Unpeeling the layers of language: Bonobos and chimpanzees engage in cooperative turn-taking sequences

Human language is a fundamentally cooperative enterprise, embodying fast-paced and extended social interactions. It has been suggested that it evolved as part of a larger adaptation of humans’ species-unique forms of cooperation. Although our closest living relatives, bonobos and chimpanzees, show general cooperative abilities, their communicative interactions seem to lack the cooperative nature of human conversation. Here, we revisited this claim by conducting the first systematic comparison of communicative interactions in mother-infant dyads living in two different communities of bonobos (LuiKotale, DRC; Wamba, DRC) and chimpanzees (Taï South, Côte d’Ivoire; Kanyawara, Uganda) in the wild. Focusing on the communicative function of joint-travel-initiation, we applied parameters of conversation analysis to gestural exchanges between mothers and infants. Results showed that communicative exchanges in both species resemble cooperative turn-taking sequences in human conversation. While bonobos consistently addressed the recipient via gaze before signal initiation and used so-called overlapping responses, chimpanzees engaged in more extended negotiations, involving frequent response waiting and gestural sequences. Our results thus strengthen the hypothesis that interactional intelligence paved the way to the cooperative endeavour of human language and suggest that social matrices highly impact upon communication styles.

Scientific RepoRts | 6:25887 | DOI: 10.1038/srep25887 comparative research with a conversation analysis framework (termed 'CA-assisted comparative research' 14 ), Rossano 15 recently showed that the underlying structure of bonobo gesturing might be more similar to human conversation, and thus language, than previously thought. Conversation analysts have been intrigued with the question of how social actions can be made intelligible, since intelligibility is required to achieve mutual understanding and facilitates the successful engagement of cooperative interactions. One useful tool for addressing this question is outlining the sequential organization of social action via turns at talk 16 . The most fundamental structure in this organization is the adjacency pair, which can be recursively produced and extended in conversation 16,17 . In its minimal, unexpanded form, an adjacency pair has the following features: (a) It is composed of two turns, (b) by different participants, (c) that are adjacently placed, and (d) relatively ordered into first pair parts and second pair parts. First pair parts are actions in first position that initiate some exchange (e.g. a request: "shall we leave now?"), and second pair parts are actions in second position that are responsive to first pair parts (e.g. an answer: "yes, perfect timing") 17 .
There are clear tendencies for a core inventory of speech acts like questions, greetings, requests, etc. (although many actions beyond the core vary), suggesting a strong universal foundation across all cultures 18 . An additional fundamental part of the infrastructure for human conversation is the temporal relationship underlying turn transitions, which is only about 200 ms on average across languages 19 . This is extraordinary if one bears in mind that latencies involved in uttering even a single word are on the order of 600 ms 20 .
In Rossano's study, gestural sequences of two mother-infant dyads of bonobos living in captivity were investigated, with a special focus on participation frameworks (the signaller decides who is part of the interaction by for instance addressing her visual attention toward a recipient, expecting to turn over the communication role), cooperative adjacency pair-like sequences (e.g. request and answer), and temporal relationships underlying gestural performance (e.g. signaller responds immediately after a signal has been produced with a time gap of only > 0 < 0.2 sec) 15 . The results showed that both dyads regularly established and engaged in participation frameworks and cooperative adjacency pair-like sequences and communicated at a speed remarkably similar to the timing of ordinary human conversation (≤ 0.2 sec).
Although cooperative communication between conspecifics has only been studied in bonobos thus far, other studies provide evidence for communicative differences between the two Pan species 10,21 . For instance, Pollick and de Waal 10 found that bonobos in captivity are culturally more diverse in their gesture use and display a higher responsiveness to combinatorial signalling than chimpanzees, giving rise to the speculation that bonobos are a better model for understanding the prerequisites of human communication. Support for this bonobo-chimpanzee dichotomy also stems from other research avenues showing that bonobos are more tolerant and cooperative 11 and outperform chimpanzees in 'theory of mind' tasks that require attention to social causality 22 . Recently, these differences in social cognitive abilities and social make-up have been explained not only in relation to brain regions responsible for aversive emotional stimuli eliciting fear and anxiety 23 , but also neural circuitry that may increase empathic sensitivity and prosocial behaviour 24 . The latter findings are especially important in light of shared intentionality 25 , which refers to collaborative interactions in which participants share psychological states with one another 25,26 . Shared intentionality has been suggested as the driving force and "the small psychological difference" in human cognitive evolution that paved the way for the cooperative endeavor of language 26,27 .
The aim of the present study was twofold: First, we revisited the claim that communicative interactions of our closest living relatives, bonobos and chimpanzees, lack the cooperative nature of human communication 1,2 . Second, we investigated whether bonobos are the better model species for understanding the precursors of human communication 10 .
To target these aims, we tested and expanded some of the parameters used by Rossano 15 in situ (i.e. in bonobos and chimpanzees living in their natural environments) by focusing on the core niche in which communication is learned-face-to-face interactions of mother-infant dyads. Since it recently had been emphasized that due to relatively high degrees of behavioural plasticity both Pan species show considerable inter-site variation (for an overview see 28 ), we investigated communicative interactions of two bonobo and two chimpanzee communities at four different sites and locations. In the case of behavioural differences between communities, this study design then offered the possibility to distinguish between within-versus between-species differences 29 . Data were collected at LuiKotale, Salonga National Park, Democratic Republic of the Congo (DRC), Wamba, Luo Scientific Reserve, DRC, Kanyawara, Kibale National Park, Uganda (Pan troglodytes schweinfurthii) and Taï South, Taï National Park, Côte d'Ivoire (Pan troglodytes verus). We focused on the single communicative function of mother-infant joint travel, since previous studies suggested that this context is a promising candidate for the occurrence of frequent turn-taking sequences to achieve a joint goal (leaving a location) 15,30 . The following criteria were analysed: (i) establishment of participation frameworks (i.e. the initiator establishes for instance via gaze, body direction who is addressed and will be part of the communicative interaction before performing a gestural signal) by focusing on the parameters of gaze, body orientation and initiation distance; (ii) adjacency pair-like sequences (i.e. analysing gestures and their respective responses) by examining the parameters number of gestural requests and their respective responses (so called gesture-response pairs) leading to joint travel and response waiting (i.e. signaller pauses after the signal has been produced for at least two seconds waiting for a response) after each solicitation; and (iii) the temporal relationships between joint-travel-initiating behaviour and response, by focusing on the parameters of delayed (> 2 sec), immediate (≥ 0 < 2 sec) and overlapped responses (< 0 sec, for further details see also methods). Overall, the results of the present study provide evidence that bonobo and chimpanzee mother-infant dyads frequently engage in cooperative turn-taking sequences, and thereby exhibit many similarities to human social action during conversation. The two species differed significantly with regard to some of the examined parameters: While gaze, close initiation distance and fast-paced responses characterised mother-infant joint travel interactions in bonobos, chimpanzees exhibited a higher number of gesture-response pairs, a higher frequency of response waiting and more delayed responses. By using a combination of methods thus far predominantly applied to human social interactions 16,17 , with an unprecedented within-group, between-group and between species comparison of apes living in natural environments, we show that (a) cooperative communication has a more ancient evolutionary origin than previously thought, and (b) social matrices strongly influence communicative preferences and styles. Our study strengthens the view that human communication represents an ensemble of layers of abilities of different types and different antiquity, with precursor adaptations in turn-taking behaviours in which gestures (and possibly other signals) are embedded 31 .

Results
For our analyses, we distinguished gestures fulfilling only one key characteristic of intentional communication 5,30 (single-criteria or SC-gestures) from those that conformed to several key characteristics of intentional communication (multiple-criteria or MC-gestures). These criteria included parameters such as sensitivity to recipients' attentional states, response waiting and goal persistence 5,30 (for further details see methods and video clips 1-4 in the SM).
The  Table S1. Establishment of participation frameworks. We investigated whether mother-infant dyads established participation frameworks before the start of joint travel by analysing the parameters of gaze, body orientation and initiation distance both within and between species. To test the extent to which species and site but also the variables of dyadic role and infant age influenced these three parameters, we used Generalized Linear Mixed Models (GLMM 32 ) for each parameter. Overall, the test predictors had a clear impact in all three models (likelihood ratio tests comparing null and the full model for gaze: χ 2 = 16.8, df = 8, p = 0.03; body orientation: χ 2 = 178.9, df = 8, p < 0.001; distance: χ 2 = 26.1, df = 8, p = 0.001). The behavioural differences that we found provided evidence for species differences, but not for within-species variability (likelihood ratio tests comparing null and the reduced model lacking site effects; gaze: χ 2 = 0.2, df = 4, p = 0.99; body orientation: χ 2 < 0.001, df = 4, p = 1; distance: Concerning gaze, species was the only significant predictor, with bonobos accompanying more initiatory behaviours with gaze than did chimpanzees (estimate ± SE = − 0.485 ± 0.155, χ 2 = 8.288, df = 1, p = 0.004; see Fig. 1). With respect to body orientation, we found a significant interaction between dyadic role and infant age (role*between-infants age: − 0.483 ± 0.128 χ 2 = 16.307, df = 1, p < 0.001). While infants oriented more towards the recipient the older they were, there was no effect for mothers with increasing age of their infants. Additionally, chimpanzees more frequently oriented their bodies toward the recipient before signalling than did bonobos (0.543 ± 0.179, χ 2 = 4.872, df = 1, p = 0.027). Chimpanzee dyads initiated joint travel from a significantly larger distance than did bonobo dyads (0.426 ± 0.103, χ 2 = 7.382, df = 1, p = 0.007; see Fig. 2). Furthermore, initiation distance significantly increased with infant age in both species (between-infants age: 0.145 ± 0.041, χ 2 = 9.919, df = 1, p = 0.002). No other effects in the three models reached significance (see Table 1; sections 1.a-1.c).
Adjacency pair-like sequences. To examine whether mother-infant dyads form successful adjacency pair-like sequences to initiate joint travel, we investigated two parameters: number of gesture-response pairs and response waiting (i.e. signaller pauses at the end of a given signal for at least two seconds waiting for a response; see also methods). To test the influence of species, site, dyadic role and infant age on these three parameters, we used Generalized Linear Mixed Models (GLMM 32 ) for each of them. The test predictors had a clear impact on both models (likelihood ratio tests comparing the null and the full model for gesture-response pairs: χ 2 = 20.9, df = 8, p = 0.007; response waiting: χ 2 = 42.2, df = 8, p < 0.001). The differences we found could only be ascribed to inter-species, rather than intra-species, variability (gesture-response pairs: χ 2 < 0.001, df = 4, p = 1; response waiting: χ 2 < 0.001, df = 4, p = 1).
Regarding the number of gesture-response pairs, a significant interaction between species and between-infants age was found (0.179 ± 0.071, χ 2 = 6.381, df = 1, p = 0.012; see Fig. 3). While in bonobos the number of gesture-response pairs decreased, it increased with age in chimpanzees. In addition, mothers produced significantly more gesture-response pairs than infants (0.18 ± 0.078, χ 2 = 5.447, df = 1, p = 0.02). For the parameter of response waiting, we found a significant interaction between dyadic role and between-infants age (− 0.433 ± 0.114, χ 2 = 14.856, df = 1, p < 0.001). While infants were more likely to wait for a response with increasing age, the age of infants did not influence the occurrence of response waiting in mothers. In addition, chimpanzees were more likely to wait for a response by the recipient than bonobos (0.893 ± 0.173, χ 2 = 19.845, df = 1, p < 0.001; see Temporal relationships between signal and response. We examined whether there were interor intra-species differences concerning the timing of a given response after a travel-initiating gesture (i.e. SC-gestures and MC-gestures) had been produced. We differentiated between immediate (Δt [start response -end initiatory gesture] > 0 < 2 s), delayed (start response ≥ 2 s after end initiatory gesture) and overlapping responses (Δt [start response -end initiatory gesture] < 0; for detailed definitions see methods). To test the extent to which species, site, dyadic role and infant age influenced the three response types, we used Generalized Linear Mixed Models (GLMM 32 ) for each parameter. The test predictors had a clear impact on the occurrence of overlapping and delayed responses, but not on the occurrence of immediate responses (overlapping response: χ 2 = 29.6, df = 8, p = 0.005; immediate response: χ 2 = 5.6, df = 8, p = 0.694; delayed response: χ 2 = 35.9, df = 8, p < 0.001). The likelihood ratio test comparing the full models with the reduced models revealed that behavioural differences mirrored species differences but not within-species variability (overlapping: χ 2 = 0, df = 4, p = 1; delayed: χ 2 < 0.001, df = 4, p = 1; see also Supplementary Material, Figure S1).
Overlapping responses were significantly more frequent in bonobos (− 0.65 ± 0.127, χ 2 = 11.656, df = 1, p < 0.001; see Fig. 5) than in chimpanzees. Furthermore, infants of both species were more likely to use overlapping responses than were mothers (0.382 ± 0.135, χ 2 = 8.256, df = 1, p = 0.004). Overlapping responses were   Table 1. Effects of species, role and infant age on the investigated parameters. (1) Not shown as lacking a meaningful interpretation. Infant sex and parity were included as control predictors; ID and site were included as random effects.  significantly more frequent in dyads with younger infants than in dyads with older infants (− 0.124 ± 0.063,  interaction between species and between-infants age, with chimpanzees producing more delayed responses across infant ages than did bonobos (− 0.51 ± 0.187, χ 2 = 7.564, df = 1, p = 0.006; see Fig. 6). Mothers of both species were also more likely to produce delayed responses than their infants (− 0.484 ± 0.137, χ 2 = 12.416, df = 1, p < 0.001). No other effect in the models reached significance (see Table 1, sections 3.a-3.b).

Discussion
The aim of the present study was twofold: First, we wanted to revisit the claim that human cooperative communication evolved as part of a larger, uniquely human, adaptation for cooperation and cultural life in general 1,2 . Second, we examined whether bonobos are the better model species for understanding the prerequisites of human communication 10 . To do so, we investigated whether bonobos and chimpanzees, both of which engage in general cooperative activities [11][12][13] , use distinct features characteristic of human social action in conversation 16,17 . By taking into consideration intra-and inter-species variability and by focusing on the mother-infant dyad, our results showed that all observed dyads across groups frequently engaged in turn-taking sequences to negotiate joint travel. They established participation frameworks via gaze, body orientation and the adjustment of initiation distance, and they used adjacency pair-like sequences characterized by gesture-response pairs and response waiting. Regarding temporal relationships between signals and responses, we found that mother-infant dyads of both species used the whole spectrum of responses, including immediate, overlapping and even delayed responses. Immediate responses match the temporal relations between turns in human speech consisting of relatively little cultural variation (e.g. overall cross-linguistic median of 100 ms, ranging from 0 ms in the English and Japanese culture, for instance, to 300 ms in the Danish and Lao culture) 19 . Our findings therefore support and expand the results of Rossano 15 , by demonstrating that gestural exchanges observed in mother-infant dyads of bonobos and chimpanzees are often very similar in timing to human action in conversation and embody the most crucial features of human cooperative conversation 4,27 : These gestural exchanges are bidirectional coordination devices, comprising the two implicit roles of signaller and recipient. In learning to use these gestures, individuals learn to play and to comprehend both roles no matter which role they are performing (soliciting to leave a location or being solicited to leave a location). Although we recently showed that chimpanzee mothers and their infants differ in many of the gesture types employed to initiate joint travel 30 , gestures shared by mothers and infants might involve role-reversal imitation (i.e. when one uses a gesture toward others the way others have used this gesture toward oneself), but also take the other's perspectives on the event of joint travel. Furthermore, to reach the joint goal (of leaving a location), signallers made efforts to communicate in ways that were comprehensible to the recipient, for instance by combining initiatory behaviours with gaze and orienting the body to recipients. In addition, they seemed to 'clarify' the intended goal by using several adjacency pair-like sequences composed of the same or different gestures, when the first communicative attempt had not been successful. Turn-taking sequences of pre-linguistic human children go a step further in that recipients ask for clarification when needed and employ 'negotiation of meaning' 33 . Furthermore, the rapid turn-taking in human conversation involves indefinite varying contents of turns, multi-modal deployment of vocal and gestural signals, and also seems without parallel, given the sheer amount of time and effort invested in communication 31 (but see for nonhuman primates and birds 34,35 ). Overall, our findings strengthen a recent proposal by Levinson and Holler 31 emphasizing the role of turn-taking behaviour for evolutionary scenarios of human language. They suggest that human language, despite its tight integration of speech and gesture, is a system composed of layers of abilities of different types and different antiquity. Thus, unpeeling the layers should enable us to understand the evolution of human language from an original rapid exchange of gestural or vocal material, into a system where the complexity of the linguistic and gestural material that is expressed in relatively short bursts has grown to the very limits that human cognition can process 31 .
In sum, sequentially organized, cooperative social interactions are not simply by-products of individuals living in human enculturated environments 15 , but play a crucial role in communicative exchanges of mother-infant dyads of bonobos and chimpanzees living under active selection pressures. These results challenge the human-ape divide, which suggests that human cooperative communication evolved as part of a larger adaptation of humans' species-unique forms of cooperation 2 ratcheted via existing and simpler components of primate cognition, such as group action and manipulative communication 26 . Our findings indicate that cooperative communicative interactions seem to play a crucial role in mother-infant dyads of bonobos and chimpanzees and, more generally, in nonhuman animals, for which shared goals and relatively low levels of competition prevail. Similarly to the universally organized social-interaction matrix of human conversation 19 , the results suggest that our closest living relatives have a strong universal infrastructure underlying their gestural interactions, which serves to minimize gaps and overlaps and allows for efficient information exchange. Further research on the methods and model species commonly used to draw inferences about evolutionary precursors to human communication is warranted (see for recent developments in other areas of cognitive ethology 36 ), to enable (i) higher sensitivity to the social characteristics and/or ecology of a given species and (ii) a vital understanding of the structure and cognitive complexity underlying turn-taking sequences 14 and communicative exchanges such as vocal alternations 34 and duetting 37 .
Concerning our second aim of examining whether bonobos are the better models for precursors to human communication, a comparison of the investigated parameters showed behavioural differences between species, but not within species. Specifically, bonobo dyads (i) accompanied their signals more frequently with gaze, (ii) stayed in closer spatial proximity to each other for mother-infant coordination, and (iii) preferred to use overlapping and immediate responses. In contrast, chimpanzee dyads (i) were more likely to orient their bodies toward a recipient before signalling, (ii) showed a higher number of gesture-response pairs and response waiting, (iii) displayed overall more 'communicative persistence' to obtain the desired goal of joint travel, and (iv) used all three response tempi with relatively similar frequencies.
Three hypotheses may account for these observations. First, differences in communicative patterns may be explained by differences in ecological environments. For example, habitat characteristics such as thickness and growth of terrestrial herbaceous vegetation (THV) may differ considerately between the sites, resulting in different degrees of visibility and thus communication space and eye contact. Although THV is clearly more prevalent in bonobo habitats than in chimpanzee habitats 38 , differences in THV might also exist between the two chimpanzee habitats, resulting in relatively higher levels of visibility at Kanyawara 38 compared to Taï South 39 . If this hypothesis were true, we would have expected to find differences in communicative behaviours between the Kanyawara and the Taï South community and between bonobos and chimpanzees in general. This does not accord with our observations. Second, differences in communication styles between bonobos and chimpanzees are a by-product of the studied age range, particularly because chimpanzee infants may generally develop more quickly than do bonobo infants. For instance, Kuroda 40 suggested that growth rates of bonobos and chimpanzees differ considerably, such that bonobos undergo a slower development of (i) spatial independence, (ii) locomotor skills (e.g. climbing, walking quadrupedally, riding on mother's back), and (iii) social interactions with conspecifics (e.g. approaching, playing). This proposed delay in general development in bonobos may also have a crucial impact on the speed of communicatory skill development. If this hypothesis were true, we would have expected to find that age had a significant impact on the investigated parameters, with bonobo infants showing certain parameters such as use of gaze, adjustment of body orientation and response waiting, as well as overlapping responses, significantly later than chimpanzee infants. However, this was not the case. Body orientation and initiation distance were the only parameters for which a developmental effect was found, with increases of adjustment of body orientation towards mothers and initiation distance with age in both bonobo and chimpanzee infants. Overall, chimpanzee dyads initiated joint travel from larger distances than did bonobo dyads. These results are in line with findings of de Lathouwers and colleagues 41 , who showed that immature chimpanzees spend more time at larger distances from their mothers than do immature bonobos. Our study confirms that chimpanzees indeed develop spatial independence more quickly than do bonobos.
Third, bonobos and chimpanzees might employ different communication styles. Consistent with this hypothesis and based on our investigated parameters, bonobos and chimpanzees could be characterized by two clearly distinguishable communication styles accompanied by different temporal relationships: Bonobos frequently combined their communicative signals with gaze while in close proximity to the addressee, and they often used speedy responses. The underlying temporal relationships often matched those underlying human turn transition during speech 19 , with predominantly single adjacency pairs but also recipients responding before signals had been fully articulated. Chimpanzees, on the other hand, adjusted their body orientation toward recipients and used a generally slower mode of communication that involved more gesture-response pairs, higher frequencies of response waiting and delayed responses. Bonobo communication thus seems to resemble a subtle dance coined of flowing movements by signallers and recipients, while chimpanzee communication is structured with temporally separated and clearly recognizable units such as signal, pause and response. Chimpanzee signalling mirrors the structure of other social interactions, such as aggressive and grooming interactions, which are also characterized by typical negotiation sequences 42,43 , thereby demonstrating the significance of clearly structured interactions in chimpanzee society.
While future studies with additional age classes, communicative functions and dyads are of course mandatory, our study provides the first evidence that mother-infant dyads of bonobos and chimpanzees living in their natural environments employ different communication styles to convey the same message. Moreover, if certain Scientific RepoRts | 6:25887 | DOI: 10.1038/srep25887 communicative patterns are already observable in mother-infant coordination, the first step of co-regulated social interaction 44 , it is likely that these patterns are also crucial for general communication abilities of the species. Thus, generalization to behaviours of other dyads of a given community may be possible to some extent.
Although the long-standing bonobo-chimpanzee dichotomy has been challenged by new data emphasizing intra-species over inter-species variability 28 , bonobos and chimpanzees still seem to differ considerably concerning distinct characteristics of their social matrices. Males are more influential in chimpanzee society than in bonobo society 45,46 , with male chimpanzees heavily competing within their communities to gain indirect and direct fitness benefits 13,46 . This competition results in linear dominance hierarchies, male harassment and male-female dominance, but also strong social bonds and cooperative behaviour between males in the form of short-and long-term alliances (e.g. in the form of coalitionary behaviour, grooming, meat sharing and border patrols 13,46 ). High levels of aggression, including lethal attacks, characterize intergroup encounters in chimpanzees, and infanticide has been observed within and between communities 47 . In contrast, bonobo society is characterized by co-dominance between the sexes, prolonged mother-son relationships 45 , and strong bonds between unrelated females 45,48 , resulting in a more flexible choice of coalition partners. Although between-group encounters in bonobos are usually friendly and peaceful 49 , there is anecdotal evidence of attempts of infanticide by males 50 but also females 51 . Given the gregariousness of bonobo females, the threat of female infanticide could explain the need for close range communication between mothers and their dependant offspring.
It has been argued that species-specific social matrices and behaviours have been evolutionarily shaped by the distinctive morphology, connectivity and molecular biology of brain regions and pathways involved in social and environmental appraisal of threats and vigilance and control of emotional responses 23,24 . For instance, bonobos have more gray matter in the dorsal amygdala and a larger pathway linking the amygdala with the ventromedial prefrontal cortex (VMPFC) 24 . This neural circuitry has been implicated in both top-down control of aggressive impulses and bottom-up biases against harming others, as well as increased empathic sensitivity and prosocial behaviour 24 . In addition, bonobos have approximately twice the density of serotonergic axons in the amygdala compared to chimpanzees 51 , contributing to appraisal of the emotional context and significance of the environment 52 . These differences in neural circuitry are in line with recent experimental findings showing that bonobos tend to exhibit more cautious temperaments 53 , reduced 'emotional reactivity' 11 , and greater tolerance when competing over food-resources 36 . The results of our study suggest that crucial features characterizing human communication, such as gaze and anticipation of recipients' behaviour 2,5 , may be more significant in bonobo than in chimpanzee communication. Bonobos appear to exhibit a higher social awareness of the communicative situation and the anticipated meaning of a given signal, strengthening recent results demonstrating a bonobo-chimpanzee divergence in tasks requiring attention to social causality 22 .
Bonobos may therefore represent the most representative model for understanding the prerequisites of human communication 10 . However, additional analyses of the communicative and cognitive abilities of our closest living relatives are compulsory for a complete understanding of the impact of social and possibly cultural matrices on communication styles and tendencies. In addition, examples of convergent evolution in distantly related species can provide clues to the types of problems that particular communicative mechanisms are 'designed' to solve 54,55 . We thus hope to inspire future research that not only incorporates additional dyads and contexts, but also conducts taxonomically informed comparisons of species engaging in turn-taking behaviour during general interactions and communicative exchanges.
In sum, our results provide substantial evidence that the two primary model species for the origins of human behaviour, bonobos and chimpanzees, differ in their communication styles. While bonobos seem to anticipate and respond to signals before they have been fully articulated, chimpanzees engage in more time-consuming communicative negotiations. Both species use sequentially organized, cooperative social interactions to engage in a joint enterprise: Leaving together to another location. Their communicative interactions thus show the hallmarks of human social action during conversation and suggest that cooperative communication arose as a way of coordinating collaborative activities more efficiently. Our results strengthen a recent proposal by Levinson and Holler 31 suggesting that the apparent gulf between animal and human communication may be bridged by looking for precursors adaptations to human language in turn-taking interactions.  56 , 1973 in Wamba 57 , 1987 in Kanyawara 58 and 1979 in Taï South 39 . It was therefore possible to observe the community members during dawn-till-dusk follows and to collect high-quality video and audio footage. In addition, we had access to long-term data on demography and relatedness for all four field sites. We observed communicative interactions between mothers and their youngest dependent offspring in a total of 12 bonobo dyads and 13 chimpanzee dyads: Six dyads were observed at LuiKotale, six at Wamba, seven at Scientific RepoRts | 6:25887 | DOI: 10.1038/srep25887

Methods
Kanyawara and six at Taï South. The age of the offspring ranged from ten to 56 months in bonobos and nine to 69 months in chimpanzees (see Table 2 for detailed information on subjects and data sets).

Data collection.
We used a focal behaviour sampling approach 59 , while maintaining a record of the frequency with which a particular dyad had been observed. In situations where we could choose which of several dyads to film, we targeted those individuals previously sampled least often. All social interactions involving mothers and infants (i.e. mother-infant as well as mother-conspecific and infant-conspecific interactions) that were judged to have any potential for communicative interactions were recorded using a digital High-Definition camera (Canon Legria HF M41) with an externally attached unidirectional microphone (Sennheiser K6). During approximately a total of 2200 hours of observation (1033 hours for bonobos; 1189 hours for chimpanzees), we collected a total of 238.5 hours of video footage on the communicative behaviour of 12 bonobo (Wamba: 41.5 h; LuiKotale: 51.9 h) and 13 chimpanzee (Taï South: 73.4 h; Kanyawara: 95.5 h) mother-infant dyads (mean ± SD per dyad = 9.5 ± 3.5 h; see Table 2 for further details on data collection). However, since this paper focuses only on the communicative context of joint travel, our analysis is based on a total of 319 bonobo and 410 chimpanzee video recordings of mother-infant joint travel interactions (mean recordings per bonobo/chimpanzee dyad: 26.6/31.9; see Supplementary Material, Table S1). In addition, we included five joint-travel interactions that were recorded with a Pocket PC (HP iPAQ rx1959), resulting in a total of 415 interactions for chimpanzees.

Coding of behaviours.
To establish the repertoires employed to initiate joint travel and to enable subsequent analyses, a total of 729 high-quality video files of mother-offspring joint-travel-initiations (i.e. carries with clear visibility of joint-travel-initiating behaviours) were coded using the program Adobe Premiere Pro CS4 (version 4.2.1). Behavioural definitions were based on established ethograms of bonobo 57 and chimpanzee behaviour 60 . The coding scheme was designed by using parameters developed in previous work on great ape gesturing 30,61 . We only coded successful, agent-initiated joint travel interactions, meaning those interactions leading to infants leaving a location attached to their mothers in a dorsal or ventral carry position. We differentiated between joint-travel-initiations via intentionally produced gestures showing more than one key characteristic of intentional communication and gestures showing only one key characteristic of intentionality, which are respectively called multiple-criteria gestures (MC-gestures) and single-criteria gestures (SC-gestures). Gestures were defined as directed, mechanically ineffective movements of the body or body postures that elicited ("requested") a voluntary response by the recipient 62 .
The following key characteristics of intentional communication were measured 5,30 : Sensitivity to the attentional state of the recipient. The signaller shows signs of being aware of the recipient's state of attention, e.g. by using visual gestures only when the recipient is looking.
Response waiting. The signaller pauses at the end of the production of a signal and waits for at least two seconds expecting a response, while maintaining visual contact with the recipient.
Apparent satisfaction of signaller. The signaller's communication ceases when the apparent goal has been met by the recipient (leaving the location together).
Goal persistence. The signaller elaborates her signalling when thwarted, e.g. by repeating and exaggerating the signal or by using a different means of communication.
Establishment of participation frameworks. We examined three different parameters: Gaze. Signaller looks at the recipient while executing the gesture.
Body orientation. Recipient is positioned directly in front of and in the visual field of the signaller.
Initiation distance. Physical distance (in arm lengths) between signaller and recipient at first joint-travel-initiating gesture.
Adjacency pair-like sequences. To investigate whether bonobos and chimpanzees use adjacency pair-like sequences, we focused on gesture-response pairs and response waiting (see paragraph above for definition) after each solicitation gesture. Regarding gesture-response pairs, we analysed the gestures and gestural bouts that included single gestures or sequences, separated by periods of response waiting lasting more than a second and followed by a recipient's response.
Temporal relationships between turns. We assessed the temporal relationships between signals and their respective responses by differentiating between three types of responses (although see for exact times Supplementary Material, Figure S1): Immediate response. The joint-travel-initiating behaviour is followed by a response of the recipient (Δ t [start response -end of initiatory gesture] > 0) less than two seconds after the behaviour has been articulated.
Delayed response. The joint-travel-initiating behaviour is followed by a response of the recipient more than two seconds after the behaviour has been articulated (end of initiatory gesture).
Scientific RepoRts | 6:25887 | DOI: 10.1038/srep25887 Overlapping response. The joint-travel-initiating behaviour (SC-or MC-gesture) is followed by a response of the recipient either before the behaviour has been fully executed (Δ t [start response -end initiatory gesture] < 0), or less than one second after it has been fully articulated. The full articulation was deciphered by the observer when a given gesture was followed by the immediate reaction (on behalf of the recipient) or response waiting (on behalf of the signaller). For each gesture used to solicit joint travel, we coded the interaction role of the actor (2 levels: mother, infant), infant age (range = 9-69 months), infant sex (2 levels: female, male), and mother's parity (range = 1-5 offspring). Fifteen per cent of all mother-infant interactions were coded for accuracy by a second observer and tested using the Cohen's Kappa coefficient to ensure inter-observer reliability 59 with the following results: A 'very good' level of agreement for initiatory gesture type/gaze (κ = 0.80), joint-travel-initiator (κ = 0.84), orientation and distance to recipient (κ = 0.89), gesture-response pairs (κ = 0.89), and temporal relationships including response waiting (κ = 0.85).
Analyses. To test the extent to which species and/or site, but also other parameters such as dyadic role and infant age, influenced i) the establishment of participation frameworks before signalling (response variables: gaze, body orientation, initiation distance), ii) successful adjacency pair-like sequences (response variables: number of gesture-response pairs, response waiting), and iii) the timing of signal and response (response variables: immediate response, delayed response and overlapping response), we used Generalized Linear Mixed Models (GLMM 32 ) with a binomial error structure and logit link function. For the number of gesture-response pairs and distance, we used a Poisson error structure and log link function. We included species, dyadic role and infant age as our key test predictors. Since age varied considerably between infants, we used the method of within-subject centering 63 to determine whether the effect of infant age was particularly relevant within and/or between infants. Hence, we included in the model the average age of each infant (constant across all data points of the respective mother-infant pair; 'between-infants age') and the difference between the infant's actual age and its average age ('within-infants age'). Because we predicted differences between the species and assumed that infants would take a more active role throughout ontogeny, we also included four two-way interactions between both species and role with the two variables representing infant age in the model. To control for the effect, we also included infant's sex and mother's parity as fixed effects in the model. We included study site and identity of the mother and infant as random effects (intercepts). To keep type 1 error rates at the nominal level of 5%, we also included role, within-infants age, parity and infant sex within subject identity, and site 64 as random slopes components. We did not include any other random slopes components within mother ID because, with a single exception, each mother was recorded with only a single infant, so the random slopes of these fixed effects within mother ID would be highly redundant with those within infant identity. For the other fixed effects, we did not include random slopes since they were usually constant within mother and infant ID. To keep model complexity at an acceptable level, and because neglected random slopes do not compromise type 1 error rates 64 , we did not include correlations between random slopes and random intercepts.
The models were implemented in R 65 using the function glmer in the package lme4 66 . To test the overall significance of our key test predictors, we used a likelihood ratio test 67 to compare the full models with a null model that contained only the control predictor with fixed effects and all random effects. If either interaction with withinand between-infants age was non-significant, they were removed from the model. To test whether inter-site differences had a significant effect on the response variables, we excluded site (and all random slopes within site), and ran a second likelihood ratio test comparing the full model with this reduced model. Prior to running the models, we z-transformed between-infants age, within-infants age and parity 68 . To control for collinearity, we determined Variance Inflation Factors (VIF 69 ) from a model that included only the fixed main effects, using the function vif of the R package car. This revealed that collinearity was not an issue (maximum VIF = 1.23). In the models with Poisson error structure, overdispersion was not an issue (dispersion parameters for distance: 1.03, gesture-response pairs: 0.52). To estimate model stability, we excluded the levels of random effects one at a time, ran the models again and compared the resulting estimates derived with those obtained from the respective models based on all data. This revealed that all models were at least 'moderately' stable, particularly for those estimates that were not close to zero. 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 R-version R.3.1.1 65 , with the level of significance set to 0.05.