White-handed gibbons discriminate context-specific song compositions

Study site and subjects

This study was conducted in the Mo Singto-Klong E-Tau area of Khao Yai National Park, Thailand (101°22′E, 14°26′N), 130 km North-East of Bangkok. Data were collected from December 2012 to August 2014. Thirteen fully habituated groups of white-handed gibbons were monitored, each comprising a primary male, his mated female with her offspring and (in 5 cases) a secondary male, totalling N = 53 individuals at the time of the study. Due to a number of constraints, it was only possible to conduct playback experiments with six of the 13 groups (Table 1).

Terminology

Following Raemaekers, Raemaekers & Haimoff’s (1984) terminology we distinguished three sequence types within each song: the introductory sequence (series of soft ‘hoo’ notes, followed by combinations of other note types, such as ‘oo’, ‘wa’, ‘leaning wa’ and ‘wa-oo’); the great call sequence (idiosyncratic female call sequence, usually followed by her male’s ‘coda’ response). In duet songs, the first great call sequence usually appears within the first 2 min. Great call sequences can be repeated multiple times (about once every 1–2 min) (Raemaekers, Raemaekers & Haimoff, 1984; Clarke, Reichard & Zuberbühler, 2006; Terleph, Malaivijitnond & Reichard, 2016), in which case they are separated by an interlude sequence (any notes given after a great call sequence, including the final one) (Ellefson, 1968; Raemaekers, Raemaekers & Haimoff, 1984) (Figs. 1 and 2A).

The same three sequence types can also be found in predator songs although, overall, they differ in length and are produced with the contribution of most group members. When comparing predator songs with duet songs, for the introductory sequence the initial ‘hoo’ notes series are longer and contain more ‘hoo’ notes, followed by fewer ‘leaning wa’ notes and more ‘hoo’ notes (Clarke, Reichard & Zuberbühler, 2006). The great call sequence is also different, mainly because males respond more rapidly with their answering coda (Clarke, Reichard & Zuberbühler, 2006). Regarding the interlude sequence, predator songs contain more ‘sharp wow’ notes, especially towards the end of the song, compared with duet songs (Clarke, Reichard & Zuberbühler, 2006) (Figs. 1 and 2B).

Stimulus collection

Duet songs were recorded on an all-occurrence basis during all-day follows of study groups (Table 1) until at least one song suitable as playback stimulus was recorded, that is, a high-quality song with minimum background noise, singing individuals at a maximum distance of 30 m from the recording device. Predator songs were induced by presenting a realistic, life-size clouded leopard (Neofelis nebulosa) model to each group following an established protocol (Fig. 3, Clarke, Reichard & Zuberbühler, 2006). Once a group was located and before positioning the model, we ensured that on the same day the group had (a) already produced at least one duet song more than one hour earlier (to verify a basic motivation to sing), (b) not yet produced a predator song (nor its direct neighbours), (c) not had a natural predator encounter since the beginning of the day-follow, nor heard other species’ alarm calls within the last hour and (d) not had an intergroup encounter with a neighbouring group. If these conditions were met, we positioned the predator model on the group’s anticipated travel direction outside their visual range. We then continuously recorded their vocal behaviour and scored the presence of any non-vocal anti-predator behaviour on an all-occurrence basis (branch dropping, defaecation, vigilance). Duet and predator songs were recorded using directional microphones (Sennheiser MKH 815T & Sennheiser ME66) with windshields connected to a digital stereo recorder (Marantz PMD660; settings 44.1 kHz, 16 bits) from December 2012 to August 2014.

Experimental protocol

Each group was tested once with each stimulus type, which resulted in a total of 12 trials (N = 6 duet songs; N = 6 predator songs, Table S1; minimum interval between trials: 1 week), all broadcasted before 12:00 local time (to match timing of natural duet song production). Prior to playback experiments, we measured the peak intensity of female great call climaxes in spontaneous duet songs (i.e. loudest notes, Terleph, Malaivijitnond & Reichard, 2016) at an estimated recording distance of 10–20 m using a REED ST-805 (REEDinstruments, Wilmington, NC, USA) sound pressure metre (frequency range 31.5 Hz–8 kHz, measuring level range 30–130 dB, 0.1 dB resolution, accuracy ± 1.5 dB). We measured three great call climaxes per female from the six song-providing groups (Table 1), which resulted in a mean sound pressure level of 78.2 ± 8.0 dB (n = 18; dB SPL, A-weighting sound pressure levels for general sound level measurements, and 125 ms fast time weighting). We then broadcasted songs such that subjects always heard recordings from one of their direct neighbours (Table S2), with comparable natural audibility (tested at each playback location with a decibel metre, matching climaxes SPL measurements, with real time adjustments in coordination with both experimenters depending on weather conditions on the testing day) and from spatially realistic locations 15–20 m within the canopy from where the song providing group had been seen before within the respective territories.

We standardised the distance between the speaker and subjects to about 150 m (mean ± SD: 149 ± 17 m), with playback conditions randomly counterbalanced (Table S2). Stimuli were broadcasted when the same conditions as for predator model presentation had been met, using a Climate CL60-T2 speaker connected to a Kenwood KAC-5203 amplifier, in conjunction with a Roland R-05 digital player.

Playback trials were carried out from spatially realistic locations, that is, from the home range of the song-providing group towards the home range of the target group. In doing so, we took a number of precautions such that the song-providing group could not overhear its own song. Before each trial, we ensured that the song-providing group was not in the vicinity of the speaker (>100 m radius). We then monitored the area for a period of 1 h to further ensure that the song-providing group was not nearby. For each trial, the speaker was positioned in the overlapping zone between the song-providing and target group, such that it was facing away from the home range centre of the song-providing group towards the target group.

Data collection

Due to the difficult visual conditions in the forest, it was impossible to continuously video-tape the entire duration of trials nor to film all group members simultaneously. We therefore decided to restrict observations to the primary male of each group. Males are easily identifiable by their body hair colouration, facial features and genitals. Primary males were video recorded as long as possible (i.e. until they moved out of sight) using a Panasonic SDR-S26 Camcorder. Videos were coded using ELAN software (ELAN (V5.2) Nijmegen: Max Planck Institute for Psycholinguistics). Because the speaker location was not visible on the video clips (outside camera range) it was necessary for the experimenter to comment on the male’s gazing direction during filming, which made blind coding redundant. All video recordings are available on figshare (https://doi.org/10.6084/m9.figshare.12363050.v1).

Regarding long-term effects, we collected 5-min scan samples of the primary male’s behavioural activities, gaze directions, body positions, elevations (m) and proximities to their female partner (m) during 1 h after each trial (i.e. 13 scans per trial; Table 2). Furthermore, we scored all defaecation/urination and branch dropping events over a two-hour period using all occurrence sampling.

Vocal responses

We digitised, analysed and compared songs given in response to both playback conditions, using Raven Pro 64 1.4 (Cornell laboratory of Ornithology, Ithaca, NY, USA). For the introductory sequence, we determined the duration of the initial ‘hoo’ notes series (s) and the corresponding number of ‘hoo’ notes, the type of the first ten notes following the ‘hoo’ series, and the duration of the introductory sequence (i.e. latency to the first female great call). We measured the interval between the female great call and the male coda reply (s), the total song duration (s), and determined whether a neighbouring group also produced a song and its type. Finally, we identified the presence of ‘sharp wow’ notes and we measured the latency to the first ‘sharp wow’ note (i.e. time elapsed in seconds between the onset of the song bout and the first ‘sharp wow’ emitted).

This study was approved by the School of Psychology Ethics Committee of St. Andrews University. Approval was given on the understanding that the ASAB guidelines for the Treatment of Animals in Behavioural research and Teaching are adhered to (n°16112011). The research permit was delivered by the National Research Council of Thailand (NRCT, n°0002/5841).

Behavioural responses

We compared behavioural responses within subjects and across playback conditions; the primary male’s latency and duration of first looks towards the speaker, the occurrence of defecations/urinations and branch droppings, the average distance to their female mate and the canopy heights (medians across all scan samples; Table 2). For categorical data (i.e. activity, body position and gaze), we summed up and calculated for each individual the proportion of each behaviour within the categories (see Table 2) and compared the behavioural pattern across playback conditions.

Vocal responses

We compared the number of introductory ‘hoo’ notes and the duration of the introductory ‘hoo’ notes series, the number of other relevant ‘hoo’ and ‘leaning wa’ notes within the first ten notes following the introductory ‘hoo’ series, and the introductory sequence duration, within groups and across conditions. For the great call sequence, we compared male response delays to the female great calls. Finally, we compared the total song duration between playback conditions, identified the presence of ‘sharp wow’ notes, and measured the latency to first ‘sharp wow’ note produced.

Statistical procedures

Due to small sample sizes we opted for non-parametric statistics. Wilcoxon matched-pair signed-rank tests were performed for behavioural data analysis, with exact significance levels reported (Siegel & Castellan, 1988; Mundry & Fischer, 1998). For vocal data, we used Kruskal–Wallis rank sum tests with a Benjamini & Hochberg procedure to correct for multiple testing (Benjamini & Hochberg, 1995). Post-hoc tests were either Wilcoxon rank sum tests with Benjamini & Hochberg p-value adjustments or Dunn (1964)’s tests with Benjamini & Hochberg p-value adjustments for eventual ties. To compare the type of the first 10 notes produced across contexts we used a Pearson’s Chi-squared test followed by Chi-squared post-hoc tests with Benjamini & Hochberg p-value adjustments. Statistical analyses were performed using R V3.5.1 (R core Team, 2018) with the significance level set at 0.05.

Vocal behaviour

Non-vocal behaviour

We were able to record the immediate behavioural responses of primary males in 5 of 6 groups (Table 1). All males responded by turning their heads towards the speaker, albeit with no latency differences across playback conditions (median duet: 1.1 ± 1.8 s, predator 2.8 ± 3.3 s, V = 2, P = 0.188, N_duet = 5, N_Predator = 5, Wilcoxon matched-pair signed-rank test, Fig. 4A). Additionally, we found a trend (although not significant) towards longer gaze duration in the predator than the duet song condition (median duet: 2.0 ± 1.4 s, predator: 12.6 ± 6.1 s, V= 0, P = 0.063, N_duet = 5, N_Predator = 5, Wilcoxon matched-pair signed-rank test, Fig. 4B).

For long-term behavioural responses, we collected data on all six primary males and found no differences across playback conditions in grooming, resting and displacement activities but a significant difference in feeding, with individuals less likely to engage in feeding activities after predator than duet song playbacks (Table 4). Regarding anti-predator behaviours, we found no differences in canopy use, distance between mates, and number of branch droppings across conditions. However, males were more vigilant and defaecated significantly more often following predator compared with duet song playbacks (Table 4).

Following playback of a predator song, males increased their vigilance activity (Fig. 5), directed more gazes towards the ground (Fig. 6A) and less towards the upper canopy (Fig. 6B) compared with duet treatment (Table 4).

Summary

White-handed gibbons produce two structurally distinct songs in context-specific ways; duet songs (in non-predatory contexts) and predator songs (to clouded leopards and other predators). The two song types differ in the overall duration, frequency and distribution of specific notes (‘hoo’, ‘leaning wa’, ‘sharp wow’) and in the location of the female great calls and male replies within each song. In this study, we investigated whether individuals discriminated between these two structurally different song types and whether they could infer meaning from them. We found several lines of evidence in favour of such an ability. First, playbacks of duet songs reliably elicited natural duet song replies (identifiable by several acoustic parameters) in neighbouring groups and in more distant groups, similar to how natural duet song spread throughout the forest (Raemaekers & Raemaekers, 1985b; J. Andrieu, 2012–2014, unpublished data a). Second, playbacks of predator songs never triggered duet songs in any group, but occasionally predator song replies (identifiable by several acoustic parameters) in one of six neighbouring groups and two non-neighbouring distant groups. Finally, subjects consistently showed anti-predator behaviours (vigilance, ground scanning, defaecation) and a tendency for longer first look towards the speaker after predator compared to duet song playbacks. Based on these data, we concluded that white-handed gibbon song conveys key information about the world, which is made accessible to recipients by a number of structural regularities. This conclusion fits with previous research by Clarke, Reichard & Zuberbühler (2006) who first demonstrated the presence of structural differences in white-handed gibbon songs.

Singing as anti-predator behaviour

Similar to other large cats, clouded leopards are opportunistic predators that attack both terrestrial and arboreal species, including primates (Rabinowitz, Andau & Chai, 1987; Grassman, 2001). Hence, a somewhat surprising finding was that subjects remained mostly silent to others’ predator songs, despite showing strong anti-predator behaviour (males and females appeared to behave in the same way, i.e. ground scanning, vigilance, defaecation). The lack of vocal response may be part of a cryptic strategy to conceal the group’s location when a dangerous stalking predator is presumed in the vicinity (Aguilar de Soto et al., 2012; Grow, 2019). However, this does not explain why 1 of 6 target groups and two distant groups still responded with predator songs to the playbacks. It is possible that gibbons pursue a flexible vocal strategy, altering between ‘crypsis’ and ‘perception advertisement’ depending on perceived personal risk, the ability to benefit neighbouring relatives, and the likely dissuasive effect on the predator itself (Zuberbühler, Jenny & Bshary, 1999; Clarke, Reichard & Zuberbühler, 2006).

Equally relevant is the fact that the three predator song responses were shorter than natural predator songs (Tables S3 and S4). We can think of several explanations for this finding. First, as mentioned already, it is possible that groups tried to minimise their own exposure to the predator if they decided to respond to another group’s predator song. Second, differences in predator song duration may function as indicators for perceived urgency, with longer songs indicating more serious threats than shorter songs. We find this less likely to be an evolved function since listeners would have to wait for (and compare) considerable amounts of time periods before extracting the relevant information. Finally, differences in song duration may be linked to how callers perceive the predator (visually, linked to mobbing the predator vs. acoustically, linked to localising the predator). A Direct observation of a real encounter with a tiger is in line with this hypothesis (Uhde & Sommer, 2002). In this instance, group A uncommonly travelled backward towards the tiger’s location (spotted 50 m away) and sang for at least 1 h and a half, suggesting that singing primarily serves first and foremost as a predator deterrence device and second as a conspecific warning signal if the exact location of the predator is unknown and groups feel reasonably safe.

Singing as territorial behaviour

In related research (J. Andrieu, 2012–2014, unpublished data), we have shown that spatial proximity between two neighbouring groups tends to lead to duet song overlap, due to the fact that the second group refuses to delay singing until the first group has finished their duet song. This behaviour is attenuated by kinship, to the effect that related individuals are more likely to respect each other’s duets, even if produced at close distances. In the current study, all study groups started producing duet songs while the playback duet song was still being broadcast, suggesting that the manipulation was perceived as a territorial threat. Unfortunately, we could not statistically analyse the effect of genetic relatedness in this study because the sample size was too small (N = 6 groups).

Singing as compositional behaviour

Although our study has focussed on song comprehension, it has also generated a more detailed picture of the structural composition of white-handed gibbon songs. Clarke, Reichard & Zuberbühler (2006) already noted that the duet songs of gibbon groups that were not well habituated to human observers contained elements that were normally found in predator songs, notably ‘sharp wows’. In our study, all groups were fully habituated to human presence, yet some groups still produced ‘sharp wow’ notes in their duet song replies to playbacks of neighbouring duet songs, but also in 4 of 5 natural duet songs (Table 3), of which 3 were involved in duet counter-signing exchanges with previous duetting direct neighbours. Another structural subtlety concerned the use of ‘wa-oo’ notes. This note type was near absent in predator songs but common in the early parts of the duet songs, especially the ones given in response to duet song playbacks. We attribute these findings to the fact that our experimental design consisted of playbacks of song recordings at relatively close distances (about 150 m), which may have been perceived as a social threat by some groups, either territorial or risk of partner defection. Future work is required to test whether these notes are actively used to describe events in hierarchically structured ways (main: predatory threat y/n; subsidiary: social threat y/n), similar to how humans represent natural events as tree structures in both cognition and language (Zuberbühler, 2019b).