flap outward. Either action would allow the subject to
insert her trunk into the box to retrieve the alfalfa reward.
The Smash Box (SB) was constructed from a plastic
enrichment box with a weight attached to the bottom face.
The box was modified to pivot end-over-end when
manipulated or struck. On the front face of the SB, a balsa
wood panel covered a 20-cm-diameter hole, and on the top
face of the SB, a rope bung was inserted to plug a 5-cm-
diameter hole. Subjects could solve the SB either by
shattering the balsa wood panel with their tusk or by
pulling out the rope bung. Once the wood was broken or
the bung removed, the box could be spun end-over-end to
collect the food reward.
Procedures
There were two types of trials: modeled, in which a subject
watched the dominant elephant (Swazi; see below) suc-
cessfully operate the apparatus before being allowed to
interact with the apparatus herself, and unmodeled, in
which a subject interacted with the apparatus without
having watched Swazi. Each of the subjects interacted with
three of the apparatus in the modeled condition and three of
the apparatus in the unmodeled condition.
Pilot data from the bull elephants were used to roughly
categorize the six apparatus into easy (K and PP), medium
(CW and SB), and hard (PR and BX) levels of difficulty.
We then attempted to shuffle the order of experimental
trials to allow each subject access to an apparatus of easy,
medium, and hard difficulty in both the modeled and un-
modeled condition, while simultaneously preventing each
subject from acting in the same experimental condition
(modeled or unmodeled) more than two consecutive times.
Additionally, when assigning subjects to condition, we
avoided duplicating any one set of three apparatus (e.g., in
the modeled condition only one subject saw the Counter-
weighted Feeder, the Pop and Roll, and the Smash Box).
Subjects participated in trials about once a week (except for
Swazi, who participated in her own trial and then acted as
model in other trials). One of us (BG) conducted all of the
trials with the aid of an assistant. All trials were videotaped
for later analysis.
Experimental trials: unmodeled condition
Prior to each trial, the apparatus to be tested was mounted
in place. BG called the subject over to the apparatus and
drew her attention to it. He then stepped away, giving the
subject up to 45 min to interact with and solve the appa-
ratus. Once the subject solved the apparatus, she was
allowed to collect any reinforcement distributed by the
apparatus and was given additional food reward to rein-
force her participation. Subjects who failed to solve the
apparatus were also given a food reward at the end of the
45-min trial period, to reinforce participation.
Experimental trials: modeled condition
The most dominant female in the herd, Swazi, served as the
model in all of the modeled trials. Out of view of the rest of
the herd, she was given access to an apparatus in an un-
modeled trial. During this trial Swazi was allowed to solve
each apparatus and was given time to practice that solution
several times. Swazi’s first interaction with each apparatus
was used as data for the analysis of responses in the un-
modeled condition. The decision to use just one model
animal was based on two factors. First, little is known
about elephant social learning and how individual differ-
ences among elephants may influence whether, when, and
how one elephant learns from another. Using just one
model for all subjects controlled for at least one aspect of
the social learning situation (i.e., the identity of the model)
in this study. Second, social hierarchy plays an important
role in elephant herds; if the roles were reversed, a sub-
dominant female may not feel comfortable using an
apparatus that is within close proximity to a more dominant
member of the herd. For this reason, the most dominant
female was selected as the model.
During modeled trials, the apparatus was mounted in
place, and Swazi (the model) was allowed to solve the
462 Anim Cogn (2013) 16:459–469
123
apparatus, while the subject was present in an adjacent
enclosure separated by a fence. Swazi always used the
same solution option as she used on her first successful
trial. Once Swazi successfully solved the apparatus, the
apparatus was reset and the observation process was
repeated. A subject was allowed to view Swazi demon-
strate each apparatus two or three times. The number of
demonstrations was kept constant for each individual
apparatus.
Following her demonstrations, Swazi was removed, the
apparatus reset, and the subject given access to the appa-
ratus as if she were participating in an unmodeled trial.
Approximately 5 min passed between Swazi’s demonstra-
tion and observer participation in the modeled trial.
Dependent variables and data coding
Two assistants, blind to condition and subject, indepen-
dently viewed all of the videotaped trials. The assistants
were trained by BG using videos from pilot trials that made
use of apparatus that were not used in the experiment. The
assistants each scored all 30 trials. Fifteen of these sessions
were unmodeled, and fifteen were modeled. Swazi’s un-
modeled sessions also were scored, but these data were not
used to calculate any of the results.
The times scored by the two assistants were averaged for
use in all of the relevant analyses described below. Inter-
rater reliability was high (r [ 0.90) across scorers.
Interaction time was scored from the unmodeled trials
and modeled trials. Because the subjects sometimes turned
their attention away from the apparatus or got distracted
with something unrelated to the apparatus, interaction time
included only those times when the elephant was actually
touching the apparatus in some way. A touch was defined
as any contact with the apparatus that appeared to be
intentional (i.e., trunk, rostrum, foot, or body contact nec-
essary to manipulate an apparatus, but not incidental con-
tact associated with proximity). Solution time is the
duration of interaction time on a particular trial. If an
individual subject solved an apparatus, her solution time
and interaction time would be the same.
Modeled trials were scored for the behavior of both the
model and the observer. First, the assistants scored Swazi’s
interaction time to give us one measure of how much
information was available to the observing elephant in
modeled trials. This is referred to as demonstration time for
a particular trial. Then the interaction time of the subjects
was scored, as above.
In addition to the scoring by the research assistants, BG
also viewed all of the videotapes to extract the following
information: (a) Exposure time: The total amount of time
elapsed between a subject’s first touch of an apparatus and
the solve event. (b) Focus score: Each subject’s interaction
times were divided by her appropriate exposure times to
calculate a focus score. When interaction and exposure
times are similar (a focus score approaching 1.0), it means
that an elephant was very engaged in the task. A low focus
score means that the subject often did not focus her
attention on the task (a focus score greater than 1.0 is not
possible, as interaction time cannot be greater than expo-
sure time). (c) Solution method: which of the two possible
solutions was chosen by the subjects in each trial.
(d) Number of touches: the number of times a subject
intentionally touched an apparatus. High rater reliability
was reached for number of touches. (e) Bouts to solution:
number of bouts each subject required to solve each
apparatus. A bout was defined as any string of continuous
touches occurring within 30 s of one another. Note that a
bout could consist of just one touch or up to 49 touches (the
maximum number of touches recorded per bout). High
rater reliability was reached for bouts to solution. As an
assessment of his reliability and consistency in scoring, BG
scored and later rescored a subset of the trials and achieved
nearly perfect agreement (r = 0.99) on both touches and
bouts. (f) Initial interest: The duration of each subject’s
first bout on an apparatus was divided by her total inter-
action time with that apparatus. Subjects that showed
maximum interest in an apparatus upon first given access to
it have initial interest scores approaching 1.0.
Finally, BG examined the video data from the three
apparatus that featured two separate solution mechanisms
(the PR, BX, and SB) in order to determine whether the
observers touched the apparatus in locations that the model
also touched. This was done by mapping the location of
touches onto schematic drawings of the apparatus. Using
these touch maps, we could estimate the number of times an
observing subject’s touches overlapped Swazi’s touches.
Results
Trials
Copying fidelity
Contrary to our hypothesis, subjects in the modeled con-
dition did not regularly copy the methods used by the
model to solve the novel apparatus. In the thirteen modeled
trials with the apparatus that offered two solutions, four
trials were solved using the same method as the model, six
were solved using the different method, and in three trials
the subject failed to solve the apparatus at all (Figs. 2, 3). If
the initial difficulty ratings are accounted for, solution
failures occurred on two difficult apparatus, one moderate
apparatus, and one easy apparatus. The solutions used by
the subjects in the modeled and unmodeled conditions were
Anim Cogn (2013) 16:459–469 463
123
distributed fairly evenly between the two available options
(Fig. 3).
Solution time
The subjects solved the apparatus on 25 of the 30 trials.
Three of the five failures were in the modeled condition.
Ndlula failed to solve apparatus 4, 5, and 6, Umoya failed
at solving the third apparatus, and Litsemba failed to solve
the sixth apparatus. When an elephant failed to solve a trial
within the time limit (45 min), its solution time was
recorded as the highest solution time scored by a subject on
that particular apparatus.
Within-subject analysis revealed no significant differences
for average solution time between the modeled (M = 189.07,
SD = 63.69) and unmodeled conditions (M = 184.93,
SD = 67.81) (paired samples t test: t(4) = 0.08, N = 5,
P = 0.94) (Fig. 4). We hypothesized that subjects would
score better (i.e., lower average solution times) in the modeled
condition than in the unmodeled condition. For one of the
subjects, this was true. However, two of the remaining sub-
jects showed similar average solution times in the modeled
and unmodeled condition, and two showed lower average
solution times than unmodeled subjects.
Focus scores
On average, when subjects participated in the modeled
condition, they displayed focus scores (interaction time
divided by exposure time) (M = 0.46, SD = 0.08) that
were 50.2 % greater than the focus scores they achieved in
unmodeled trials (M = 0.31, SD = 0.08) (paired samples
t test: t(4) = 3.71, N = 5, P = 0.02) (Fig. 5). Indeed, all
five subjects had higher focus scores in the modeled than in
the unmodeled condition. It is interesting to note that those
individuals who failed to solve an apparatus often dis-
played lower focus scores for that apparatus.
Touches and bouts
There was no significant difference in the average number
of touches when subjects were in the modeled (M = 53.7,
SD = 25.9) versus unmodeled conditions (M = 51.6,
SD = 28.6) (paired samples t test: t(4) = 0.2, N = 5,
P = 0.9) (Fig. 6), nor was there a significant difference
in the average number of bouts in the two conditions
(modeled: M = 10.8, SD = 7.3; unmodeled: M = 9.1,
SD = 4.1) (paired samples t test: t(4) = 0.6, N = 5,
P = 0.6).
Touch-map analysis revealed that Swazi touched the
apparatus in solution-relevant and non-solution-relevant
locations during all demonstrations. The total number of
touches ranged greatly, from 10 to 51, and touches directed
toward solution-relevant locations were always weighted
toward the solution mechanism she demonstrated (average
ratio: Pop and Roll = 2:1; Boxall = 21:1; Smash
Box = 5:2). In contrast, the solution-relevant touches of
the modeled condition subjects were more evenly distrib-
uted (average ratio: Pop and Roll = 17:11; Boxall = 19:7;
Smash Box = 7:8) (Fig. 7). On average, 49.8 % of subject
touches overlapped the locations touched by Swazi (the
model) during demonstration, and of these touches, an
average of 25.3 % overlapped solution-relevant locations
and an average of 22.2 % overlapped non-solution-relevant
areas. Additionally, an average of 58.3 % of touches
occurred in areas of the apparatus that had small openings
to the food chamber (e.g., seams, cracks, and tie-down
eyelets), and an average of 37.9 % of these touches over-
lapped areas touched by Swazi.
Initial interest
On average, when subjects participated in the modeled
condition (M = 0.48, SD = 0.22), their initial interest in
an apparatus (interaction time of the first bout divided
by total interaction time) was 102 % greater than the
initial interest of subjects who participated in the unmod-
eled condition (M = 0.24, SD = 0.10) (paired samples
t test: t(4) = 3.11, N = 5, P = 0.036) (Fig. 8). As was
the case with focus scores, all subjects displayed higher
initial interest in the modeled than in the unmodeled
condition.
Umoya
Ndlula
Umngani
Litsemba
Lungile
SF
SF
SF
Fig. 2 The copying fidelity of observing subjects according to
apparatus. Apparatus are represented by the icons from left to rightCounterweighted Feeder, Pop and Roll, Push Pop, Boxall, and SmashBox. Stars indicate a match in solution method. Crossed out boxesindicate solution methods used other than the model’s. Blank cellsindicate that the subject interacted with an apparatus in the
unmodeled condition. SF stands for solution failure. Only two
subjects observed the model demonstrate the Push Pop and the Boxall
464 Anim Cogn (2013) 16:459–469
123
Additional analyses
Concordant with the similar solution times across the two
(modeled and unmodeled) conditions, Swazi’s average
demonstration times were not correlated with the observ-
ers’ average solution times (Pearson’s r = 0.15, N = 5,
P = 0.81). These correlations were calculated again to
compare data from apparatus that had been demonstrated
twice by the model, and then again for data demonstrated
three times by the model. In comparing these correlations,
there was no indication that extra demonstration time
influenced subjects’ performance in the modeled condition.
Discussion
We predicted that the elephants would demonstrate social
learning by using the same technique to solve a foraging
task as they had seen in a demonstration session by a
familiar conspecific. However, the subjects were just as
likely to use an alternate solution as they were to copy the
Mouth Pull ekirtStooF llaBhsuP Thrust Flap Rope Panel
SF: 1
1
3
2
SF: 1 SF: 1
SF: 1
Mod
eled
Unm
odel
ed
1
2
ALT:1
3
SF: 1
Fig. 3 This figure demonstrates
the use of solution options by
apparatus in both modeled and
unmodeled conditions
(including the options Swazi
learned first during model
training). Apparatus are
represented by the icons from
left to right: CounterweightedFeeder, Pop and Roll, PushPop, Boxall, and Smash Box. In
each case the total adds up to six
trials. SF stands for solution
failure. One subject solved the
Pop and Roll via an alternate
method (ALT). Only two
subjects observed the model
demonstrate the Push Pop and
the Boxall
Fig. 4 Average solution times per subject are listed according to
dominance rank, with the most dominant on the left and least on the
right
Fig. 5 Focus score are shown in unmodeled and modeled conditions
Anim Cogn (2013) 16:459–469 465
123
actions of the model. Copying fidelity is a measure of true
imitation, whereby an observer uses only the actions of the
model, and not changes associated with the apparatus itself,
to guide its subsequent interactions with that apparatus.
Had we seen evidence of copying fidelity, it would not
have been certain that the criteria for true imitation had
been met, insofar as our apparatus, with the possible
exception of the PP and CW apparatus, were not true ‘‘two-
action’’ apparatus (Dawson and Foss 1965), which are
designed to test for true imitation (Voelkl and Huber 2000).
However, the data from our two (possible) two-action tasks
are not more suggestive of true imitation than the data from
the other tasks; in each case, half of the subjects did not use
the solution method they had seen in the demonstration by
the model.
In addition to copying fidelity, we examined various
measures of performance that can shed light on non-imi-
tative forms of social learning. For example, if subjects
solve a problem more quickly after witnessing a model
successfully gain food reinforcement from an apparatus, it
could indicate that the observation of a model has increased
the motivation of the observers or drawn attention to the
aspects of the apparatus that are relevant to a solution
(Fragaszy and Visalberghi 1990). However, when com-
paring subjects’ scores across the two conditions, the
solution times of the subjects were not significantly dif-
ferent. Furthermore, on three occasions the elephants failed
to solve the apparatus at all, even though they had the
benefit of watching the model successfully solve the
apparatus two or three times in a row.
The ways and extent to which a subject physically
interacts with an apparatus (number/location of touches
and numbers/duration of bouts of touches in the current
study) also can help to elucidate forms and mechanisms of
both imitative and non-imitative learning in a social
learning situation. For example, despite the fact that none
of their subjects was able to solve a novel ‘‘two-action’’
Fig. 6 The mean number of touches are diagrammed in modeled and
unmodeled conditions
26Swazi44Umoya
03Swazi01Lungile
46Swazi42Litsemba
016Swazi38Umngani
226Swazi411Lungile
05Swazi136Umoya
23Swazi30Umngani
68Swazi628Lungile
SR2SR1Model IDSR2SR1Modeled Condition Subject ID
Fig. 7 The number of touches to solution-relevant components of the
Pop and Roll, Boxall, and Smash Box are demonstrated above.
Modeled condition subject touches are tallied on the left, with touches
demonstrated by the model (Swazi) tallied on the right. Each of the
examined apparatus featured two solution-relevant components (SR):
Pop and Roll SR1 = pull plug; SR2 = push trolley; Boxall:
SR1 = ball; SR2 = flap; Smash Box: SR1 = pull rope; SR2 = break
panel
466 Anim Cogn (2013) 16:459–469
123
foraging apparatus, Caldwell and Whiten (2004) found that
common marmosets (Callithrix jacchus) that had observed
a model’s demonstration more frequently touched the
apparatus in locations used by the model than marmosets
that had not seen a model’s demonstration. Such demon-
stration-consistent touching could arise from stimulus
enhancement or social facilitation, both of which are forms
of non-imitative social learning.
In contrast to the findings of Caldwell and Whiten
(2004), the present study failed to reveal evidence of
demonstration-consistent touching by the subjects. Despite
the fact that Swazi’s demonstrated touches were always
weighted (at least 2:1, and usually much greater) toward
the solution mechanism she demonstrated, the attention of
the observing subjects was more evenly distributed. We did
find that an average of 49.8 % of all touches by modeled
condition subjects overlapped locations where Swazi had
touched during demonstration. However, 37.9 % of these
touches overlapped in areas of the apparatus that had small
openings to the food chamber (e.g., seams, cracks, and tie-
down eyelets). Therefore, we believe that the overlap of
subjects’ touches with those of the model is more consis-
tent with an interest in food-based odors emanating from
the apparatus than they are suggestive of learning from the
model.
We created two measures, the focus score and the initial
interest score, to investigate the effects of modeling that
were not addressed by any other variables alone. Focus
scores—a measure of the proportion of time elephants
spent interacting with the apparatus—address a subject’s
engagement in the task at hand. Focus scores were, on
average, 50.2 % higher in modeled trials than in unmod-
eled trials, and all five elephants had higher focus scores in
the former. Initial interest—a measure of the proportion of
interaction time used in the first bout—is similar to that of
focus scores in that it addresses a subject’s engagement in a
task, but it may be a more accurate measure of the
immediate affects of observing a modeled demonstration.
On average, subjects in the modeled condition demon-
strated 102 % more initial interest in apparatus than sub-
jects in the unmodeled condition. Together, these results
point to the possibility that observing the demonstrations of
a model may increase an elephant’s general motivation to
approach and engage a novel foraging task. Interestingly,
the observing subjects in our study showed higher average
focus scores and initial interest despite the fact that they
did not show higher average numbers of touches.
Thus, it seems possible that observational conditioning
may explain the difference in focus scores and initial
interest across conditions. Observational conditioning may
have occurred because the subjects learned that the appa-
ratus could be made to deliver food, motivating them to
interact with the apparatus with the expectation that they
could be rewarded with pellets or hay. Given the social and
spatial cohesiveness of elephant groups, there is ample
opportunity for elephants to benefit from simply directing
their attention to the consequences and contexts of a con-
specific’s actions.
However, we must also consider that when interacting
with the apparatus during a modeling trial, the model
(Swazi) may have left odors on the apparatus that inter-
fered with or facilitated a subject’s use of the apparatus
during the subsequent trial. It was not feasible to clean the
apparatus (which were large, difficult to handle and store,
and included porous materials) between trials, and cleaning
agents are strictly regulated by the Safari Park. We think it
is unlikely that any odors that Swazi may have deposited
on the apparatus prior to or during a modeled trial would
have been particularly salient, given that the apparatus was
used repeatedly by the subjects, touched by keepers, and
stored on the ground in areas where they may have come in
contact with other olfactory stimuli. Nonetheless, we
reviewed the tapes for evidence that there may have been
olfactory biases in the data. For example, if Swazi’s scent
cues were salient for subjects in the modeled condition, we
should expect to see higher focus scores and initial interest
in subjects who received apparatus that were more fre-
quently touched by Swazi during modeled sessions. This
was not the case. Likewise, Swazi’s scent cues also might
have directed subjects toward Swazi’s chosen solution if
the majority of her scent cues were in proximity to parts of
the apparatus relevant to that particular solution. This also
appears not to be the case. As mentioned previously, when
demonstrating an apparatus, Swazi’s touches were always
Fig. 8 Initial interest is illustrated above in unmodeled and modeled
conditions
Anim Cogn (2013) 16:459–469 467
123
weighted (at least 2:1) toward the solution mechanism she
demonstrated, whereas the attention of the subjects was
more evenly distributed. Furthermore, the subjects failed to
solve the apparatus via Swazi’s chosen mechanism on nine
of the 13 trials. Thus, while we cannot rule out odor cues,
we do not believe that they accounted for the pattern of
results we obtained.
The failure of our elephants to demonstrate copying
fidelity may have been related to the nature of the exper-
imental task. Inspired by studies with primates, we used
apparatus that required the elephants to manipulate con-
tainers that held hidden food rewards. This presents a sit-
uation that is more ecologically valid for primate species
(e.g., nut cracking, seed extraction, and ‘‘fishing’’ for
insects) than it is for elephants, whose foods do not typi-
cally require processing (Johnston 1981; Tomasello and
Call 1994; Sukumar 2003; Hart et al. 2008; Hermes et al.
2008). Furthermore, adult elephants probably only rarely
come across novel foraging situations, limiting the need for
social transmission of new feeding techniques. Elephants
develop the majority of their feeding/foraging skills within
the first year of life (Sukumar 2003), suggesting that
immature elephants may be the best subjects to use when
looking for evidence of true imitative social learning in
feeding contexts. Also, we conformed with primate studies
of social learning by using vision as the sensory modality
in which we tested the elephants. Although we attempted to
account for the elephants’ rather poor visual acuity (Shyan-
Norwalt et al. 2010) by building large apparatus and sta-
tioning the observing subjects close to the model, it may be
the case that elephants rely primarily on their superior
olfactory and/or auditory senses to glean information in
social situations and are unused to visually scrutinizing
their groupmates’ behavior. Studies of social learning
should try to tap the most appropriate sensory modalities
for the species under investigation.
In sum, our data suggest that elephants may benefit from
social learning, but not necessarily in the form of imitating
the actions of conspecifics in foraging contexts. Adult ele-
phants may learn from one another primarily in non-imitative
ways such as observational conditioning. Ours is the first
study to attempt a controlled investigation of social learning
in elephants, and our data should be taken with some caution,
considering the small sample and variables that we could not
control, such as possible odor cues. In future research, it will
be important to consider the ecological validity of social
learning experiments with elephants.
Acknowledgments We are very grateful to the elephant training
staff at the San Diego Zoo Safari Park for their assistance, support,
and encouragement, and to Melissa Ritzer and Erin Lane for their
diligent coding of the videotapes. Dr. Doree Fragaszy generously
offered advice on an earlier version of this paper, as did two anon-
ymous reviewers.
References
Andrews J, Mecklenborg A, Bercovitch FB (2005) Milk intake and
development in a newborn captive African elephant (Loxodontaafricana). Zoo Biol 24(3):275–281. doi:10.1002/zoo.20048
Byrne RW, Bates LA, Moss CJ (2009) Elephant cognition in primate
perspective. Comp Cogn Behav Rev 4:65–79. doi:10.3819/ccbr.
2009.40009
Caldwell CA, Whiten A (2004) Testing for social learning and
imitation in common marmosets, Callithrix jacchus, using an
artificial fruit. Anim Cogn 7(2):77–85. doi:10.1007/s10071-003-
0192-9
Call J, Carpenter M (2002) Three sources of information in social
learning. In: Dautenhahn K, Nehaniv CL (eds) Imitation in
animals and artifacts. MIT Press, Cambridge, pp 211–228
Campbell FM, Heyes CM, Goldsmity AR (1999) Stimulus learning
and response learning by observation in the European starling, in
a two-object/two-action test. Anim Behav 58(1):151–158
Coussi-Korbel S, Fragaszy D (1995) On the relation between social
dynamics and social learning. Anim Behav 50:1441–1453
Dawson BV, Foss BM (1965) Observational learning in budgerigars.
Anim Behav 13(4):470–474
Dindo M, Thierry B, Whiten A (2008) Social diffusion of novel
foraging methods in brown capuchin monkeys (Cebus apella).
P Roy Soc Lond B Bio 275:187–193
Dindo M, Whiten A, de Waal FB (2009) Social facilitation of
exploratory foraging behavior in capuchin monkeys (Cebusapella). Am J Primatol 71(5):419–426. doi:10.1002/ajp.20669
Fragaszy D, Visalberghi E (1990) Social processes affecting the
appearance of innovative behaviors in capuchin monkeys. Folia
Primatol 54:155–165
Hart BL, Hart LA, Pinter-Wollman N (2008) Large brains and
cognition: where do elephants fit in? Neurosci Biobehav R
32(1):86–98. doi:10.1016/j.neubiorev.2007.05.012
Hermes R, Saragusty J, Schaftenaar W, Goritz F, Schmitt DL,
Hildebrandt TB (2008) Obstetrics in elephants. Theriogenology
70(2):131–144. doi:10.1016/j.theriogenology.2008.04.003
Heyes CM (1993) Imitation, culture and cognition. Anim Behav
46:999–1010
Hopewell LJ, Leaver LA, Lea SE, Wills AJ (2010) Grey squirrels
(Sciurus carolinensis) show a feature-negative effect specific to
social learning. Anim Cogn 13(2):219–227. doi:10.1007/s10071-
009-0259-3
Horner V, Whiten A, Flynn E, de Waal FB (2006) Faithful replication
of foraging techniques along cultural transmission chains by
chimpanzees and children. Proc Nat Acad Sci USA
103(37):13878–13883. doi:10.1073/pnas.0606015103
Huber L, Gajdon GK (2006) Technical intelligence in animals: the
kea model. Anim Cogn 9(4):295–305. doi:10.1007/s10071-
006-0033-8
Johnston TD (1981) An ecological approach to a theory of learning.
Behav Brain Sci 4:162–173
Krueger K, Heinze J (2008) Horse sense: social status of horses
(Equus caballus) affects their likelihood of copying other horses’
behavior. Anim Cogn 11:431–439
Krueger K, Flauger B, Farmer K, Maros K (2011) Horses (Equuscaballus) use human local enhancement cues and adjust
to human attention. Anim Cogn 14(2):187–201. doi:10.1007/
s10071-010-0352-7
Krutzen M, Mann J, Heithaus MR, Connor RC, Bejder L, Sherwin
WB (2005) Cultural transmission of tool use in bottlenose
dolphins. Proc Nat Acad Sci USA 102(25):8939–8943. doi:
10.1073/pnas.0500232102
Langbauer WR (2000) Elephant communication. Zoo Biol 19:
425–445
468 Anim Cogn (2013) 16:459–469
123
Moss C (1988) Elephant memories: thirteen years in the life of an
elephant family. Fawcett Columbine, New York
Oostindjer M, Bolhuis JE, Mendl M, Held S, van den Brand H, Kemp
B (2011) Learning how to eat like a pig: effectiveness of
mechanisms for vertical social learning in piglets. Anim Behav
82(3):503–511. doi:10.1016/j.anbehav.2011.05.031
Page RA, Ryan MJ (2006) Social transmission of novel foraging
behavior in bats: frog calls and their referents. Curr Biol
16(12):1201–1205. doi:10.1016/j.cub.2006.04.038
Ray ED, Heyes CM (2002) Do rats in a two-action test encode
movement egocentrically or allocentrically? Anim Cogn
5(4):245–252. doi:10.1007/s10071-002-0154-7
Rendell L, Whitehead H (2001) Culture in whales and dolphins.
Behav Brain Sci 24(2):309–382
Schuster S, Wohl S, Griebsch M, Klostermeier I (2006) Animal
cognition: how archer fish learn to down rapidly moving targets.
Curr Biol 16(4):378–383. doi:10.1016/j.cub.2005.12.037
Shyan-Norwalt MR, Peterson J, Milankow King B, Staggs TE, Dale
RH (2010) Initial findings on visual acuity thresholds in an
African elephant (Loxodonta africana). Zoo Biol 29(1):30–35.
doi:10.1002/zoo.20259
Stowe M, Bugnyar T, Loretto MC, Schloegl C, Range F, Kotrschal K
(2006) Novel object exploration in ravens (Corvus corax):
effects of social relationships. Behav Process 73(1):68–75. doi:
10.1016/j.beproc.2006.03.015
Sukumar R (2003) The living elephants: evolution, ecology, behavior,
and conservation. Oxford University Press, New York, NY
Tomasello M, Call J (1994) Social cognition of monkeys and apes.
Yearb Phys Anthropol 37:273–305
Voelkl B, Huber L (2000) True imitation in marmosets. Anim Behav
60(2):195–202. doi:10.1006/anbe.2000.1457
Whiten A, Custance DM, Gomez J, Teixidor P, Bard KA (1996)
Imitative learning of artificial fruit processing in children (Homosapiens) and chimpanzees (Pan troglodytes). J Comp Psychol
110(1):3–14
Whiten A, Horner V, Litchfield C, Marshall-Pescini S (2004) How do
apes ape. Learn Behav 32(1):36–52
Zentall TR (2004) Action imitation in birds. Learn Behav 32(1):
15–23
Zentall TR (2006) Imitation: definitions, evidence, and mechanisms.
Anim Cogn 9(4):335–353. doi:10.1007/s10071-006-0039-2
Zentall TR, Sutton JE, Sherburne LM (1996) True imitative learning
in pigeons. Psychol Sci 7(6):343–346
Anim Cogn (2013) 16:459–469 469
123