Bonobos Extract Meaning from Call Sequences
Zanna Clay1,2, Klaus Zuberbühler1*
1 School of Psychology, University of St Andrews, St Andrews, Scotland, United Kingdom, 2 Twycross Zoo, Twycross, Atherstone, United Kingdom
Abstract
Studies on language-trained bonobos have revealed their remarkable abilities in representational and communication tasks.
Surprisingly, however, corresponding research into their natural communication has largely been neglected. We address
this issue with a first playback study on the natural vocal behaviour of bonobos. Bonobos produce five acoustically distinct
call types when finding food, which they regularly mix together into longer call sequences. We found that individual call
types were relatively poor indicators of food quality, while context specificity was much greater at the call sequence level.
We therefore investigated whether receivers could extract meaning about the quality of food encountered by the caller by
integrating across different call sequences. We first trained four captive individuals to find two types of foods, kiwi
(preferred) and apples (less preferred) at two different locations. We then conducted naturalistic playback experiments
during which we broadcasted sequences of four calls, originally produced by a familiar individual responding to either kiwi
or apples. All sequences contained the same number of calls but varied in the composition of call types. Following
playbacks, we found that subjects devoted significantly more search effort to the field indicated by the call sequence.
Rather than attending to individual calls, bonobos attended to the entire sequences to make inferences about the food
encountered by a caller. These results provide the first empirical evidence that bonobos are able to extract information
about external events by attending to vocal sequences of other individuals and highlight the importance of call
combinations in their natural communication system.
Citation: Clay Z, Zuberbühler K (2011) Bonobos Extract Meaning from Call Sequences. PLoS ONE 6(4): e18786. doi:10.1371/journal.pone.0018786
Editor: Martine Hausberger, University of Rennes 1, France
Received August 20, 2010; Accepted March 18, 2011; Published April 27, 2011
Copyright: ß 2011 Clay, Zuberbühler. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This research was funded by a Leverhulme Trust Research Leadership Award and the Wissenschaftskolleg zu Berlin. The funders had no role in study
design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: kz3@st-andrews.ac.uk
which can be encoded by changes in call rates [12,15–17]
or acoustic structure [18–19]. For example, Rhesus monkeys
(Macaca mulatta) produce up to five different food-associated call
types [20], which receivers discriminate based on differences in
their referential features, rather than their acoustic properties
alone [21].
Among the great apes, chimpanzees (Pan troglodytes) produce
specific calls when discovering food, the ‘rough grunts’ [22–23].
The morphology of this call type co-varies with the caller’s
personal food preference both in captivity and the wild [24]. In a
naturalistic playback experiment, it was demonstrated that
acoustic variation in this call influences the foraging decisions of
receivers, suggesting that the acoustic structure of this graded
signal provides meaningful information to other chimpanzees
about the quality of food encountered by the caller [25].
What exactly governs receiver responses, however, is a matter of
ongoing debate. For instance it is not clear whether receivers
respond directly to the calls’ physical features or their referential
nature, i.e. the causal relation between calls and contexts [26–27].
Similarly, signalling is often said to be non-cooperative with
signallers merely producing ‘natural’ information in response to
biologically relevant events, while any representational content is
largely generated by the listeners [28]. These problems are
unsolved because the psychological states experienced by primates
during call production and perception are rarely investigated. In
one recent study, however, food call production in wild
chimpanzees was found to co-vary with the presence and arrival
of long-term allies, suggesting these calls may be used as a flexible
social strategy [10].
Introduction
A growing body of research on the communicative behaviour of
non-human primates has demonstrated that their vocalisations can
convey a considerably rich amount of information that is
meaningful to receivers (e.g. [1]). For instance, field experiments
with various primate species have shown that acoustically distinct
alarm calls can inform listeners about specific types of dangers (e.g.
[2–5]). In some species, there is evidence that signallers produce
strings of acoustically variable calls composed in context-specific
ways (e.g. [6–8]). For example, black-and-white Colobus monkeys
(Colobus polykomos, C. guereza) produce two types of vocalisations to
predators, which are arranged in event-specific sequences that are
seemingly meaningful to others [8].
Food discovery is another event type during which some
primates produce highly context-specific vocalisations. Since food
is often patchily distributed and seasonally dispersed, food calls
can provide listeners with a useful means to access foraging
patches more effectively, while callers appear to gain mainly
social benefits [9–10]. The production of food-associated calls is
not restricted to primates but found in other mammals and some
birds, e.g. Gallus gallus [11], although relatively little is still known
about the type of information conveyed by the calls. At the
simplest level, food calls are a basic physiological response
indicating that the caller has found something desirable, as
demonstrated by receivers approaching food calls more rapidly
than other calls [12–13] or by triggering foraging behaviour [14].
In some species, food calls appear to provide more detailed
information about the food item itself, such as its quality or divisibility,
PLoS ONE | www.plosone.org
1
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
mixed sequences. The composition of these food call sequences
related reliably to food quality, suggesting that listeners can gain
information from attending to the call sequences. The hypothesis
of meaningful call combinations has been put forward before for
bonobos, but has never been tested formally [32–33].
In the current study, we examined whether listeners were able
to extract meaningful information relating to food quality by
attending to the composition of these heterogeneous call
sequences. To this end, we conducted naturalistic playback
experiments in which subjects heard different types of food all
sequences, and their subsequent foraging responses were analysed.
Very little is known about how our other closest relative, the
bonobo (Pan paniscus), naturally communicates about events in the
external world. This is despite the fact that some individuals have
been remarkably capable in mastering artificial language systems
[29–30]. In one recent systematic study, Clay & Zuberbühler [31]
demonstrated that bonobos also vocalize upon encountering food,
but that there are important differences between the two Pan
species. Whilst both species produce food ‘grunts’, bonobos
additionally give four acoustically distinguishable tonal calls when
finding food (barks, peeps, peep-yelps, and yelps), which although
lying on a graded scale, can be statistically discriminated from one
another [31]. Barks are longest in duration, characterised by a
distinctive pointed shape and numerous visible harmonic bands.
Whilst peeps are also high pitched, they are temporally shorter
than barks (and all other calls), with few harmonic bands and had
a flat frequency contour. Yelps and peep-yelps are lower in pitch
and although share acoustically similarities, yelps possess a marked
downward stroke frequency contour, in contrast to the arched
contour of the peep-yelps. In terms of production, peeps and barks
are most frequently, but not exclusively, given to preferred foods,
while yelps and grunts are more often, but not exclusively, given to
less preferred foods. Peep-yelps are produced broadly, although
they also tend to occur more to mid- and lesser-preferred foods. In
sum, the link between individual call types and perceived food
quality is only probabilistic in bonobos. One important consequence of this is that different food calls themselves do not appear
to allow listeners to make strong predictions about the type of food
encountered by the caller. However, in contrast to chimpanzees,
bonobos regularly combine different call types together into longer
Results
The study was conducted at Twycross Zoo, UK, between April
and July 2009. Individuals were permanently separated into two
subgroups that occupied different indoor facilities but shared the
same outdoor area via two separate doors in the morning or
afternoon, respectively (fig. 1; table 1). We conducted playback
experiments in which we broadcast food-calls of an individual of
the morning subgroup responding to either high-quality food (kiwi)
or low-quality food (apple) to all members of the afternoon
subgroup before releasing them into the outdoor enclosure. Our
general aim was to simulate a morning group individual
discovering food of high or low quality in the outdoor area shortly
before the release of the afternoon subgroup. We also included a
control condition in which no food calls were played. We then
assessed the foraging behaviours of receivers at previously learned
locations for high-quality and low-quality food in the outdoor
enclosure (fig. 1).
Figure 1. Schematic layout of the bonobo facility at Twycross Zoo, including location of playback equipment and artificial food
sites.
doi:10.1371/journal.pone.0018786.g001
PLoS ONE | www.plosone.org
2
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
Finally, hearing playbacks of food-associated call sequences had
a significant effect on the foraging time devoted by the group at
both the kiwi site (x2 = 6.902, df = 2, p = .026; two-tailed exact
Kruskal-Wallis test) and the apple site (x2 = 10.876, df = 2,
p = .002; two-tailed exact Kruskal-Wallis test; fig. 2c). Pair-wise
comparisons (Bonferroni corrected alpha = .0169) revealed that
individuals spent more time at the kiwi location after hearing ‘kiwi’
call sequences compared to control condition (mediancontrol =
2.25s; mediankiwi = 16.50s; medianapple = 5.75s; Kiwi site, ‘kiwi’
playback vs. control: Mann-Whitney U = 5, p = .022,) or hearing
‘apple’ call sequences (‘kiwi’ playback vs. ‘apple’ playback:
U = 15.5, p = .058). Likewise, individuals spent more time in the
apple field after hearing playbacks of ‘apple’ call sequences
compared to control trials (mediancontrol = 0.0s; median apple =
9.5s; mediankiwi = 1.5; apple field, ‘apple’ playback vs. control:
Mann-Whitney U = 6, p = .015). Although there was a trend for
spending more time foraging at apple after hearing apple
playbacks compared to kiwi playbacks, the result did not reach
significance (apple vs. kiwi playbacks: U = 20, p..05).
All previous analyses were based on non-parametric comparisons at the group level. We selected this analysis strategy to avoid
problems with data interdependency and type-two clustering
errors [34] at the cost of a substantial reduction in statistical
power. Thus, in a second set of analyses we relied on Generalized
Linear Mixed Models based on Poisson distributions and log link
functions (GLMM), which allowed us to maximise the amount of
data available by considering the contributions of individuals. We
accounted for individual identity by entering it as a random factor,
although it may be argued that combining individuals into one
model will not completely eliminate the problem of interdependency, The GLMM procedure is an extension of the more widely
known General Linear Model (GLM) and is particularly useful for
this study because it relaxes assumptions of normal data
distribution and an identity link [35].
This GLMM analysis revealed the same main effects, with a
significant interaction between playback type and the number of
visits to the two fields (two-tailed GLMM, F (2,178) = 5.037,
p = .007; fig. 2b; table S2). Pair-wise comparisons revealed that
individuals visited the kiwi field significantly more often after
hearing ‘kiwi’ call sequences compared to control trials (p = 0.028).
They also visited the apple field significantly more often after
hearing ‘apple’ call sequences compared to control and kiwi
conditions (p,.001; p = 0.008, respectively).
Finally, hearing playbacks of food-associated call sequences had
a significant effect on the foraging time devoted by each of the four
individuals in the two fields (two-tailed GLMM: F (2, 178) =
120.772, p,.001; fig. 2c, table S3). Pair-wise comparisons
revealed that individuals spent more time at the kiwi location
after hearing ‘kiwi’ call sequences than ‘apple’ call sequences or
compared to control trials (both p,.001). Likewise, individuals
spent more time in the apple field after hearing playbacks of
‘apple’ call sequences compared to ‘kiwi’ call sequences or
compared to control trials (both p,.001).
Table 1. Composition of the two bonobo subgroups at
Twycross Zoo (UK) in April 2009.
Subgroup A (call producers)
Subgroup B (call receivers)
Name
Name
ID
Sex
Age
ID
Sex
Age
Kakowet
KT
M
07.06.1980
Diatou
DT
F
21.10.1977
Banya
BY
F
16.02.1990
Kichele
JS
M
02.08.1980
Keke
KK
M
02.01.1994
Jasongo
KH
F
19.04.1989
Maringa
MR
F
05.05.1998
Cheka
CK
F
18.03.1996
Bokela
BK
F
14.10.2003
Luo
LU
M
01.12.2002
Gemena
GM
F
07.11.2005
doi:10.1371/journal.pone.0018786.t001
Hearing food-associated calling sequences influenced
foraging behaviour
Following release, there was a strong baseline preference for the
highly preferred ‘kiwi’ field. This was particularly evident in the
control condition (when no food was presented), in which
individuals were more likely to visit the kiwi field first and more
often, as well as devoting more foraging effort to it compared to
the apple field (fig. 2). Despite this baseline bias, playbacks of foodassociated calls had an overall significant effect on the individuals’
first choice of fields (x2 (2) = 16.347, p,.001; Pearson chi-square,
two-tailed; fig. 2a). Playback of call sequences originally given to
kiwi resulted in an increase in first visits to the kiwi field, compared
to baseline or apple trials. However, this change failed to reach
significance due to a ceiling effect caused by the strong baseline
bias for kiwi (First arrivals to kiwi site per individual: median N
trials: control condition = 3.0 (50% of trials); kiwi playback
condition = 6.0 (86%); apple playback condition = 5.0 (50%), all
one-way x2 tests: p..05). Playback of call sequences originally give
to apple resulted in a significant increase in the number of first
visits to the apple field, compared to baseline or kiwi trials (First
arrivals to apple site, per individual: median N trials: control
condition = 0.0; kiwi playback condition = 0.0; apple playback
condition = 4.0 (40%); both control and kiwi vs. apple:
x2 (1, N = 17) = 13.235, p,.001, with Bonferroni corrected alpha
.0169). Hearing food-associated call sequences, in other words,
influenced the bonobos’ foraging decisions against their preexisting food preference biases. We conducted this first analysis at
the group level because individuals almost always foraged as a
cohesive unit and, when released, entered the outside enclosure
almost simultaneously (table S1).
Next, we determined whether hearing playbacks influenced the
number of visits the group made to the two fields (fig. 2b). Again,
we found a significant effect of playback condition on the median
number of visits made by the group to both the kiwi field
(x2 = 6.486, df = 2, p = .034; two-tailed exact Kruskal-Wallis test)
and the apple field (x2 = 10.532, df = 2, p = .002; two-tailed exact
Kruskal-Wallis test). Post-hoc, pair-wise comparisons using a
Bonferroni correction (corrected alpha = .0169) revealed that
individuals visited the ‘kiwi’ field more often after hearing ‘kiwi’
call sequences compared to control trials (mediancontrol = 0.5;
mediankiwi = 1.0; medianapple = 1.0; N visits to kiwi field, ‘kiwi’
playback vs. control: Mann-Whitney U = 4.5, p = .015). Conversely, we found that individuals visited the apple field more
often after hearing playback of ‘apple’ call sequences compared to
the control condition (mediancontrol = 0.0; mediankiwi = 0.5; medianapple = 1.0; N visits to apple field, ‘apple’ playback vs control:
U = 3, p = .002).
PLoS ONE | www.plosone.org
Foraging errors and integration across the call sequence
A key indicator of representationally-based signal processing is
that subjects sometimes make mistakes, particularly with signals
that are ambiguous or only weakly correlated with specific external
events [28]. In our sample, some call sequences were better
indicators of high or low food quality than others in terms of call
composition, suggesting that if subjects made mistakes then this
should happen in response to the more ambiguous sequences (e.g.
visiting the apple field after hearing a kiwi sequence). To address
this, we assigned a cumulative value to each sequence, based on its
3
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
Figure 2. Box plots indicating foraging responses of bonobos (N = 4) following playbacks of food-associated calls given to high
value (kiwi) or low value (apple) foods. (a) Site of first entry expressed as a median proportion of the individual’s median choices per condition;
(b) median number of visits per trial; (c) median time spent foraging following playback (s). Box plots illustrate medians, inter-quartile ranges, and
highest and lowest values, excluding outliers.
doi:10.1371/journal.pone.0018786.g002
call composition. Each call contributed with a value that reflected
its association strength with high preference food (table 2). For
instance, in natural calling sequences, ‘barks’ were given six times
more frequently to high than low preference food (proportion of
‘barks’ in calling sequences to high preference foods = 0.24, vs.
low-preference foods = 0.04). Similarly, peeps were given 1.86
times more frequently to high than low preference food
(proportion of ‘peeps’ in calling sequences to high preference
foods = 0.52 vs. low-preference food = 0.28), and so on. These
relative differences resulted in the following cumulative values:
B = 6.00, P = 1.86, PY = 0.52, Y = 0.12. We also assigned an
ordinal scale, based on the strength of their relationship to highpreference foods, with ‘barks’ being given most frequently to highpreference foods and yelps most infrequently: B = 4, P = 3, PY = 2,
Y = 1 (see fig. 3, table 3).
There was a significant positive correlation between subjects’
foraging effort in the kiwi field and the overall cumulative food value
as assessed by the composition of the sequence (time spent: cardinal
PLoS ONE | www.plosone.org
scale Spearman’s rho: N = 17 rs = 0.585, p = .014, fig. 3; ordinal
scale: N = 17, rs = 0.575, p = .016). Inspection at the level of
individual trials indicated an almost perfect separation of sequences
given to apples and kiwis by the cumulative sequence value
generated by the constituent calls. One exception was a call
sequence given to apples (PY-B-B-PY), which interestingly also
triggered almost twice as much searching in the (wrong) kiwi
compared to the apple field. Also interesting were two responses to
kiwi sequences, which only triggered weak searching in the kiwi
field. However, in both cases, search effort in the apple field was also
unusually low, suggesting that subjects were generally unmotivated
to forage (table 3). In sum, the foraging effort was a strong reflection
of the cumulative ‘good food’ score encoded by the sequence.
Discussion
Human-enculturated bonobos have long been known for
their extraordinary representational and communication skills
4
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
[29–30,36–37], but their natural communication behaviour has
hardly been investigated. Our study provides progress to this end
in showing that bonobos can increase their foraging success by
attending to each other’s call sequences. Our key finding was that
subjects were able to direct their foraging effort to specific
locations according to the call sequence presented to them. Whilst
we found an unsurprising baseline preference to the highpreference food site, we found that playbacks of high-preference
food call sequences resulted in an even greater amount of foraging
effort at this site, indicating the calls were meaningful to the
receivers. Furthermore, although lack of interest at the low-food
preference apple site was to be expected (see baseline trials) we
found a significant increase in search effort at this site that only
occurred after hearing sequences associated with low-preference
food. This result further suggests that individuals incorporated
information extracted from the food call sequences to optimise
their foraging strategies, in some cases against pre-existing foraging
biases.
Although it would be interesting to investigate the mechanism of
decision making within the group, we were not able to address this
point in our study due to the constraints of working with subjects
that were part of a public display (with Zoo regulations prohibiting
separation). It is likely that there were individual differences in the
ability to assign meaning to the different call sequences used as
playback stimuli, either due to individual differences in knowledge
or other more immediate factors, such as receiver attention or
motivation during the trials. Despite these constraints, we are able
to draw the conservative conclusion that at least one individual in
Table 2. Relative frequency of food-related call types within
natural call sequences given to high and low-value foods by
bonobos at Twycross Zoo and the corresponding playback
stimuli.
Sequence
Natural
Playback
Food
type
Call
type
Bark
Peep
Peepyelp
Yelp
Grunt
High
0.24
0.52
0.22
0.03
0.00
Low
0.04
0.28
0.42
0.26
0.02
High
0.29
0.50
0.18
0.04
0.00
Low
0.05
0.13
0.40
0.43
0.00
The values are calculated for the first four calls only; with mean values produced
by three individuals in subgroup A (KK, KT, BK) and three other individuals in
subgroup B (DT, CK, KH). Results mirror what was previously described from a
larger data set from two groups in San Diego (31) showing that although all
calls were produced at least once in the high and low contexts, their relative
frequency varies with food preference. Barks are produced most typically in
response to high value foods; peeps occur often to high value foods, peepyelps are produced more often to medium-lower value foods, while yelps were
typically produced the lower value foods. Grunts were rarely produced in
natural call sequences at Twycross, with only 2 individuals producing them,
both for low value foods.
doi:10.1371/journal.pone.0018786.t002
Figure 3. Scatter plot showing the relationship between the food type encountered (highly preferred = kiwi, less preferred = apple)
and the cumulative value of the stimuli sequence. Calls were assigned a cardinal score based on how frequently they were produced in
response to high vs. low value food (e.g. barks were six times more frequent to high than low preference foods, so that: B = 6.00, P = 1.86, PY = 0.52,
Y = 0.12). The relationship between foraging responses and stimuli scores are indicated in table 3.
doi:10.1371/journal.pone.0018786.g003
PLoS ONE | www.plosone.org
5
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
Whilst there is a growing body of evidence that numerous
monkey species produce strings or sequences of acoustically
variable calls composed in context-specific ways, evidence for
meaningful signal combinations in apes has been poor (although
see [38]). A recent study of gorilla gestural sequences failed to find
evidence of syntactic organisation or corresponding semantic
content [39]. Results from the current study provide the first
empirical evidence that call combinations do play a role in bonobo
communication in the foraging context. However, it is important
to note that we also did not find any evidence for syntactic rules or
that the sequencing structure itself was itself semantically relevant.
Thus, although call combinations appear to represent a useful
means of communicating information in bonobos, the manner in
which bonobos use call combinations contrasts the way linguistic
units are combined and structured in human language. This
finding highlights the importance of studying non-human primate
communication as a means to identify the features of the language
faculty that are uniquely human.
One recurrent topic in the animal communication literature is
whether signals given in response to external events, such as in this
study, should be conceptualised as ‘referential’ or a mere readout
of a caller’s motivational or internal state [26,40–41]. Great apes,
especially chimpanzees and bonobos, are often described as
exceedingly ‘emotional’, suggesting that arousal-based explanations may be more in line with the nature of the phenomenon
described here (e.g. [26]). In particular, sequences containing a
greater amount of calls with presumably high emotional valence
may lead receivers to search at the high-value food site. Although
we do not discount the integral role motivation plays in animal
communication [26,41], gaining meaningful measurements of
internal state or arousal have so far proved very challenging, and
thus it has often proved more empirically fruitful to focus on the
relation between receiver response and external variables that can
be manipulated and measured experimentally [1,42]. Furthermore, even calls with presumably high motivational content (as
may be the case for food discovery) are still able to inform receivers
about the external world. This has been demonstrated by studies
showing that, regardless of the signaller’s motivational state during
call production, calls can provide listeners with representational
information about external objects and events, in way that can be
studied experimentally [1,27,41,43–45]. Recent work on the alarm
call responses of meerkats (Suricata suricata) for example, has
demonstrated that both emotional and referential information are
encoded into the same signal and develop on different ontogenetic
time scales [45]. Meaningful progress will focus more specifically
on the motivational experience of the caller and how this
influences signal production.
How do receivers extract information from these call sequences?
Are they attending to individual calls or do they perceive the
sequence as a holistic unit? For example, it could be argued that an
increasing number of high-pitched calls, such as peeps and barks,
increases the perceived gestalt of the sequence in a discrete way,
enabling individuals to make foraging decisions without paying
attention to individual calls. Further research will be necessary to
address this issue in more detail.
Another related question is whether receivers process vocalisations purely based on their acoustic properties, or whether they
attach some semantic value to them. For example, subjects may
have learned the contingencies of this particular experiment, e.g.
that high-pitched vocalisations were associated with food at one
specific location, without any understanding of more general
relations between a caller’s perceived food quality and the vocal
signals produced in response. However, research on foodassociated calls of rhesus macaques (Macaca mulatta), another
Table 3. Composition of different call stimuli and resulting
behavioural responses in receivers.
Signaller behaviour
Receiver foraging effort (s)
Food
Sequence
CVO CVC
Kiwi field
Apple field Kiwi bias
Kiwi
BBPB
15
19.86
21.0
2.5
Kiwi
BBPB
15
19.86
6.5
2.5
3.6
Kiwi
B B P PY
13
14.38
28.3
5.0
6.7
9.4
Apple
PY B B PY
12
13.05
20.0
12.0
2.7
Kiwi
P P PY P
11
6.10
79.0
18.8
5.2
Kiwi
PY P P P
11
6.10
20.8
1.8
12.6
Kiwi
P PY PY P
10
4.76
1.3
2.5
1.5
Kiwi
P P PY Y
9
4.35
16.5
6.5
3.5
Apple
PY P PY PY
9
3.43
14.8
3.3
5.5
Apple
Y PY PY P
8
3.02
0.0
15.8
1.0
Apple
Y PY P Y
7
2.61
9.3
2.0
5.7
Apple
PY PY Y PY
7
1.69
5.8
20.8
1.3
Apple
YPYY
6
2.20
6.5
14.3
1.5
Apple
YYYP
6
2.20
3.8
40.8
1.1
Apple
PY PY Y Y
6
1.28
9.5
9.0
2.1
Apple
PY Y PY Y
6
1.28
6.5
17.3
1.4
Apple
PY Y Y Y
5
0.87
2.5
10.3
1.2
Receiver foraging effort represents mean time spent foraging at per individual.
Cells marked in bold represent ’response errors’ where individuals exerted more
foraging effort in the incongruent field. Details of how cumulative values were
calculated are found in Figure 3. (B = bark, P = peep, PY = peep yelp, Y = yelp).
CVO: Cumulative value (ordinal); CVC: Cumulative value (cardinal); Kiwi bias:
Relative bias towards the kiwi field.
doi:10.1371/journal.pone.0018786.t003
our group was able to comprehend the information conveyed by
the call sequences, though the results of some trials suggested that
several or all group members made their own independent
foraging decisions prior to being released into the enclosure. For
example, we found that, across trials, different individuals arrived
at the food sites first and, in some trials, individuals diverged in
their first choices (table S1).
Our results also show that, although chimpanzees and bonobos
are phylogenetically closely related, they appear to communicate
about food in considerably different ways. In contrast to
chimpanzees, who produce an acoustically graded call type that
co-varies with food quality [26], bonobos regularly mix several
acoustically distinct food call types into heterogeneous strings of
vocalisations. Rather than at the level of individual calls, food
quality appears related to the probabilistic composition of
heterogeneous call sequences [31]. Results from our playback
experiment indicate that rather than attending to individual call
types, receivers took into account the relative proportions of
different calls within the sequence and extracted meaning by
integrating information from across the call units.
In addition, the generation of more foraging errors in
structurally ambiguous call sequences (which were less strongly
indicative of high or low preference foods) indicated that the
information extracted from the stimuli sequences was influencing
the foraging decisions of the receivers. In a recent discussion,
Stegmann [28] argued that in contrast to ‘natural information’,
which does not allow for errors, the generation of misrepresentations and errors is a defining feature of what we consider as
‘semantic information’ in animal signals.
PLoS ONE | www.plosone.org
6
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
1. Food preference tests. First, we conducted food
preference tests to identify a highly valued and a lesser-valued
food. Equal sized piles of two foods were placed next to each other
on the ground and the first choice was recorded for each
individual, repeated across four days, once per day. For each
individual, we created a preference matrix based on the
percentage of trials in which each food was selected over the
other food type across the four sessions. These percentages were
then combined to give a cumulative preference score per food, and
ranked accordingly (table S4). From eight familiar foods, kiwis and
apples were consistently ranked as high and low by all individuals,
while both still reliably triggering food calls. We thus selected kiwis
and apples as our experimental foods.
2. Recording calls. From April to May 2009, we recorded
food-associated call sequences given by all individuals feeding in
the outdoor enclosure (fig. 4). This allowed us to build up a sound
library of call sequences given to kiwi and apples by individuals of
the ‘morning’ subgroup for the subsequent playback experiments
and to compare their behaviour with a previous study on bonobos
[31] (table 2). We recorded vocalizations using a SENNHEISER
MKH816T directional microphone and a MARANTZ PMD660
solid-state recorder. Verbal comments were given and later
transcribed. We transferred recordings onto a TOSHIBA Laptop
(Equium 1.8 GHz) at a sampling rate of 44.1 kHz with 16-bit
accuracy. To control for hunger levels, novelty, and other factors,
we only recorded calls produced during first morning feeds. We
excluded calls produced by individuals interacting with more than
one type of food or when caller identity was uncertain. Calls were
recorded from a range of locations from a distance of 2–15 m.
3. Foraging training. During this same period, we
established two outdoor foraging patches for the afternoon
subgroup in a daily foraging task. Before their midday release, a
caretaker hid finely cut pieces (1 cm2 pieces, total 300 g) of either
apple or kiwi in the grass in one of two 30 m2 fields so that they
were not visible from a distance. The two fields were on slopes,
equidistant to the door (21 m); the distance between them was
8 m. Both field areas were equal (length top = 6.5 m;
width = 4.0 m; length bottom = 8.5 m), starting with a flat
descent and finishing at the concrete border of the enclosure
wall. Kiwi and apple feeds were presented in random order so
individuals could not predict which patch was baited so had to
inspect the two areas separately. Only one food type was ever
provided and no other food or enrichment was provided during
training. The keeper always visited both areas, even if no food was
placed, to prevent individuals from learning noises associated with
scattering food. There were 16 training days for each food type,
and 10 control days during which no food was provided. We
filmed the individuals’ foraging behaviour and kept a daily record
of each individual’s food encounters (text S1, table S5). Individuals
soon learned the two feeding fields and quickly formed a
preference for the kiwi field.
4. Playback experiments. In the final step, we conducted
playback experiments in which call sequences of members of the
morning subgroup were played to individuals in the afternoon
subgroup (text S1). The experimental routine was as follows.
Around midday, the morning subgroup was brought inside for a
seed feed. Live radio broadcasting was played via an inside keeper
door to prevent subjects from the morning group from hearing the
stimuli (i.e. their own calls) in the subsequent experiment. This was
effective as no vocal responses were elicited from any individual
during playback trials (except for one apple trial, which was
excluded from analysis). Meanwhile, individuals from the
afternoon subgroup were waiting to be released through their
own door. Beforehand, three key manipulations were carried out.
primate with a graded vocal system, has demonstrated that
individuals categorise calls based on their meaning, not just their
acoustic structure alone [31]. Whether bonobos process their own
calls in the same way remains open as a topic for future research.
A final unresolved question concerns the function of food call
production in bonobos. In capuchins (Cebus capucinus), food calls
are thought to provide ecological benefits, functioning to
announce food ownership and a willingness to defend, thereby
resulting in reducing foraging competition from others [46]. In
red-bellied tamarins (Saguinas labiatus), it has been suggested that
food calls are not solely a function of arousal in the presence of
highly desirable food patches, but may provide social benefits by
attracting allies, even at the cost of increasing feeding competition
[9]. A similar effect has recently been indicated in wild male
chimpanzees who were found to call more in the presence of close
allies and even recommenced calling upon their arrival [10]. For
bonobos, it has also been suggested that individuals receive
benefits from producing food calls, for instance by attracting mates
or potential allies [47]. Further work investigating the interplay
and influence of social and ecological variables on the production
of food-associated calls in bonobos is required to explore the
adaptive significance of these calls in this species.
Materials and Methods
Study site and subjects
The study was conducted at Twycross Zoo, UK, between April
and July 2009.Individuals were permanently separated into two
subgroups that occupied separate indoor facilities but shared the
same outdoor area via two separate doors (fig. 1). Subgroup A
consisted of 5 individuals (2 adult males and females, 1 juvenile
female; range 6–29 years); subgroup B consisted of 6 individuals (1
adult male, 3 adult females, 1 juvenile male, 1 juvenile female;
range 4–32 years; table 1). Each subgroup was housed in one of
two separated heated indoor halls (62 m2) with additional sleeping
areas (22 m2) and both facilities were separately connected to an
outdoor enclosure (588 m2). There was no visual contact between
indoor and outdoor enclosures, although vocalizations produced
outside could be heard indoors. Both subgroups were fed a range
of fruits and vegetables twice per day in a scatter feed. Water was
freely available. Bonobos were provided with regular enrichment
materials and feeds (such as branches, seeds, grapes or frozen juice)
as well as supplements such as yoghurt, egg and bread.
Ethical statement
The Twycross Zoo Ethics and Management Committee and the
Zoo Research Coordinator and gave full ethical approval to this
behavioural, non-invasive study, which complied with the ethical
guidelines set out by the British and Irish Association of Zoos and
Aquariums (BIAZA). During all stages of the study, we took steps
to ensure that the welfare of all animals was not compromised. No
individual showed distress during any part of this study and their
participation throughout was voluntary. In order to reduce stress
and to comply with Zoo guidelines, we did not separate any
individual in any stage of this study.
Experimental Design
The basic design was to simulate a member of the morning
subgroup A finding food shortly before the midday switchover, in
order to investigate whether this influenced the subsequent
foraging behaviour of subgroup B members. The study consisted
of four stages: (1) food preference tests, (2) recording of foodspecific calling sequences, (3) establishment of two feeding areas,
and (4) playback experiments.
PLoS ONE | www.plosone.org
7
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
Figure 4. Spectrographic illustrations of two playback stimuli. (a) High value sequence originally given to kiwi consisting of bark/peep/bark/
peep and (b) low value sequence originally given to apple consisting of peep-yelp/peep-yelp/yelp/yelp. Recordings of the corresponding call
sequences are available as audio S1 and audio S2.
doi:10.1371/journal.pone.0018786.g004
The remaining two individuals (DT, JS) were excluded. JS did
not complete the training phase and was not motivated to go
outdoors due to his low social rank. DT showed no evidence of
having learned the food locations during the training phase and
showed little interest during the playback phase, only completing 5
of 23 trials, not enough for statistical analyses.
We extracted systematic data on three dependent variables across
the different conditions: (a) patch first visited (kiwi vs. apple), (b) time
spent actively foraging in each patch (time trespassing, sitting,
resting, or sleeping were subtracted), (c) total number of visits per
patch (N times entering and exiting the patch areas interrupted by at
least one bout of foraging). Because data from individuals were
interdependent (we were unable to separate individuals so all
individuals foraged together), we conducted our principal analyses
at the group level using the median scores for individuals combined
per trial. The nature of the data distribution for this method meant
that only non-parametric statistics were employed.
Whilst measuring the central tendency of the group across trials
reduces the problems of interdependency of the data and type-two
clustering errors [34], the cost is a substantial reduction in
statistical power. Furthermore, rather than using the foraging
behaviours of receivers (upon which our hypothesis is based) as the
unit of analysis, the unit of analysis becomes the trials in which the
responses of a groups of receivers were measured. We therefore
conducted a second analysis using Generalized Linear Mixed
Models (Poisson distribution-log link) in order to address the
problem of statistical power. The Generalised Linear Mixed
Model Procedure (GLMM) is an extension of the General Linear
Mixed Model, characterised by a flexible generalization of
ordinary least squares regression. The procedure relaxes the
assumptions of normal distribution and identity link [35], a crucial
prerequisite for the ordinal data of this study. The GLMM
procedure enables individual identity to be accounted for (by
entering it as a random factor), although this does not completely
address the problem of potential inter-dependency in the
individuals’ foraging decisions.
First, a keeper entered the outdoor enclosure from a side door to
mimic placing food (none was provisioned). Individuals were
familiar with this routine from the previous foraging training.
They could not see the event, but could hear the associated
sounds. After the keeper’s return, subjects heard the opening
sounds of the door, which connected the morning subgroup to the
outdoor enclosure (to suggest a re-entry of the morning subgroup),
although, in reality, no subject was released. A trial was conducted
only if (a) no vocalizations had been produced by the morning
subgroup for at least 1 minute, (b) individuals of the afternoon
subgroup were waiting close (,1–2 m) to the door and not
distracted by social activities (play, agonistic, sex) for at least 1
minute; (c) there was no rain or excessive wind outdoors.
Communication between a keeper who stayed indoors with the
bonobos and the experimenter, who stayed outdoors, was
maintained with two-way radios. If these conditions were not
met, the trial was either delayed or, in some cases, abandoned. If
conditions were met, the afternoon subgroup then heard a 4 s
playback of a series of four equally spaced calls extracted from a
natural call sequence to either apple or kiwi (to simulate a morning
subgroup member finding apple or kiwi) played from their outdoor
enclosure (fig. S1). During control trials, all features of the
procedure remained the same except that no stimulus was played.
One minute after the playback (a sufficient time period for
morning subgroup to ‘return’ indoors), the afternoon subgroup
was released and their foraging behaviour was monitored for up to
10 min using a camcorder with additional verbal comments. We
simultaneously recorded all vocal responses with professional
sound recording equipment as previously described. During
experimental and control trials, no food was ever provided on
either field (to rule out visually-based foraging). To reduce the
possibility of extinction, we interspersed a number of refresher
days between each trial, i.e. between 1–4 days during which we
provided either kiwi or apple pieces on the corresponding fields in
random order (N = 28 total).
Zoo regulations prohibited separation of group members (due to
stress provoked) so all individuals were released simultaneously
into the outdoor enclosure and behavioural response measures
were collected while individuals interacted with each other as a
group. A total of 28 trials were conducted; three were discarded
due to poor weather (preventing the bonobos from being released),
one due to unexpected vocalizations (see before), and one due to a
communication problem between keeper and experimenter. The
remaining 23 trials consisted of N = 10 apple, N = 7 kiwi, and
N = 6 control trials, which were completed by four individuals
(GM, CK, KH, LU).
PLoS ONE | www.plosone.org
Supporting Information
Images depicting (a) the playback speaker positioned
during the experimental phase, (b) the view of the sloped outdoor
enclosure from the bonobo exit door.
(TIF)
Figure S1
Order that individuals first arrived to one of
the two fields per trial, with their choice of field
indicated in parentheses (k = kiwi field, a = apple field).
Table S1
8
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
If two or three individuals arrived simultaneously, each of these
individuals was given their number plus 0.5 or 0.3, respectively.
(DOC)
Audio S1 Recording of a call sequence produced by a bonobo in
response to finding kiwi, a high-ranked food.
(WAV)
Tables S2 Median number of visits by each individual
Audio S2 Recording of a call sequence produced by a bonobo in
response to finding apples, a low-ranked food.
(WAV)
to the two fields after playbacks of food-associated call
sequences. The top number indicates the median value, with the
bottom numbers, indicating the 25 and 75 percentiles. PB =
Playback condition.
(DOC)
Text S1
(DOC)
Table S3 Mean time spent (sec) by each individual at
Acknowledgments
the apple and kiwi slope after hearing food- associated
call playbacks. The top number indicates the median value
with the 6 indicating the standard errors.
(DOC)
We are very grateful to Twycross Zoo Management Committee for
permission to conduct research at Twycross Zoo. Our special gratitude
goes to Jackie Hooley, Donna Lundy, Emma Swaddle, Liz Cubberley and
Kris Hern whose help and support made this study possible. We are
grateful to Maren Mende for help with data collection, to Brian Kirk and
Andy Burnley for technical support, and to Katie Slocombe for valuable
advice. We thank Mike Oram, Jim McNicol, Brian Leung and Eric
Bowman for statistical advice, as well as to the reviewers for their helpful
comments.
Table S4 Results of food preference tests conducted on
two groups of captive bonobos at Twycross Zoo, UK.
(DOC)
Table S5 Direct and indirect experiences by subgroup B
individuals during foraging training phase. Direct experience indicates a foraging event where the individual had physical
contact with a food item at the location; indirect experience
indicates a foraging event where the individual witnessed another
individual eating or on contact with a food item, but themselves
did not.
(DOC)
Author Contributions
Conceived and designed the experiments: KZ ZC. Performed the
experiments: ZC. Analyzed the data: ZC. Contributed reagents/materials/analysis tools: KZ. Wrote the paper: KZ ZC.
References
19. Benz J (1993) Food-elicited vocalizations in golden lion tamarins: design features
for representational communication. Anim Behav 45: 443–455.
20. Hauser M, Marler P (1993) Food-associated calls in rhesus macaques (Macaca
mulatta): I. Socioecological factors. Behav Ecol 4: 194.
21. Hauser M (1998) Functional referents and acoustic similarity: field playback
experiments with rhesus monkeys. Anim Behav 55: 1647–1658.
22. Goodall J (1963) Feeding behaviour of wild chimpanzees: a preliminary report;
1963. pp 39–47.
23. Goodall J (1986) The chimpanzees of Gombe: Patterns of behavior. Cambridge:
Cambridge Univ. Press. 673 p.
24. Slocombe K, Zuberbühler K (2006) Food-associated calls in chimpanzees:
responses to food types or food preferences? Anim Behav 72: 989–999.
25. Slocombe K, Zuberbühler K (2005) Functionally referential communication in a
chimpanzee. Curr Biol 15: 1779–1784.
26. Rendall D, Owren M, Ryan M (2009) What do animal signals mean? Anim
Behav 78: 233–240.
27. Seyfarth R, Cheney D (2003) Signalers and receivers in animal communication.
An Rev Psych 54: 145–173.
28. Stegmann U (2009) A Consumer Based Teleosemantics for Animal Signals. Phil
Sci 76: 864–875.
29. Savage-Rumbaugh E, Murphy J, Sevcik R, Brakke K, Williams S, et al. (1993)
Language comprehension in ape and child. Monog Soc Res Child Dev 58.
30. Savage-Rumbaugh S, McDonald K, Sevcik R, Hopkins W, Rubert E (1986)
Spontaneous symbol acquisition and communicative use by pygmy chimpanzees
(Pan paniscus). J Exp Psych: Gen 115: 211–235.
31. Clay Z, Zuberbühler K (2009) Food-associated calling sequences in bonobos.
Anim Behav 77: 1387–1396.
32. de Waal FBM (1988) The communication repertoire of captive bonobos (Pan
paniscus), compared to that of chimpanzees. Behaviour 106: 183–251.
33. Bermejo M, Omedes A (1999) Preliminary vocal repertoire and vocal
communication of wild bonobos (Pan paniscus) at Lilungu (Democratic Republic
of Congo). Fol Primatol 70: 328–357.
34. Galbraith S, Daniel J, Vissel B (2010) A Study of Clustered Data and
Approaches to Its Analysis. J Neurosci 30: 10601.
35. Nelder J, Wedderburn R (1972) Generalized linear models. J Roy Stat
Soc A (Gen) 135: 370–384.
36. Savage-Rumbaugh S, Sevcik RA, Brakke KE, Rumbaugh DM (1990) Symbols:
Their communicative use, comprehension, and combination by bonobos (Pan
paniscus). Adv Infan Res 6: 221–278.
37. Savage-Rumbaugh S, Lewin R (1994) Kanzi. The ape at the brink of the human
mind. NY: Wiley.
38. Clarke E, Reichard U, Zuberbühler K (2006) The syntax and meaning of wild
gibbon songs. PLoS One 1: E73.
39. Genty E, Byrne R (2010) Why do gorillas make sequences of gestures? Anim
Cogn 13: 287–301.
1. Seyfarth RM, Cheney DL, Bergman T, Fischer J, Zuberbuhler K,
Hammerschmidt K (2010) The central importance of information in studies
of animal communication. Anim Behav 80: 3–8.
2. Seyfarth R, Cheney D, Marler P (1980) Monkey responses to three different
alarm calls: evidence of predator classification and semantic communication.
Science 210: 801.
3. Zuberbühler K, Cheney D, Seyfarth R (1999) Conceptual semantics in a
nonhuman primate. J Comp Psych 113: 33–42.
4. Kirchhof J, Hammerschmidt K (2006) Functionally referential alarm calls in
tamarins (Saguinus fuscicollis and Saguinus mystax)–evidence from playback
experiments. Ethology 112: 346–354.
5. Fichtel C, Kappeler P (2002) Anti-predator behavior of group-living Malagasy
primates: mixed evidence for a referential alarm call system. Behav Ecol
Sociobiol 51: 262–275.
6. Ouattara K, Lemasson A, Zuberbühler K (2009) Campbell’s monkeys
concatenate vocalizations into context-specific call sequences. Proc Natl Acad
Sci U S A 106: 22026.
7. Arnold K, Zuberbuhler K (2006) Language evolution: compositional semantics
in primate calls. Nature 441: 303.
8. Schel A, Tranquilli S, Zuberbühler K (2009) The alarm call system of black-andwhite colobus monkeys. J Comp Psychol 123: 136–150.
9. Caine N, Addington R, Windefelder T (1995) Factors affecting the rates of food
calls given by red-bellied tamarins. Anim Behav 50: 53–60.
10. Slocombe KE, Kaller T, Turman L, Squibbs P, Papworth S, Townsend SW,
et al. (2010) Production of food-associated calls in wild chimpanzees is dependent
on the composition of the audience. Behav Ecol Sociobiol, In press.
11. Evans C, Evans L (1999) Chicken food calls are functionally referential. Anim
Behav 58: 307–319.
12. Gros-Louis J (2004) Responses of White-Faced Capuchins (Cebus capucinus) to
Naturalistic and Experimentally Presented Food-Associated Calls. J Comp Psych
118: 396–402.
13. Di Bitetti M (2003) Food-associated calls of tufted capuchin monkeys (Cebus
apella nigritus) are functionally referential signals. Behaviour 140: 565–
592.
14. Kitzmann C, Caine N (2009) Marmoset (Callithrix geoffroyi) food-associated
calls are functionally referential. Ethology 115: 439–448.
15. Elowson A, Tannenbaum P, Snowdon C (1991) Food-associated calls correlate
with food preferences in cotton-top tamarins. Anim Behav 42: 931–937.
16. Roush R, Snowdon C (2000) Quality, Quantity, Distribution and Audience
Effects on Food Calling in Cotton-Top Tamarins. Ethology 106: 673–690.
17. Boinski S, Campbell A (1996) The huh vocalization of white-faced capuchins: a
spacing call disguised as a food call? Ethology 102: 826–840.
18. Benz J, Leger D, French J (1992) Relation between food preference and foodelicited vocalizations in golden lion tamarins (Leontopithecus rosalia). J Comp
Psychol 106: 142–149.
PLoS ONE | www.plosone.org
9
April 2011 | Volume 6 | Issue 4 | e18786
Semantic Integration in Bonobos
44. Manser MB, Seyfarth RM, Cheney DL (2002) Suricate alarm calls signal
predator class and urgency. Trends Cogn Sci 6: 55–57.
45. Hollén L, Manser M (2007) Motivation before meaning: motivational
information encoded in meerkat alarm calls develops earlier than referential
information. Am Nat 169: 758–767.
46. Gros-Louis J (2004) The function of food-associated calls in white-faced
capuchin monkeys, Cebus capucinus, from the perspective of the signaller. Anim
Behav 67: 431–440.
47. Van Krunkelsven E, Dupain J, Van Elsacker L, Verheyen RF (1996) Food
calling in captive bonobos (Pan paniscus): an experiment. Int J Primatol 17:
207–217.
40. Owren M, Rendall D, Ryan M (2010) Redefining animal signaling: influence
versus information in communication. Biol Phil. pp 1–26.
41. Seyfarth R, Cheney D (2003) Meaning and emotion in animal vocalizations.
Annals New York Acad Sci 1000: 32–55.
42. Zuberbuhler K (2003) Referential signaling in non-human primates: Cognitive
precursors and limitations for the evolution of language. Adv Stud Behav 33:
265–307.
43. Manser MB (2001) The acoustic structure of suricates’ alarm calls varies with
predator type and the level of response urgency. Proc R Soc Lond B 268:
2315–2324.
PLoS ONE | www.plosone.org
10
April 2011 | Volume 6 | Issue 4 | e18786