Вы находитесь на странице: 1из 9


Animal Behaviour 134 (2017) 229e237

Contents lists available at ScienceDirect

Animal Behaviour
journal homepage: www.elsevier.com/locate/anbehav

Special Issue: Communicative Complexity

Structural variability and communicative complexity in acoustic

Julia Fischer a, b, c, *, Philip Wadewitz a, Kurt Hammerschmidt a, c
Cognitive Ethology Laboratory, German Primate Center, Go €ttingen, Germany
Georg-August-University Go€ttingen, Go
€ttingen, Germany
Leibniz ScienceCampus Primate Cognition, Go €ttingen, Germany

a r t i c l e i n f o
The notion that social complexity may drive communicative complexity has invigorated the research
Article history: interest in the question of how to assess the structural features of a species' communication system. This
Received 15 February 2016 applies to both the level of the signal repertoire and the level of potential rules governing the succession
Initial acceptance 29 February 2016 of elements. This review first provides an overview of some of the most influential studies in the realm of
Final acceptance 20 May 2016 acoustic communication, before turning to a key problem at the foundation of many analyses. Many
Available online 9 August 2016 biological signal repertoires reveal intermediate forms between specific signal types as well as variation
MS. number: 16-00124 within signal types. Therefore, it is often difficult to identify the specific number of signal types (and
consequently, their sequential relationships). Nevertheless, subjective classification or ‘hard clustering’
Keywords: approaches force items into specific categories. Yet, given the graded nature of many repertoires, it may
be more appropriate to measure the degree of differentiation within a repertoire, instead of the number
call type
of call types, which may also be strongly affected by sampling artefacts. ‘Fuzzy clustering’ provides
fuzzy clustering
hard clustering
measures to capture the overall structural variability of a repertoire, i.e. whether they are rather graded
information theory or discrete. Because with fuzzy clustering it may also be difficult to identify a single best cluster solution,
primate methods are needed that transcend the number of clusters identified with the cluster analysis. One such
repertoire approach is the assessment of the distribution of typicality coefficients, which are derived from fuzzy
signal clustering. For the time being, these provide an alternative route to quantitatively test hypotheses
vocalization regarding the evolution of signal repertoires. Future research should aim to establish a solid mathe-
matical foundation to link the properties of graded repertoires to measures derived from complexity
theory. Until then, the notion of complexity to describe the structure of a repertoire should be used with
© 2016 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal
Behaviour. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/

Why do some birds just produce three or four different song better ways to quantitatively capture communicative complexity,
types, and others more than 200? Why do some monkey species so that it can be estimated reliably across species and signal mo-
mainly grunt, scream and bark, while others evolved an intriguing dalities, for use in comparative analyses.
variety of twitters, whistles, caws and shrieks? Numerous re- In this paper, we focus on the acoustic domain and largely
searchers have aimed to provide answers to such questions, and restrict ourselves to the question of how to quantify vocal
have invoked habitat characteristics, predation pressure, sexual complexity in nonhuman primates. Nonhuman primates (here-
selection and more recently social complexity as key drivers of after: ‘primates’) provide particular challenges because most spe-
vocal complexity (Bradbury & Vehrencamp, 2011; Freeberg, cies' repertoires can be considered as graded, with substantial
Dunbar, & Ord, 2012). An unresolved issue, however, is that for a variation within and between call types, such as in Japanese ma-
formal test of any of the hypotheses put forward to explain varia- caques, Macaca fuscata (Green, 1975), Barbary macaques, Macaca
tion in signal repertoires between species, we need to develop sylvanus (Hammerschmidt & Fischer, 1998), rhesus macaques,
Macaca mulatta (LePrell, Hauser, & Moody, 2002), chimpanzees,
Pan troglodytes (Crockford & Boesch, 2005), as well as gorillas,
* Correspondence: J. Fischer, Cognitive Ethology Lab, German Primate Center, Gorilla gorilla gorilla and Gorilla beringei beringei (Hedwig, Robbins,
Kellnerweg 4, 37077 Go € ttingen, Germany. Mundry, Hammerschmidt, & Boesch, 2014). This probably also
E-mail address: jfischer@dpz.eu (J. Fischer).

0003-3472/© 2016 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

230 J. Fischer et al. / Animal Behaviour 134 (2017) 229e237

holds for the majority of other terrestrial mammals, due to the affected by signal variation but also by other available information,
sound production mechanism in these taxa (Fitch & Hauser, 1995). such as contextual cues and signaller identity, among others
We therefore assume that the core findings apply to a broader (Fischer, 2013; Wheeler & Fischer, 2012). In addition, it is known
range of study species, and may also be valid in other communi- that receivers may recode graded variation into discrete categories
cative domains, such as facial expressions (Parr & Waller, 2006; (see Fischer, 2006 for a review). Therefore, for both practical and
Scheider, Liebal, On~ a, Burrows, & Waller, 2014) or gestures conceptual reasons, a quantitative assessment of communicative
(Hobaiter & Byrne, 2014; Pika, Liebal, Call, & Tomasello, 2005). complexity that includes both signaller and receiver characteristics
We begin this review by summarizing different approaches that can be achieved for a limited selection of an entire repertoire, at
have been taken to conceptualize vocal complexity, and stress the best.
importance of clarifying the level of analysis. We then provide an When applying measures derived from information theory, it is
overview of the results of some of the most influential studies, important to consider in which way statistical information is
before we turn to our own attempt to characterize the structure of related to biological information. Statistical information is maxi-
primate vocal repertoires, paying particular attention to the issue of mized when signals are maximally diverse and/or when sequences
graded versus discrete repertoires. Finally, we suggest a novel are truly random (Shannon & Weaver, 1949). Communication sys-
method to capture the structural variability of repertoires, to tems, however, require a balance between redundancy and di-
complement (or replace) previous attempts to settle on a specific versity (see McCowan et al., 1999). But there is more to consider
number of call types. than redundancy and diversity. As analyses of the evolution of
communication have revealed (Dawkins & Krebs, 1978; Maynard
MEASURING COMMUNICATIVE COMPLEXITY Smith & Harper, 2003), signallers are selected to produce signals
that serve their own best interests, and that are sufficiently cheap.
There are different levels at which communicative complexity At the same time, signals only evolve when receivers respond to
can be captured, with increasing difficulties in terms of oper- them (Fischer, 2013; Maynard Smith & Harper, 2003; Scott-Phillips,
ationalization. Specifically, one needs to distinguish between the 2008). For this, signals need to be sufficiently informative, in the
identification of elements on the one hand and analyses of higher- sense that they either correlate with a specific state, e.g. resource-
order relationships between elements that appear in succession holding potential, or can be used to predict upcoming behaviours or
(sequence analyses) on the other. At the level of the elements, this events, such as imminence of attack, so that it pays the receiver to
would amount to the identification of the number of call types in attend to them (Bradbury & Vehrencamp, 2011). Correspondingly,
the repertoire (Kershenbaum, Freeberg, & Gammon, 2015). Previ- two affordances of a communication system emerge, namely a
ously, a higher number of different call types had been equated sufficient degree of consistency of signal occurrence with a specific
with a higher degree of complexity (e.g., McComb & Semple, 2005). state or context and sufficient capacity to encode as much (po-
The central issue here, as we show below, is that the identification tential) information as possible (for an in-depth discussion of the
of the number of units in a signalling repertoire can be extremely term information, see Fischer, 2013).
challenging when intermediates between different signal types
exist. The problem of identifying the number of units or call types is DRIVERS OF REPERTOIRE STRUCTURE
further exacerbated because other factors, such as individual sig-
natures or variation in signaller quality or state add to the structural Habitat
variability in the repertoire.
Another way to measure vocal complexity is based on infor- Several factors have been assumed to impact the structure of a
mation theory (Shannon & Weaver, 1949). Information theory species' vocal repertoire. Peter Marler suggested that in species that
provides a formal approach to characterize a communicative event live in habitats with poor visibility between signaller and receiver
in terms of its statistical properties. A key measure in information and/or high background noise such as dense rainforests, discrete
theory is entropy, which provides an estimation of the amount of repertoires should be favoured, because a clear discriminability of
uncertainty in a communicative system (Shannon & Weaver, 1949). call types facilitates signal recognition (Marler, Kavanaugh, &
The greater the variation within a signal space, the greater the Cutting, 1975). In contrast, in species that live in open habitats
uncertainty. Information theoretical concepts, such as Zipf plots, with visual access to each other, graded repertoires with variation
have been used to describe the structure of vocal repertoires based within and between call types should be favoured because they
on the frequency of occurrence of different elements in a repertoire have a higher capacity to encode potential information, as calls may
(McCowan, Hanser, & Doyle, 1999). The Zipf statistic is derived from vary with regard to arousal level. Furthermore, in case of ambiguity,
a logelog plot of the frequency of occurrence of signalling units visual information can be used to disambiguate the situation.
against their rank order (Zipf, 1949). In many communication sys- Similarly, Marler assumed that within a species' repertoire, grada-
tems, an approximate slope of -1 emerges (Zipf's law), and there is tion was more likely to occur in close-range signals, whereas long-
considerable debate about the significance of this relationship (see distance signals should be more distinct (Marler, 1967).
McCowan, Doyle, Jenkins, & Hanser, 2005 for a summary). Others Others have made more specific predictions regarding the
have used information theoretical approaches to analyse the acoustic features of long-distance calls in relation to habitat. It has
sequential composition of communicative signals (Freeberg & been proposed that in closed habitats longer signals, signals with a
Lucas, 2012; Hailman, Ficken, & Ficken, 1985; Kershenbaum, lower repetition rate, a lower frequency modulation and a lower
2014; Ord & Martins, 2006). Importantly, information theoretical frequency range should be favoured (see Ey & Fischer, 2009 for a
approaches also rely on an estimation of the number of different review). While the vocalizations of Japanese macaques conformed
units in the system, and thus require a solution to the problem of to the predictions (Sugiura, Tanaka, & Masataka, 2006), this was not
identifying the number of call types in the repertoire. the case in other species (e.g. in marmosets: Daniel & Blumstein,
To make a full estimate of the communicative complexity of a 1998). A study of the loud calls given over long distances of four
species requires one to take into account the way variation in sig- sympatric primate species on Siberut Island also provided only
nals affects the behaviour of receivers (Freeberg et al., 2012). There mixed support for the above-mentioned predictions (Schneider,
are a number of major problems associated with estimating Hodges, Fischer, & Hammerschmidt, 2008). Although all four spe-
complexity at this level, because responses to signals are not only cies concentrated most of the energy (amplitude) of their loud calls

J. Fischer et al. / Animal Behaviour 134 (2017) 229e237 231

in the spectral window with the lowest background noise, in three elements within song types. Kroodsma (1977), for instance, found a
of the four species the structure of their calls did not appear to be relationship between the degree of polygyny and song complexity.
adapted to long-range transmission. This clearly suggests that Carolina chickadees, Poecile carolinensis, living in larger groups
signal design is the result of a complex set of selective pressures, reveal a greater diversity of note types and a greater versatility in
including the need to ensure species recognition, and that phylo- the combination of their chick-a-dee calls than individuals living in
genetic inheritance may be more decisive than adaptation to the smaller groups (Freeberg, 2006). In a further study, Freeberg and
current habitat. Indeed, there is now ample evidence that the Lucas (2012) found that the information content of Carolina
structure of calls in closely related species is highly conserved and chickadees' songs was higher than the information content of
that variation in structure is closely correlated with phylogenetic black-capped chickadee, Poecile atricapillus, songs. The authors
descent. Thinh, Hallam, Roos, and Hammerschmidt (2011) analysed related this finding to variation in social structure between the two
the song features and mitochondrial cytochrome b from 22 gibbon species, as black-capped chickadees have a more despotic social
populations representing six of the seven crested gibbon species system with fewer reversals in dominant - subordinate interactions
(genus Nomascus), finding a high concordance between song than Carolina chickadees (see also Krams et al., 2012). For mam-
structure and phylogeny (Thinh et al., 2011). Similarly, an analysis of mals, Blumstein and Armitage (1997) examined the alarm calling of
the loud calls of six species of wild male surilis, Presbytis spp., a ground-dwelling sciurid rodents, and found that the number of
genus of Asian leaf monkeys, revealed a clear correlation between alarm calls correlated with the diversity of social roles as a measure
call structure and genetic similarity (Meyer et al., 2012). Both of social complexity. This could be confirmed by a comparative
studies indicate that the acoustic structure of loud calls can be used study using a recent sciurid phylogeny (Pollard & Blumstein, 2012).
to distinguish between species and to verify phylogenetic related- To test whether social complexity drives vocal complexity in
ness and migration history. nonhuman primates, McComb and Semple (2005) extracted the vocal
repertoire size of more than 40 different species from the literature
Predation Pressure and compared the results with the mean group size of the species. As a
second measure of social complexity they calculated the grooming
Another factor that has been assumed to influence the structure time for a subset of species and used this as a proxy of social bonding.
of vocal repertoires is predation (Braune, Schmidt, & Zimmermann, Species living in larger groups and with more intense social bonding
2008; Freeberg et al., 2012). In species with multiple predator- were found to have larger vocal repertoires (McComb & Semple,
specific defence strategies, differentiated alarm calls are deemed 2005). A study on three forest-dwelling species of Cercopithecinae
beneficial because they allow receivers to select appropriate re- lent further support to the hypothesis that more socially complex
sponses without having to ascertain the type of predator or level of primate groups have evolved a more complex vocal communication.
urgency (Zuberbühler, 2006). The numerous publications on Bouchet, Blois-heulin, and Lemasson (2013) found that acoustic
predator-specific variation in alarm calls in more vulnerable species variability and individual distinctiveness were related to their
such as guenons and vervet monkeys lend support for this importance in social function. Contact calls used for intragroup
assumption. In addition, relatively large species such as baboons cohesion were most individually distinctive regardless of the species.
produce alarm calls with much lower specificity (Fischer, In addition, they found that repertoire size, diversity and usage were
Hammerschmidt, Cheney, & Seyfarth, 2001, 2002). Yet, the related to the social complexity of the species. Red-capped manga-
assessment of call specificity has to be taken with a pinch of salt, as beys, Cercocebus torquatus, which live in large and relatively despotic
it crucially depends on the range of call types included in the multimale multifemale groups with frequent interactions, had the
analysis. A recent reanalysis of the alarm calls of vervet monkeys, most complex vocal repertoire of all three species, while De Brazza's
Chlorocebus pygerythrus (Price et al. 2015), corroborated the monkeys, Cercopithecus neglectus, which live in small family groups,
assumption that calls given in different predator contexts are had the smallest vocal repertoire (Bouchet et al., 2013).
acoustically distinct (Seyfarth, Cheney, & Marler, 1980). Yet, when In sum, evidence is accumulating that higher social complexity
calls from aggressive contexts were included into the analysis, the is related to higher vocal complexity. However, all of these studies
specificity was much lower, because both ‘chutter’ and ‘rraup’ calls are afflicted by the same methodological drawback, namely that
(typically given in response to snakes and aerial predators, there is no objective way to estimate the size of a vocal repertoire
respectively) also occur during within-group and between-group when it is graded, and this may also apply to the identification of
aggression (Price et al., 2015). Similarly, in an analysis of variation graded song elements in birdsong (Hailman et al., 1985). Therefore,
in chacma baboon, Papio ursinus, grunts, the inclusion of additional it would be highly desirable to agree on common methods (and
contextual situations decreased the assumed context specificity of data sharing) to facilitate objective measures of the structure of
the calls (Meise, Keller, Cowlishaw, & Fischer, 2011). Notably, the signal repertoires.
structure of male alarm calls in members of the genus Chlorocebus
in eastern, southern and western Africa differed only marginally, IDENTIFYING THE APPROPRIATE NUMBER OF CALL TYPES
while varying with phylogenetic distance (Price, Ndiaye, & Fischer,
2014), corroborating the view that the patterns that make up a From what we said above, it would be important to come to a
vocal repertoire are highly genetically fixed. better agreement on how to identify the number of call types in a
repertoire. This is particularly challenging when there is variation
Social Complexity within and between call types, which renders the identification of
the numbers of elements difficult, and has important repercussions
The idea of a link between communicative and social complexity for analyses of the compositional nature of a repertoire when the
can be traced back to Darwin (1872). In recent years, this notion has units that make up the sequences are graded.
been termed the ‘social complexity hypothesis for communication’ Graded repertoires with substantial variation within and be-
(Freeberg et al., 2012; Krams, Krama, Freeberg, Kullberg, & Lucas, tween call types have been described in a number of species,
2012; Sewall, 2015). It postulates that individuals living in more including different macaque species such as rhesus macaques
complex social environments need a more complex communica- (Rowell & Hinde, 1962), Japanese (Green, 1975) and Barbary ma-
tion. For birdsong, much research focus has been put on song or- caques (Hammerschmidt & Fischer, 1998), as well as in apes (e.g.
ganization, that is, the relationships of successive song types or bonobos, Pan paniscus, de Waal, 1988). In contrast, the calls from

232 J. Fischer et al. / Animal Behaviour 134 (2017) 229e237

forest-dwelling guenons such as Diana monkeys, Cercopithecus results. One critical issue in the use of automated procedures is the
diana (Zuberbühler, Noe, & Seyfarth, 1997), putty-nosed monkeys, choice of the acoustic features that are used. In a study that aimed
Cercopithecus nictitans (Arnold & Zuberbühler, 2006), and blue to characterize the vocal repertoire of chacma baboons, Wadewitz
monkeys, Cercopithecus mitis (Papworth, Bo € se, Barker, Schel, & et al. (2015) assessed the effects of the selection of input variables
Zuberbühler, 2008) have been labelled as discrete. There are two on the resulting cluster solutions. They extracted 118 acoustic var-
caveats, however: first, not all researchers agree whether a specific iables from FFT-derived spectrograms of approximately 1000 calls,
variation within and between call types should warrant classifica- and used different sets of acoustic features, containing nine, 38 and
tion as graded or discrete; and second, not all of the above analyses 118 variables, as well 19 factors derived from principal component
included the complete repertoires of the species, so that assess- analysis of the 118 variables. They found that the data sets with 38
ments of the classification of the repertoire may still require revi- and 118 acoustic features led to better clustering results than the
sion. Notably, more recently, the alarm calls of Campbell's monkeys, data set with only nine features for k-means and hierarchical
Cercopithecus campbelli, were classified as being graded, not clustering approaches. Importantly, the clustering result based on
discrete (Keenan, Lemasson, & Zuberbühler, 2013). It is also factor loadings yielded the poorest results (Wadewitz et al., 2015).
important to keep in mind that the classification of entire reper- One of the most frequently used measures to validate results
tories as either discrete or graded represents an oversimplification, gained from unsupervised clustering is the Silhouette value, which
since there may be both gradation and discrete differences, and call ranges from -1 to 1 (Rousseeuw, 1987). Silhouette values reflect
types may vary to different degrees (Fichtel & Kappeler, 2002; both the tightness within a cluster and the separation between
Manser, Seyfarth, & Cheney, 2002). Only a few studies have used different clusters (Rousseeuw, 1987). Very low Silhouette values
reproducible quantitative approaches such as cluster analysis to indicate that a given data point is equidistant to two cluster centres,
estimate the number of call types in the repertoire (Gamba et al., whereas positive values indicate that the data points are closer to
2015; Hammerschmidt & Fischer, 1998; Maciej, Ndao, the centre of one cluster than the second closest cluster. The overall
Hammerschmidt, & Fischer, 2013). Yet, such studies also have Silhouette width, S(A), is defined as the average of the Silhouette
some pitfalls because the number of available recordings, the se- values over the entire data set and can be taken as a measure of the
lection of acoustic features and the cluster algorithm may strongly clustering quality. Silhouette values can be used to compare the
influence the resulting number of call types (Wadewitz et al., 2015). quality of different cluster solutions; their sensitivity to different
sets of acoustic variables limits their value, however.
Feature Extraction An alternative method to describe the quality of a cluster solu-
tion is the reduction in the total variance of the acoustic features of
The first step in the analysis of vocal repertoires is to define the calls (Bacher, 1994). This can be expressed in two ways, either as the
signal. Calls may consist of a single call unit or a combination of reduction in the total variance (the variance of the null cluster
several units. These elements may be acoustically similar, as in model) in relation to the respective cluster solution
Barbary macaque copulation calls (Pfefferle, Brauch, Heistermann, (eta(k) ¼ 1  SS(k)/SS(1)), with SS(k) ¼ sum of squares in the kth
Hodges, & Fischer, 2008) or vervet monkey alarm calls (Price cluster, and SS(1) the sum of squares in the unpartitioned data set, or
et al., 2015), or they may reveal substantial variation between el- as the difference between two successive cluster solutions
ements, as in gibbon song (Dallmann & Geissmann, 2001; Thinh (pre(k)¼1  SS(k)/SS(k1)). Although both measures together allow
et al., 2011) or chimpanzee pant hoots (Arcadi, 1996; Crockford & for an assessment of the explanatory power of different cluster
Boesch, 2005; see Fig. 1 for spectrographic depictions). A fast solutions, it may still be difficult to decide what the best solution
Fourier transform (FFT) is the most common foundation for would be (Hammerschmidt & Fischer, 1998). This also applies to
extracting acoustic features. The FFT converts the amplitude signal Silhouette values (Ben-David, Luxburg, & Pal, 2006; Wadewitz
into the frequency domain, resulting in a spectrogram that repre- et al., 2015), when applied to graded data structures. In such
sents the amplitude over frequency and time. From the spectro- cases, different cluster solutions may be equally suited to reflect the
gram, temporal and spectral features can be extracted using various structure of the data set. The problem is illustrated in Fig. 2. In
algorithms. To estimate the fundamental frequency (F0), additional Fig. 2a, the data have a discontinuous distribution (discrete struc-
algorithms that are mostly based on autocorrelation of the spec- ture); thus, a ‘hard’ cluster algorithm is able to assign each case to
trum are available. An alternative approach is linear predictive one category. Fig. 2b depicts a case with a graded data structure,
coding (LPC; Owren & Bernacki, 1998). LPC is based on the source- where it becomes difficult to assign all of the cases to one of the
filter model and is used to measure the resonance (formant) fre- three categories unambiguously; some cases appear to fall in the
quencies. It has its origin in human speech analysis, but is also used middle between two (or three) categories. An alternative approach
in the analysis of primate calls. From the vocal tract length of the therefore is to characterize single cases with regard to multiple
signaller, the number of formants can be determined and, subse- category memberships. This method is known as ‘fuzzy clustering’.
quently, formant dispersion can be calculated (Owren, Linker,
Zimmermann, Newman, & Jürgens, 1995; Pfefferle & Fischer, Fuzzy Clustering
2006; Rendall, Kollias, Ney, & Lloyd, 2005).
Fuzzy set theory (Zadeh, 1965) acknowledges that the mem-
Call Classification and Clustering bership to a given cluster may be imperfect. It is therefore
appealing in cases where the separation between different cate-
In earlier years, the descriptions of vocal repertoires relied on a gories is gradual rather than discrete (Zadeh, 2008). Accordingly,
visual classification of spectrograms (Struhsaker, 1967), supple- fuzzy c-means clustering can be applied to describe systems that do
mented with information about the context of calling (Moody & not exhibit clearly separated categories. Each case is assigned a
Menzel, 1976; Palombit, 1992). Yet, such subjective classifications membership value for each of the clusters, ranging from m ¼ 1
are not suited for quantitative comparative analyses, and therefore, (fully in accordance with the properties of the cluster) and m ¼ 0
the usage of algorithm-based approaches such as unsupervised (none of the properties of the cluster in question). Intermediate
clustering is now considered a requirement (Deecke & Janik, 2006; membership values are assigned to cases that do not fully belong to
Pozzi, Gamba, & Giacoma, 2010). Nevertheless, subjective decisions any of the clusters, but appear to be intermediates between
in terms of the settings in the analysis may strongly affect the different types.

J. Fischer et al. / Animal Behaviour 134 (2017) 229e237 233

10 (a)

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5

10 (b)
Frequency (kHz)

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

10 (c)

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 10

5 (d)

5 10 15 20 25 30
Time (s)

Figure 1. Spectrograms of primate calls comprising multiple elements. (a) Female Barbary macaque copulation call, (b) a bout of male vervet monkey alarm calls, which may
comprise either single or multiple elements; (c) chimpanzee pant-hoot with build-up and climax parts (recorded from P. t. verus); (d) song of a Kloss gibbon. Note the variation in
time and frequency resolution on the X and Y axes.

(a) (b)
Discriminant feature 2

Discriminant feature 2

m=0.7 m=0.1 c


Discriminant feature 1 Discriminant feature 1

Figure 2. (a) When the distributions of discriminant features are discontinuous, discrete categories can be unambiguously identified. (b) When the discriminant features are
continuously distributed, no clear categories emerge. Fuzzy clustering classifies each item (call) with regard to the probability of membership (m) to multiple clusters, in this
example to three clusters ‘c’.

Two parameters of the fuzzy c-means algorithm (Jang & Sun, If m ¼ 1, the extracted clusters are crisp and membership values of
1997; Xu, Keller, Popescu, & Bondugula, 2008) can be adjusted to the cases in the analysis are either 1 or 0 (under these circum-
identify which number of clusters is most suited to describe the stances, fuzzy c-means clustering converges exactly to k-means
data set. The first parameter is the maximum number of clusters clustering). By increasing m, clusters become fuzzier and adjacent
allowed, while the second parameter is the fuzziness parameter m. clusters eventually merge, decreasing the number of clusters. In a

234 J. Fischer et al. / Animal Behaviour 134 (2017) 229e237

similar fashion to k-means, the fuzzy c-means algorithm creates gradually increased m, which is also known as the temperature
clusters by randomly seeding cluster centroids and subsequent parameter. With increasing fuzziness, clusters begin to merge
iterative optimization. In this regard, fuzzy c-means clustering (Fig. 3a). One critical piece of information in fuzzy clustering is the
suffers from the same sensitivity to the initial cluster centroids as k- stability, which refers to the range of values of m over which a
means clustering. specific cluster solution will remain stable (Fig. 3a). Up to a value of
Once the fuzziness parameter m is chosen and the clusters are m ¼ 1.44, clusters were not very stable and rapidly merged into
computed, the main cluster a for each case is the cluster with the fewer clusters. At m ¼ 1.505, a five-cluster solution appeared that
highest membership value (see Wadewitz et al., 2015 for further was relatively stable (Fig. 3d) and consisted of grunts, barks,
details). By subtracting the second largest membership value from weaning calls and two structurally different scream clusters. At
ð1stÞ ð2ndÞ
the first, the so-called typicality coefficient dðiÞ ¼ mi  mi is m ¼ 1.600, the two scream clusters merged, and the weaning calls
computed. The average d of all typicality coefficients and their and barks merged, resulting in a three-cluster solution (Fig. 3c).
distribution, quantified by the halved mean absolute deviation Finally, at m ¼ 2.000, two clusters remained, one consisting of
D ¼ d  d/2 can then be calculated for the entire data set. Based on screams and the other of all other calls (Fig. 3b). This solution
the observed distribution of typicality coefficients, cases can then emerged as the most stable one, although it is clear that this so-
be considered as typical if d > dtypical ¼ d þ D and as atypical if lution is not appropriate to describe the structure of the vocal
d < datypical ¼ d  D. repertoire of chacma baboons.
We assessed the performance of two commonly used hard The stability over change in fuzziness is not the only way to
clustering approaches, centre-based k-means and hierarchical inspect the result of fuzzy clustering. A two-dimensional visualiza-
Ward's clustering, to fuzzy c-means clustering (Dunn, 1974), using tion of how calls are scattered in membership space is a further
the vocalizations of chacma baboons as a worked example possibility to see which calls are typical for a call type and in which
(Wadewitz et al., 2015). The goal was to test whether fuzzy c-means cases calls grade from one call type into another. In Fig. 4a, each call
clustering would be superior to deal with the graded variation is represented by its membership to the closest and the second
observed in the vocalizations of this species (Fischer et al., 2001, closest cluster. We found intermediates between weaning calls and
2002). We compared the results to a visual classification by expe- barks, as well as between weaning calls and grunts. Intermediate
rienced human observers, which identified five main call types: calls were also observed between the two scream clusters and more
grunts, barks, noisy screams, tonal screams and weaning calls. We rarely between the bark and the noisy scream cluster. Note that the
started with a maximum number of allowed clusters of 15, because tonal screams exhibited generally low typicality scores. We
this was well above the expected number of clusters, and then observed typical grunts and barks, as well as intermediates between

(a) (b) µ = 2.000

13 Screams
Number of clusters

5 Lumped
2 0.8
1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5 0.6
Fuzziness parameter (µ)

(c) µ = 1.600 (d) µ = 1.505 0.4

Tonal screams

Noisy screams

Grunts Grunts



Figure 3. Partitions of the chacma baboon's vocal repertoire are visualized as so-called membership matrices. (a) Stability of cluster solutions in relation to the fuzziness parameter
m. The maximum number of allowed clusters was set to 15; the first informative solution is 14, because the stability of the maximal solution is systematically overestimated.
Partitions with more than five clusters could only be found for a very narrow range of m values. (bed) Three examples of membership matrices: rows correspond to the respective
number of clusters (two, three or five clusters); columns correspond to single calls (N ¼ 912). The typicality of membership in the respective clusters is colour-coded, with blue as
atypical and red as typical (see heat map). Cases are ordered with regard to the membership to the respective bottom category. (b) Two-cluster solution containing a cluster with
screams and one with all other call types. In the ‘scream’ row, most cases are blue and thus atypical for screams, while cases marked as red are typical members of this cluster. (c)
Three-cluster solution (with screams in one cluster, grunts in the second and all other call types in the third). (d) Five-cluster solution: most typical members can be found in the
yellow to red coloured rectangles. Modified from Wadewitz et al. (2015).

J. Fischer et al. / Animal Behaviour 134 (2017) 229e237 235

(a) 1 (b)

Frequency (kHz)

0 2
1 0.5 1 1.5 2 2.5

Time (s)

0 (c) 200
1 160
Tonal scream

Noisy scream
Tonal scream

Number of calls

Noisy scream


0.2 0.4 0.6 0.8 1
0 1 0 1 0 1 0 1
Grunt Bark Weaning Tonal scream Typicality

Figure 4. Comparison of cluster segregations. (a) Two-dimensional projections of membership of calls belonging to the grunt (red), noisy scream (green), tonal scream (orange),
weaning (yellow) and bark (blue) clusters. Every call is represented once (by the closest and second closest clusters). Diagonal lines represent identical memberships. (b) Call
example represents transitions from typical grunt calls to typical bark calls with intermediates close to the cluster boundaries. Colour codes reflect the respective membership value
(see Fig. 2). (c) Distribution of typicality coefficients (TC). TCs are calculated by subtracting the membership of the second closest cluster from the membership of the closest cluster
(see text). A TC of 1 indicates that the call lies right in the centre of a call type and has a morphological structure that is typical for the call type; a TC of 0 indicates that the call lies
exactly between the centres of two call types and has an intermediate structure. Modified from Wadewitz et al. (2015).

the two call types (see Fig. 4b). In contrast to the other pairs, no calls the current understanding of the sound production mechanisms in
at the boundary between grunts and noisy or tonal screams were primates is beyond the scope of this paper, it seems reasonable to
observed, indicating that these clusters remained well separated. assume that variation between call types is the result of different
Based on the membership values it is possible to classify each motor pattern generators underpinning the muscle movements
call as typical or atypical for a given type. Inspection of the distri- associated with phonation, while variation within call types can be
bution of typicality coefficients (Fig. 4c) reveals which call types linked to increasing lung pressure, variation in articulation, body size,
consist mainly of typical calls and which show more intermediates. hormonal state, or other physiological and neural changes associated
The majority of both grunts and barks could be clearly separated with an animal's general affective state (Ackermann, Hage, & Ziegler,
from the other call types, as evidenced by their high typicality co- 2014; Fitch & Hauser, 1995; Jürgens, 2002, 2009).
efficients. Weaning calls were not well separated from barks and For receivers, not only the variation between call types is of
grunts, and the two scream clusters showed much overlap. We interest. Individual variation (Hammerschmidt & Todt, 1995),
suggest that the distribution of typicality coefficients could be used variation in relation to arousal (Fichtel, Hammerschmidt, & Jürgens,
to compare different repertoires. Measures such as the skew could 2001) or hormonal state (Pfefferle, Heistermann, Pirow, Hodges, &
be derived. Right-skewed distributions would indicate higher dif- Fischer, 2011), as well as acoustic variation in relation to physical
ferentiation, while left-skewed distributions would indicate a distance between sender and receiver (Maciej, Fischer, &
higher degree of gradation (Wadewitz, 2015). Hammerschmidt, 2011), all add to the variation that the receiver
One insight from the application of the fuzzy clustering method has to process; at the same time this variation is a source of
was that this approach does not yield a better insight into the ‘true’ potentially important information for the receiver. A further
number of call types either. Yet, whether most or just a few calls can be important factor to consider is variation in call type usage, which
unambiguously assigned to one call type (or cluster) may prove to be will strongly affect the information value of a given signal. When
informative in comparative analyses of vocal repertoires. We suggest the basic number of patterns is of interest, however, it seems
that the inspection of the distribution of typicality coefficients may advisable to control for the other sources of variation.
provide a fruitful avenue to describe the structure of a repertoire. Finally, how does structural variability relate to communicative
Further research will be necessary to put this conjecture to a test. complexity? Previous studies have equated a higher number of call
types with higher complexity. This relationship cannot be linear,
Conclusion and Outlook however, as a completely random system would not be more
complex. Moreover, given the graded nature of many signal rep-
Depending on the research interest, future studies should ertoires, and the resulting problems in identifying the true number
explicitly state whether they are interested in identifying the number of call types, it may be more appropriate to measure the degree of
of call types or elements in the repertoire, or whether the overall differentiation within the signal repertoire, instead of the number
structural variability is of interest. In the former case, it may prove of call types. As long as there are no clear theoretical foundations to
fruitful to consider the basis of sound production. While a review of link repertoire properties of graded systems to complexity theory, it

236 J. Fischer et al. / Animal Behaviour 134 (2017) 229e237

may be best to refrain from the concept of communicative Dallmann, R., & Geissmann, T. (2001). Different levels of variability in the female
song of wild silvery gibbons (Hylobates moloch). Behaviour, 138, 629e648.
complexity altogether, at least in terms of the identification of units
Daniel, J. C., & Blumstein, D. T. (1998). A test of the acoustic adaptation hypothesis in
in the system. four species of marmots. Animal Behaviour, 56(6), 1517e1528. http://dx.doi.org/
Future research should aim to put the link between repertoire 10.1006/anbe.1998.0929.
structure and complexity theory on a firm footing. One might hy- Darwin, C. (1872). The expression of the emotions in man and animals. London, U.K.:
John Murray.
pothesize that repertoires that exhibit both a certain degree of Dawkins, R., & Krebs, J. R. (1978). Animal signals: Information or manipulation? In
structure in the form of clear prototypes of calls and noise in the J. R. Krebs, & N. B. Davies (Eds.), Behavioural ecology: An evolutionary approach
form of variation of these prototypes may be more complex than (pp. 282e309). Oxford, U.K.: Blackwell Scientific Publications.
Deecke, V. B., & Janik, V. M. (2006). Automated categorization of bioacoustic signals:
those that are highly skewed. The relationship is not trivial, how- Avoiding perceptual pitfalls. Journal of the Acoustical Society of America, 119(1),
ever, as it would be unclear whether a repertoire with a bimodal 645e653.
distribution of typicality coefficients would be more or less com- Dunn, J. C. (1974). A fuzzy relative of the ISODATA process and its use in detecting
compact well-separated clusters. Journal of Cybernetics, 3(3), 32e57. http://
plex than one with a flat distribution. Whereas it is relatively dx.doi.org/10.1080/01969727308546046.
straightforward to apply complexity theory to assess the Ey, E., & Fischer, J. (2009). The “Acoustic adaptation hypothesis” e A review of the
complexity of a system based on the relationships between items, evidence from birds, anurans and mammals. Bioacoustics, 19, 21e48.
Fichtel, C., Hammerschmidt, K., & Jürgens, U. (2001). On the vocal expression of
more work is needed to integrate information about category emotion. A multi-parametric analysis of different states of aversion in the
membership and the likelihood of category membership, at the squirrel monkey. Behaviour, 138, 97e116.
ensemble level. Within biology, this question might be of interest to Fichtel, C., & Kappeler, P. M. (2002). Anti-predator behavior of group-living Mala-
gasy primates: Mixed evidence for a referential alarm call system. Behavioral
a broader community of researchers.
Ecology and Sociobiology, 51(3), 262e275.
Fischer, J. (2006). Categorical perception in animals. In K. Brown (Ed.), Encyclopedia
of language & linguistics (2nd ed., pp. 248e251). London, U.K.: Elsevier.
ETHICAL NOTE Fischer, J. (2013). Information, inference and meaning in primate vocal behaviour. In
U. Stegmann (Ed.), Animal communication theory: Information and influence (pp.
This paper reviews different concepts and methods in 297e317). Cambridge, U.K.: Cambridge University Press.
Fischer, J., Hammerschmidt, K., Cheney, D. L., & Seyfarth, R. M. (2001). Acoustic
bioacoustics. The vocalizations used in the worked example were features of female chacma baboon barks. Ethology, 107, 33e54.
collected from fully habituated free-ranging chacma baboons be- Fischer, J., Hammerschmidt, K., Cheney, D. L., & Seyfarth, R. M. (2002). Acoustic
tween January 1998 and June 1999 in the Moremi Wildlife Reserve features of male baboon loud calls: Influences of context, age, and individuality.
Journal of the Acoustical Society of America, 111(3), 1465e1474.
in Botswana. The data collection conformed to ASAB/ABS's guide- Fitch, W. T., & Hauser, M. D. (1995). Vocal production in nonhuman primates e
lines for ethical research with animals. Acoustics, physiology, and functional constraints on honest advertisement.
American Journal of Primatology, 37(3), 191e219.
Freeberg, T. M. (2006). Social complexity can drive vocal complexity: Group size
Acknowledgments influences vocal information in Carolina chickadees. Psychological Science, 17(7),
557e561. http://dx.doi.org/10.1111/j.1467-9280.2006.01743.x.
Freeberg, T. M., Dunbar, R. I. M., & Ord, T. J. (2012). Social complexity as a proximate
We thank Simone Pika for the invitation to contribute to this and ultimate factor in communicative complexity. Philosophical Transactions of
special issue, and Brandon Wheeler for comments on the manu- the Royal Society B: Biological Sciences, 367(1597), 1785e1801. http://dx.doi.org/
script. Catherine Crockford kindly made the recording of the
Freeberg, T. M., & Lucas, J. R. (2012). Information theoretical approaches to chick-a-
chimpanzee pant-hoot available. We acknowledge financial sup- dee calls of Carolina chickadees (Poecile carolinensis). Journal of Comparative
port by the German Federal Ministry of Education and Research Psychology, 126(1), 68e81. http://dx.doi.org/10.1037/a0024906.
(BMBF) via the Bernstein Center for Computational Neuroscience Gamba, M., Friard, O., Riondato, I., Righini, R., Colombo, C., Miaretsoa, L., et al.
(2015). Comparative analysis of the vocal repertoire of Eulemur: A dynamic
(BCCN) Go €ttingen under Grant No. 01GQ1005C. We are grateful to time warping approach. International Journal of Primatology, 36(5), 894e910.
the Office of the President and the Department of Wildlife and http://dx.doi.org/10.1007/s10764-015-9861-1.
National Parks of the Republic of Botswana for permission to Green, S. (1975). Communication by a graded vocal system in Japanese monkeys. In
L. A. Rosenblum (Ed.), Primate behaviour (pp.1e102). New York, NY: Academic Press.
conduct research in the Moremi Wildlife Reserve. Hailman, J. P., Ficken, J. P., & Ficken, R. W. (1985). The “chick-a-dee” calls of Parus
atricapillus: A recombinant system of animal communication compared with
written English. Semiotica, 56(1985), 191e224. http://dx.doi.org/10.1515/
References semi.1985.56.3-4.191.
Hammerschmidt, K., & Fischer, J. (1998). The vocal repertoire of Barbary
Ackermann, H., Hage, S. R., & Ziegler, W. (2014). Brain mechanisms of acoustic macaques: A quantitative analysis of a graded signal system. Ethology, 104,
communication in humans and nonhuman primates: An evolutionary 203e216.
perspective. Behavioral and Brain Sciences, 37(6), 529e546. http://dx.doi.org/ Hammerschmidt, K., & Todt, D. (1995). Individual differences in vocalizations of
10.1017/S0140525X13003099. young Barbary macaques (Macaca sylvanus): A multi-parametric analysis to
Arcadi, A. C. (1996). Phrase structure of wild chimpanzee pant hoots: Patterns of identify critical cues in acoustic signalling. Behaviour, 132, 381e399.
production and interpopulation variability. American Journal of Primatology, Hedwig, D., Robbins, M. M., Mundry, R., Hammerschmidt, K., & Boesch, C. (2014).
39(3), 159e178. Acoustic structure and variation in mountain and western gorilla close calls: A
Arnold, K., & Zuberbühler, K. (2006). The alarm-calling system of adult male putty- syntactic approach. Behaviour, 151(8), 1091e1120. http://dx.doi.org/10.1163/
nosed monkeys, Cercopithecus nictitans martini. Animal Behaviour, 72, 643e653. 1568539X-00003175.
Bacher, J. (1994). Clusteranalyse. München, Germany: Oldenbourg Verlag. Hobaiter, C., & Byrne, R. W. (2014). The meanings of chimpanzee gestures. Current
Ben-David, S., v. Luxburg, U., & Pal, D. (2006). A sober look at clustering stability. In Biology, 24(14), 1596e1600. http://dx.doi.org/10.1016/j.cub.2014.05.066.
G. Lugosi, & H. Simons (Eds.), Proceedings of the 19th Annual Conference on Jang, J. S. R., & Sun, C. T. (1997). Neuro-fuzzy and soft computing e A computational
Learning Theory (COLT) (pp. 5e19). Berlin, Germany: Springer. http://dx.doi.org/ approach to learning and machine intelligence (Vol. 42). Upper Saddle River, NJ:
10.1007/11776420_4. Prentice Hall.
Blumstein, D. T., & Armitage, K. B. (1997). Does sociality drive the evolution of Jürgens, U. (2002). Neural pathways underlying vocal control. Neuroscience and
communicative complexity? A comparative test with ground-dwelling sciurid Biobehavioral Reviews, 26(2), 235e258.
alarm calls. American Naturalist, 150(2), 179e200. Jürgens, U. (2009). The neural control of vocalization in mammals: A review. Journal
Bouchet, H., Blois-heulin, C., & Lemasson, A. (2013). Social complexity parallels vocal of Voice, 23(1), 1e10. http://dx.doi.org/10.1016/j.jvoice.2007.07.005.
complexity: A comparison of three non-human primate species. Frontiers in Keenan, S., Lemasson, A., & Zuberbühler, K. (2013). Graded or discrete? A quanti-
Psychology, 4(July), 1e15. http://dx.doi.org/10.3389/fpsyg.2013.00390. tative analysis of Campbell's monkey alarm calls. Animal Behaviour, 85(1),
Bradbury, J. W., & Vehrencamp, S. L. (2011). Principles of animal communication (2nd 109e118. http://dx.doi.org/10.1016/j.anbehav.2012.10.014.
ed.). Sunderland, MA: Sinauer Associates. Kershenbaum, A. (2014). Entropy rate as a measure of animal vocal complexity.
Braune, P., Schmidt, S., & Zimmermann, E. (2008). Acoustic divergence in the Bioacoustics, 23(3), 195e208. http://dx.doi.org/10.1080/09524622.2013.850040.
communication of cryptic species of nocturnal primates (Microcebus ssp.). BMC Kershenbaum, A., Freeberg, T. M., & Gammon, D. E. (2015). Estimating vocal
Biology, 6(19), 1e10. http://dx.doi.org/10.1186/1741-7007-6-19. repertoire size is like collecting coupons: A theoretical framework with het-
Crockford, C., & Boesch, C. (2005). Call combinations in wild chimpanzees. Behav- erogeneity in signal abundance. Journal of Theoretical Biology, 373, 1e11. http://
iour, 142, 397e421. dx.doi.org/10.1016/j.jtbi.2015.03.009.

J. Fischer et al. / Animal Behaviour 134 (2017) 229e237 237

Krams, I., Krama, T., Freeberg, T. M., Kullberg, C., & Lucas, J. R. (2012). Linking social Pika, S., Liebal, K., Call, J., & Tomasello, M. (2005). The gestural communication of
complexity and vocal complexity: A parid perspective. Philosophical Trans- apes. Gesture, 5(1/2), 41e56.
actions of the Royal Society B: Biological Sciences, 367(1597), 1879e1891. http:// Pollard, K. a., & Blumstein, D. T. (2012). Evolving communicative complexity: In-
dx.doi.org/10.1098/rstb.2011.0222. sights from rodents and beyond. Philosophical Transactions of the Royal Society
Kroodsma, D. E. (1977). Correlates of song organization among North American B: Biological Sciences, 367(1597), 1869e1878. http://dx.doi.org/10.1098/
wrens. American Naturalist, 111(981), 995e1008. rstb.2011.0221.
LePrell, C. G., Hauser, M. D., & Moody, D. B. (2002). Discrete or graded variation Pozzi, L., Gamba, M., & Giacoma, C. (2010). The use of artificial neural networks to
within rhesus monkey screams? Psychophysical experiments on classification. classify primate vocalizations: A pilot study on black lemurs. American Journal
Animal Behaviour, 62, 47e62. of Primatology, 72(4), 337e348. http://dx.doi.org/10.1002/ajp.20786.
Maciej, P., Fischer, J., & Hammerschmidt, K. (2011). Transmission characteristics of Price, T., Ndiaye, O., & Fischer, J. (2014). Limited geographic variation in the acoustic
primate vocalizations: Implications for acoustic analyses. PLoS ONE, 6(8), structure of and responses to adult male alarm barks of African green monkeys.
e23015. Behavioural Ecology and Sociobiology, 68(5), 815e825. http://dx.doi.org/10.1007/
Maciej, P., Ndao, I., Hammerschmidt, K., & Fischer, J. (2013). Vocal communication in s00265-014-1694-y.
a complex multi-level society: Constrained acoustic structure and flexible call Price, T., Wadewitz, P., Cheney, D. L., Seyfarth, R. M., Hammerschmidt, K., & Fischer, J.
usage in Guinea baboons. Frontiers in Zoology, 10(1), 58. http://dx.doi.org/ (2015). Vervets revisited: A quantitative analysis of alarm call structure and
10.1186/1742-9994-10-58. context specificity. Scientific Reports, 5(13220), 1e11. http://dx.doi.org/10.1038/
Manser, M. B., Seyfarth, R. M., & Cheney, D. L. (2002). Suricate alarm calls signal srep13220.
predator class and urgency. Trends in Cognitive Sciences, 6(2), 55e57. Rendall, D., Kollias, S., Ney, C., & Lloyd, P. (2005). Pitch (F0) and formant profiles of
Marler, P. (1967). Animal communication signals. Science, 157(3790), 769. human vowels and vowel-like baboon grunts: The role of vocalizer body size
Marler, P., Kavanaugh, J. F., & Cutting, J. E. (1975). On the origin of speech from and voice-acoustic allometry. Journal of the Acoustical Society of America, 117(2),
animal sounds. In The role of speech in language (pp. 11e37). Cambridge, MA: 944e955.
MIT Press. Rousseeuw, P. J. (1987). Silhouettes: A graphical aid to the interpretation and
Maynard Smith, J., & Harper, D. (2003). Animal signals. Oxford, U.K.: Oxford Uni- validation of cluster analysis. Journal of Computational and Applied Mathematics,
versity Press. 20, 53e65. http://dx.doi.org/10.1016/0377-0427(87)90125-7.
McComb, K., & Semple, S. (2005). Coevolution of vocal communication and sociality Rowell, T. E., & Hinde, R. A. (1962). Vocal communication by the rhesus monkey
in primates. Biology Letters, 1, 381e385. (Macaca mulatta). Proceedings of the Zoological Society of London, 138, 279e294.
McCowan, B., Doyle, L. R., Jenkins, J. M., & Hanser, S. F. (2005). The appropriate use of Scheider, L., Liebal, K., On ~ a, L., Burrows, A., & Waller, B. (2014). A comparison of
Zipf's law in animal communication studies. Animal Behaviour, 69(1), 1e7. facial expression properties in five hylobatid species. American Journal of Pri-
http://dx.doi.org/10.1016/j.anbehav.2004.09.002. matology, 76(7), 618e628. http://dx.doi.org/10.1002/ajp.22255.
McCowan, B., Hanser, S. F., & Doyle, L. R. (1999). Quantitative tools for comparing Schneider, C., Hodges, J. K., Fischer, J., & Hammerschmidt, K. (2008). Acoustic niches
animal communication systems: Information theory applied to bottlenose of Siberut primates. International Journal of Primatology, 29, 601e613. http://
dolphin whistle repertoires. Animal Behaviour, 57, 409e419. dx.doi.org/10.1007/s10764-007-9181-1.
Meise, K., Keller, C., Cowlishaw, G., & Fischer, J. (2011). Sources of acoustic variation: Scott-Phillips, T. C. (2008). Defining biological communication. Journal of Evolu-
Implications for production specificity and call categorization in chacma ba- tionary Biology, 21(2), 387e395. http://dx.doi.org/10.1111/j.1420-
boon (Papio ursinus) grunts. Journal of the Acoustical Society of America, 129(3), 9101.2007.01497.x.
1631e1641. Sewall, K. B. (2015). Social complexity as a driver of communication and cognition.
Meyer, D., Hodges, K., Rinaldi, D., Wijaya, A., Roos, C., & Hammerschmidt, K. (2012). Integrative and Comparative Biology, 55(3), 384e395. http://dx.doi.org/10.1093/
Acoustic structure of male loud-calls support molecular phylogeny of Sumatran icb/icv064.
and Javanese leaf monkeys (genus Presbytis). BMC Evolutionary Biology, 12(1), 16. Seyfarth, R. M., Cheney, D. L., & Marler, P. (1980). Monkey responses to three
Moody, M. I., & Menzel, E. W. (1976). Vocalizations and their behavioral contexts in different alarm calls: Evidence of predator classification and semantic
the tamarin Saguinus fuscicollis. Folia Primatologia, 25, 73e94. communication. Science, 210, 801e803.
Ord, T. J., & Martins, E. P. (2006). Tracing the origins of signal diversity in anole Shannon, C. E., & Weaver, W. (1949). The mathematical theory of communication.
lizards: Phylogenetic approaches to inferring the evolution of complex behav- Urbana, IL: University of Illinois Press.
iour. Animal Behaviour, 71(6), 1411e1429. http://dx.doi.org/10.1016/ Struhsaker, T. T. (1967). Auditory communication among vervet monkeys (Cerco-
j.anbehav.2005.12.003. pithecus aethiops). In S. A. Altmann (Ed.), Social communication among primates
Owren, M. J., & Bernacki, R. H. (1998). Applying linear predictive coding (LPC) to the (pp. 281e324). Chicago, IL: University of Chicago Press.
frequency analysis of animal acoustic signals. In S. L. Hopp, & C. S. Evans (Eds.), Sugiura, H., Tanaka, T., & Masataka, N. (2006). Sound transmission in the habitats of
Recent advances in the study of animal communication (pp. 129e162). New York, Japanese macaques and its possible effect on population differences in coo calls.
NY: Springer. Behaviour, 143(8), 993e1012. http://dx.doi.org/10.1163/156853906778623617.
Owren, M. J., Linker, C., Zimmermann, E., Newman, J. D., & Jürgens, U. (1995). Some Thinh, V. N., Hallam, C., Roos, C., & Hammerschmidt, K. (2011). Concordance be-
analysis methods that may be useful to acoustic primatologists. In Current topics tween vocal and genetic diversity in crested gibbons. BMC Evolutionary Biology,
in primate vocal communication (pp. 1e28). New York, NY: Plenum Press. 11(1), 36. http://dx.doi.org/10.1186/1471-2148-11-36.
Palombit, R. A. (1992). A preliminary study of vocal communication in wild long- de Waal, F. B. M. (1988). The communicative repertoire of capitve bonobos (Pan
tailed macaques (Macaca fascicularis). 1. Vocal repertoire and call emission. paniscus) compared to that of chimpanzees. Behaviour, 106(27), 183e251.
International Journal of Primatology, 13(2), 143e182. Wadewitz, P. (2015). Processing of graded signaling systems (PhD thesis), Go €ttingen,
Papworth, S., Bo €se, A.-S., Barker, J., Schel, A. M., & Zuberbühler, K. (2008). Male blue Germany: Georg-August-University Go €ttingen.
monkeys alarm call in response to danger experienced by others. Biology Letters, Wadewitz, P., Hammerschmidt, K., Battaglia, D., Witt, A., Wolf, F., & Fischer, J. (2015).
4(5), 472e475. http://dx.doi.org/10.1098/rsbl.2008.0299. Characterizing vocal repertoiresdhard vs. soft classification approaches. PLoS
Parr, L. A., & Waller, B. M. (2006). Understanding chimpanzee facial expression: One, 10(4), e0125785. http://dx.doi.org/10.1371/journal.pone.0125785.
Insights into the evolution of communication. Social Cognitive and Affective Wheeler, B. C., & Fischer, J. (2012). Functionally referential signals: A promising
Neuroscience, 1(3), 221e228. http://dx.doi.org/10.1093/scan/nsl031. paradigm whose time has passed. Evolutionary Anthropology, 21, 195e205.
Pfefferle, D., Brauch, K., Heistermann, M., Hodges, J. K., & Fischer, J. (2008). Female Xu, D., Keller, J. M., Popescu, M., & Bondugula, R. (2008). Applications of fuzzy logic in
Barbary macaque (Macaca sylvanus) copulation calls do not reveal the fertile bioinformatics. London, U.K.: Imperial College Press.
phase but influence mating outcome. Proceedings of the Royal Society B: Bio- Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8(3), 338e353.
logical Sciences, 275(1634), 571e578. http://dx.doi.org/10.1098/rspb.2007.1499, Zadeh, L. A. (2008). Is there a need for fuzzy logic? Information Sciences, 178(13),
7P88K06U5U683210 [pii]. 2751e2779. http://dx.doi.org/10.1016/j.ins.2008.02.012.
Pfefferle, D., & Fischer, J. (2006). Sounds and size: Identification of acoustic variables Zipf, G. K. (1949). Human behavior and the principle of least effort (Vol. 6). Cambridge,
that reflect body size in hamadryas baboons, Papio hamadryas. Animal Behav- MA: Addison-Wesley. http://dx.doi.org/10.1002/1097-4679(195007)6:3<306::
iour, 72(1), 43e51. http://dx.doi.org/10.1016/j.anbehav.2005.08.021. AID-JCLP2270060331>3.0.CO;2-7.
Pfefferle, D., Heistermann, M., Pirow, R., Hodges, J. K., & Fischer, J. (2011). Estrogen Zuberbühler, K. (2006). Alarm calls. In K. Brown (Ed.), Encyclopedia of language and
and progestogen correlates of the structure of female copulation calls in semi- linguistics (pp. 143e155). Oxford, U.K.: Elsevier.
free-ranging Barbary macaques (Macaca sylvanus). International Journal of Pri- Zuberbühler, K., Noe, R., & Seyfarth, R. M. (1997). Diana monkey long-distance calls:
matology, 32(4), 992e1006. http://dx.doi.org/10.1007/s10764-011-9517-8. Messages for conspecifics and predators. Animal Behaviour, 53, 589e604.