Play-solicitation gestures in chimpanzees in the wild: Flexible adjustment to social circumstances and individual matrices

Social play is a frequent behaviour in great apes and involves sophisticated forms of communicative exchange. While it is well established that great apes test and practise the majority of their gestural signals during play interactions, the influence of demographic factors and kin relationships between the interactants on the form and variability of gestures are relatively little understood. We thus carried out the first systematic study on the exchange of play-soliciting gestures in two chimpanzee (Pan troglodytes) communities of different subspecies. We examined the influence of age, sex and kin relationships of the play partners on gestural play solicitations, including object-associated and self-handicapping gestures. Our results demonstrated that the usage of (i) audible and visual gestures increased significantly with infant age, (ii) tactile gestures differed between the sexes, and (iii) audible and visual gestures were higher in interactions with conspecifics than with mothers. Object-associated and self-handicapping gestures were frequently used to initiate play with sameaged and younger play partners, respectively. Our study thus strengthens the view that gestures are mutually constructed communicative means, which are flexibly adjusted to social circumstances and individual matrices of interactants. The study was conducted at two different chimpanzee communities in the wild: Kanyawara in Kibale National Park, Uganda and Taï South in Taï National Park, Côte d’Ivoire. Detailed descriptions of the study areas can be found in Wrangham et al. [56] and Boesch & Boesch-Achermann [46], respectively. During the two study periods, the size of the Kanyawara group varied between 53 and 56 individuals, and 26 and 33 individuals in Taï South, respectively. In addition, we had access to long-term data concerning the chimpanzees’ demography and relatedness. We observed play interactions of a total of 16 infants (10 from Kanyawara, 6 from Taï South) with their mothers and conspecifics, with ages ranging from 9 to 74 months (electronic supplementary material, table S1). Conspecifics that the infants interacted with during social play included 10 juveniles/sub-adults (three females, seven males) and 11 adult individuals (eight females, three males) at Kanyawara, and 8 juveniles/sub-adults (one female, seven males) and four adult individuals (four females, no males) at Taï South. In terms of age classes, individuals were categorized as juvenile/sub-adults if aged between 6 and 13/15 years (females/males) and adults if aged 14/16 years or older (females/males) [46]. Observations were made on chimpanzees during four periods between October 2012 and June 2014 (Kanyawara: Mar–May 2013; Mar–June 2014; Taï South: Oct–Dec 2012; Oct–Dec 2013). We used a focal next to a behaviour sampling approach, that is one individual was observed for a set period of time, while play behaviour of non-focal individuals in the immediate proximity was also recorded [57]. All play interactions of infants (i.e. mother–infant interactions and infant–conspecific interactions) were recorded using a digital high-definition camera (Canon HF M41) with an external unidirectional microphone (Sennheiser K6). This method resulted in a total of 81.9 hours (Kanyawara: 44.3, mean ± s.d. per infant = 6.6 ± 4.6; Taï South: 37.6, mean ± s.d. = 10.7 ± 5.4, see also electronic supplementary material, table S1) of video footage of play interactions recorded during approximately 1154 hours (Kanyawara: 557, Taï South: 597) of focal observations. To establish the signal repertoires of chimpanzees used to solicit play and enable subsequent analyses, a total of 643 high-quality video files of play interactions were coded using the program Adobe Premiere Pro CS4 (v. 4.2.1.). Behavioural definitions were based on established ethograms of two long-term studies of Eastern Chimpanzees at Gombe [22] and Mahale [58] and several gesture studies [21,59,60]. A specific coding scheme was developed based on parameters used in previous work on great ape gesturing [28,36,61]. ‘Play solicitations’ comprised dyadic play initiation from the start of a play bout, and also dyadic play reinitiation after social play was paused or interrupted by a third individual for at least 10 s. While coding all play interactions, we differentiated play-soliciting gestures from physical actions. An action was defined as any behaviour that resulted in play through direct manipulation of another’s body via physical force (e.g. throw on) or one’s own locomotion (e.g. move backwards). Contrarily, a gesture was defined as directed, mechanically ineffective movement of the extremities, the body or body postures that elicited (requested) a voluntary response by the recipient [62]. For our analyses, we only included play-initiating gestures that were accompanied by key characteristics of intentional communication: sensitivity to the recipient’s attentional state, response waiting, apparent satisfaction of the signaller and goal persistence (for definitions, see [36,63,64]). Signals were clustered into three categories: audible (signals generate a sound while being performed, e.g. slap ground), tactile (signals include physical contact with the recipient, e.g. touch) and visual (signals generate a mainly graphic component, e.g. raise arm) gestures [28]. To identify play solicitations, the behaviour of both the signaller and the recipient throughout the interaction, from first initiating action/gesture until the start of play, was taken into account to assess the success of communicative attempts [65]. Finally, for each signal or action case, we coded the following parameters: signaller’s and recipient’s age (range = 0–57 years), if signaller = infant: infant age (range = 9–69 months), age class relative to recipient (three levels: same age class, older age class, younger age class), sex of signaller and recipient (two levels: male, female), kin relationship between play partners (three levels: infant–mother, infant–maternal kin, infant–non-kin); play intensity (three levels: low—touching and tickling, intermediate—wrestling and biting, high—rough and tumble). Fifteen per cent of all coded interactions were coded for accuracy by a second observer and tested using the Cohen’s κ coefficient to ensure inter-observer reliability [57]. The following results were found: a ‘very good’ level of agreement for play intensity (κ = 0.868), gesture type (κ = 0.827), signal category (κ = 0.884), object-associated gesture (κ = 0.916) and self-handicapping gesture (κ = 0.886) and a ‘good’ level of agreement for intentional usage of signal (κ = 0.703). To assess the influence on sampling size, we plotted the cumulative numbers of observed gesture types over time for each study site. If an asymptote was reached (i.e. no further gesture types were observed), we concluded that we had observed the individual’s full repertoire for the communicative function of play solicitation [36]. To test to which extent sex, age and the kin relationship between the play partners influenced (1) intensity of solicited play, (2) frequency of gestures overall (response variables: audible, tactile, visual gesture), (3) frequency of object-associated gestures, and (4) frequency of self-handicapping gestures, we used generalized linear mixed models (GLMMs; [66]) with a Poisson (1) or binomial error structure (2–4) and log (1) or logit link function (2–4). Since infant age varied considerably between infants, we used the method of within-subject centring [67] to tease apart whether the effect of infant age was particularly relevant within and/or between infants. Hence, we included into the model the average age of each infant (being constant across all data points of the respective individual; between-age) and also the difference between the infant’s actual age and its average age (within-age). To rule out that age effects do not simply result from higher rates of conspecific-directed signalling with age, we initially included the interactions between relation and both age variables into the model (analyses 1 to 3, removed if non-significant). To control for confounding effects, we also included site as further fixed effect into the model. As random effects (intercepts) we included the identities (ID) of signaller, recipient and the play dyad. To keep type 1 error rates at the nominal level of 5%, we also included the random slopes components of within-age, age difference, sex of signaller/recipient and kinship within signaller ID and/or recipient ID [68,69]. For the other fixed effects, we did not include random slopes because they were usually constant within signaller and recipient ID. We also did not include correlations between random slopes and random intercepts to keep model complexity at an acceptable level and because neglected random slopes do not compromise type 1 error rates [69]. The models were implemented in R [70] using the function glmer of the package lme4 [71]. To test the overall significance of our key test predictors [72], we compared the full models with the respective null models comprising only the control predictor (study site) and all random effects using a likelihood ratio test [73]. Prior to running the models, we z-transformed between-age, within-age and age difference [74]. To control for collinearity, we determined variance inflation factors (VIF; [75]) from a model including only the fixed main effects using the function vif of the R package car [76]. This revealed collinearity to not be an issue (maximum VIF = 1.49). Confidence intervals were derived using the function sim of the R package arm [77]. Tests of the individual fixed effects were derived using likelihood ratio tests (R function drop1 with argument ‘test’ set to ‘Chisq’). All statistical analyses were performed using the R statistical package, version R.3.1.1 [70], with a level of significance set at 0.05. For each of our four analyses, we used different datasets: for analyses 1 and 2, we included play signals of infants towards all possible play partners, including mothers, and used age and sex of the signaller as well as relation to the recipient (mother, maternal kin, non-kin) as fixed effects (for details see dataset in electronic supplementary material, S2). Analysis 3 used the same dataset, but here we excluded the few gestures towards mothers, since object-associated gestures did not seem to play a role in infant–mother play solicitation. Here, we additionally included age difference and sex of the recipient as additional test predictors. Finally, for analysis 4, we included gesture cases by or towards infants to examine the role of self-handicapping (i.e. signallers could be of all age classes, for details see dataset in electronic supplementary material, S3). This analysis used the same test predictors as utilized in analysis 3.