A non-human primate combinatorial system for long-distance communication

Abstract

Complex vocal systems are thought to evolve if individuals are regularly challenged by complex social decision-making, the social complexity hypothesis. We tested this idea on a West African forest non-human primate, the Olive colobus monkey, a highly cryptic species with very little social behavior and very small group sizes, factors unlikely to favor the evolution of complex communication. The species also has an unusual fission-fusion social system, with group members regularly spending considerable amounts of time with neighboring groups. As predicted by the social complexity hypothesis, we only found a very basic repertoire of two call types in this species, produced by both males and females. However, the calls were astonishingly loud, never uttered alone but in syntactically structured sequences assembled along a set of rules. We concluded that the Olive colobus monkeys have evolved a combinatorial system to interact with distant group members.

Keywords: Behavioral neuroscience; Biological sciences; Linguistics; Social sciences.

Graphical abstract

Figure 1

Results from Olive colobus monkey call classification plotted in acoustic parameter space with corresponding spectrograms Each data point represents one call (circles = “A” calls, triangles = “B” calls). Colors represent the cluster attributions from the Partitioning Around Medoids analysis. Each ellipse was determined using the Student’s t-distribution with 95% of confidence level; center dots represent the mean position of the data points for each cluster. Spectrograms were extracted from Raven Pro software v1.6.4, background noise removed with Adobe Photoshop software v23.0.0.

Figure 2

Density histogram and kernel density estimate curve of Olive colobus inter-call intervals The x axis was log transformed. The blue and red clusters were obtained via unsupervised model-based clustering and represent the distributions of inter-call intervals (N = 1,119) within and between sequences, respectively.

Figure 3

Trie of N = 284 call sequences grouped into 48 sequence types in response to different disturbances: leopard and eagle calls presented from the ground or within a tree, falling tree sounds, and visual leopard model Each colored node represents a call (green = “A”; yellow = “B”). The “root” node represents the start and “⊣” marks the end of a sequence. The number at the end of each branch represents the sample size of the corresponding sequence in all datasets. The x-coordinates represent the ordinal position of the call in the sequence.

Figure 4

Spectrographic representations of the four basic sequence types to different types of dangers “AAA” and “ABA” sequences shown here were given in response to eagle shrieks, the “BA” sequence was given in response to a leopard growl and the “BAA” sequence was given in response to a falling tree sound. Spectrograms were extracted from Raven Pro software v1.6.4, background noise removed with Adobe Photoshop software v23.0.0.

Figure 5

Context-specificity of sequence initiation in Olive colobus call production Each data point represents one trial, and vocal production from only one individual of the group. To reduce overplotting and enhance visibility, random noise was added to the x-coordinate of data points. Embedded black dots and vertical lines indicate means and bootstrapped 95% confidence intervals from model estimation, respectively. Sample sizes: Leopard growls N = 37 trials, Eagle shrieks N = 25 trials, Falling tree sounds N = 42 trials. Three outliers not depicted (N = 2 leopard trails with 13 “B”-initiated sequences, and N = 1 leopard trial with 9 “B”-initiated sequences).

Figure 6

Context-specificity of overall sequence patterns in Olive colobus call production Each data point represents one trial, and vocal production from only one individual of the group. To reduce overplotting and enhance visibility, random noise was added to the x-coordinate of each data point. Embedded black dots and vertical lines indicate means and bootstrapped 95% confidence intervals from model estimation, respectively. Sample sizes: Leopard growls N = 37 trials, Eagle shrieks N = 25 trials, Falling tree sounds N = 42 trials. One outlier not depicted (N = 1 leopard trail with 12 “BA” sequences).

Figure 7

Context-specificity of sequence termination in Olive colobus call production Each data point represents one trial, and vocal production from only one individual of the group. To reduce overplotting and enhance visibility, random noise was added to the x-coordinate of each data point. Embedded black dots and vertical lines indicate means and bootstrapped 95% confidence intervals from model estimation, respectively. Sample sizes: Leopard growls N = 37 trials, Eagle shrieks N = 25 trials, Falling tree sounds N = 42 trials, one outlier not depicted (N = 1 leopard trail with 12 “BA-gram” terminated sequences).