A convergent interaction engine: vocal communication among marmoset monkeys

Abstract

To understand the primate origins of the human interaction engine, it is worthwhile to focus not only on great apes but also on callitrichid monkeys (marmosets and tamarins). Like humans, but unlike great apes, callitrichids are cooperative breeders, and thus habitually engage in coordinated joint actions, for instance when an infant is handed over from one group member to another. We first explore the hypothesis that these habitual cooperative interactions, the marmoset interactional ethology, are supported by the same key elements as found in the human interaction engine: mutual gaze (during joint action), turn-taking, volubility, as well as group-wide prosociality and trust. Marmosets show clear evidence of these features. We next examine the prediction that, if such an interaction engine can indeed give rise to more flexible communication, callitrichids may also possess elaborate communicative skills. A review of marmoset vocal communication confirms unusual abilities in these small primates: high volubility and large vocal repertoires, vocal learning and babbling in immatures, and voluntary usage and control. We end by discussing how the adoption of cooperative breeding during human evolution may have catalysed language evolution by adding these convergent consequences to the great ape-like cognitive system of our hominin ancestors.

This article is part of the theme issue ‘Revisiting the human ‘interaction engine’: comparative approaches to social action coordination’.

1. Introduction

To better understand the primate origins of human nature, a focus on our closest relatives, the great apes, is important because it reveals how we are similar to them. Such similarities can readily be explained by shared ancestry. However, the comparison simultaneously reveals how we are different from them, and the origin of these dissimilarities is more difficult to explain. Evolutionary biology can often help us provide an explanation by looking at more distantly related organisms, primates or non-primates, that are more similar to us than apes with regard to some features owing to convergent evolution.

Like humans, but unlike any other great ape, callitrichid monkeys (marmosets and tamarins) are cooperative breeders: mothers receive help in rearing their offspring from other group members, and this help increases infant growth and survival [1–3]. In callitrichids, group members other than the mother (i.e. allomothers, which includes sires, related and non-related adults and subadults [4]) help by carrying, provisioning and protecting immatures. Collective action and coordination are also evident in vigilance [5] and anti-predator behaviours, as well as resource and territorial defence [6,7]. These cooperative behaviours, in particular joint infant care, require regular social interactions between individuals, such as handing over infants from one carrier to the next or sharing food [8,9]. They also require coordinating mutually exclusive activities among group members, such as infant carrying versus anti-predator behaviour [10], or feeding versus being vigilant [5]. To allow for the smooth coordination of these activities, callitrichids can rely on a set of well-developed socio-cognitive skills and motivations [11,12]. Accordingly, they appear to have convergently evolved a callitrichid interaction engine. Understanding its consequences for communicative flexibility can thus contribute to an evolutionarily informed understanding of how our interactional ethology may have paved the way to the evolution of language and social cognition more generally [1,10,13–17].

This contribution has three main sections (§§2–5). In §2, we explore the idea that the cooperative interactional ethology of marmosets is supported by key elements it shares with the human interaction engine [18,19]: mutual gaze, turn-taking, prosociality and a group-wide platform of trust [20], as well as a general tendency to be in tune with others at the hormonal, communicative and behavioural level. We find that these elements are highly prevalent in marmosets and other callitrichids. To the extent that the interaction engine indeed gives rise to more flexible communication, this pattern predicts that marmosets and other callitrichids also have more elaborate communicative skills than other primates. In §3, we therefore review the vast literature on marmoset vocal communication, and find it is indeed remarkable among non-human primates, specifically regarding high volubility, large signal repertoires, some vocal learning as immatures and adults, a babbling phase during ontogeny and intentional control over vocalizations. In §4, we scrutinize in detail how the affordances of the cooperative breeding system can bring about the specific communicative skills observed in marmosets, by separately looking at the affordances of this breeding system for adults and for immatures. We end with a discussion of the implications for human language evolution.

2. A convergent interaction engine?

The communication system of humans, language, is unusual among animals, and so is human cooperation and joint action, i.e. our interactional ethology. A prominent proposal, the interaction engine hypothesis [18,21,22], posits that our interactional ethology is supported by a set of cognitive and motivational interactive abilities, summarized as the human interaction engine, which catalysed both the evolution of language and its acquisition during ontogeny. A comprehensive list of the specific abilities and traits that compose the human interaction engine is currently lacking, but it entails broadly the toleration and expectation of mutual gaze, the white sclera of the human eye, rapid turn-taking, and ‘the sheer amount of time and effort invested in communication’ [19] , as well as the ability to recognize and establish common ground, fine-tuned via repair behaviours and sometimes based on Theory of Mind and even Gricean intentions [18,21,22]. These abilities thus arguably predated the emergence of language, rather than being the result of it.

In this section, we show how the habitual cooperative interactions of callitrichid monkeys likewise require their own, but overlapping, set of capacities and motivations: the marmoset interaction engine (figure 1). Doing so will reveal to what extent habitual cooperative interactions indeed coevolve with a set of skills that feature prominently in the human interaction engine, and—given that callitrichids do not possess human language—identify skills for which we can be sure they do not require the presence of language.

Figure 1.

The marmoset interaction engine. The marmoset interactional ethology is characterized by cooperative breeding, cooperation in various contexts and high levels of interdependence. To make this system work, adults and immatures face specific affordances, or behavioural tasks. To fulfil these tasks and affordances, a set of psychological and motivational adaptations must exist (see also [11,12]). These can be summarized as a marmoset interaction engine and include several of the elements that have been highlighted [19] as critical for the human interaction engine. They may thus scaffold more flexible communication not only in humans but also in marmosets. (Online version in colour.)

(a) . Mutual gaze to coordinate joint actions

In general, non-human primates avoid making eye contact, and direct gaze is typically perceived as a threat [23]. A reduction of gaze aversion is thus a prerequisite for mutual gaze and its use to coordinate actions, as highlighted in the interaction engine framework (e.g. [24]). The extent of gaze aversion varies across primate species and contexts. For instance, bonobos make more eye contact than chimpanzees [25] and gaze is used as a social tool to start and end social interactions by several primates, in particular when recruiting others for social play [26–28] and grooming [29] or during mating [30] (see [27] for overview of other species).

Intriguingly, marmosets and other callitrichids show a striking lack of gaze aversion in almost all contexts. Captive marmosets, when looked at by people, will look straight back into their eyes, perhaps responding with a trill call (i.e. short-range contact calls in marmosets). Most other primates, such as macaques, will avert their gaze, perhaps glance back quickly, or even respond with a direct threat [31]. A systematic comparison found that marmosets showed much lower levels of gaze aversion than their closest living, but independently breeding, sister taxon, the squirrel monkey [32], and more recent eye tracking results revealed that marmosets showed longer inspection times and a stronger preference for faces compared with an independent breeder, the rhesus macaque [33]. Two recent comparative analyses covering many different primate species confirmed the generality of this reduced gaze aversion in marmosets and other callitrichids. Mearing et al. [34] found that the whiteness of the sclera had coevolved with allomaternal care across primates. Likewise, Cerrito & DeCasien [35] found that among primates, neural control of facial muscles had coevolved with the frequency of allomaternal care. Both a bright white sclera and neural control over facial muscles facilitate cooperative social interactions and signalling when individuals are tolerant of direct eye contact in a broad range of contexts, including foraging. That marmosets indeed use eye-direction independently of head orientation for close-range coordination was found in an experimental test investigating their ability to use gaze as a communicative cue to find food. They were able to use human pupil direction alone from a distance of 1 m to disambiguate which of nine containers (that were only 25 cm apart from each other) would contain food [36]. They were thus very skilful in using even very subtle eye cues to precisely extrapolate gaze direction.

Mutual gaze is even more directly relevant to active coordination of cooperative behaviours among conspecifics. A recent set of results on co-representation during joint action using the joint Simon task revealed that marmosets engage in mutual gaze when in a coordination conflict, but independently breeding capuchin monkeys and Tonkean macaques do not (29,30,31). In this task, two individuals can work on an apparatus to obtain food for both of them, but this is only successful if only the partner whose turn it is actually works on the apparatus. The marmosets would co-represent each other when jointly engaged in this cooperative coordination task [37], as did those selected dyads of capuchin monkeys and Tonkean macaques that were sufficiently tolerant to engage in the task at all [38,39]. However, intriguing differences appeared when analysing gazing behaviour: marmosets, like human children, engaged in mutual gaze prior to acting in ambiguous trials when it was unclear whose turn it was, whereas in non-ambiguous trials mutual gaze was less frequent. The independently breeding capuchin monkeys and Tonkean macaques hardly engaged in mutual gaze at all during the entire study [38]. These interactions not only highlight the cooperative use of mutual gaze in marmosets but also suggest some proclivity towards turn-taking, to which we will now turn.

(b) . Cooperative turn-taking

Callitrichids are renowned for their extensive turn-taking during antiphonal calling, which has been studied in much detail in marmosets. When separated from each other, these monkeys call back and forth using phee calls (i.e. long-distance contact calls), which facilitates the monitoring of each other's position, as well as group cohesion and reunion [10]. These vocal exchanges are intriguing because they are not merely triggered automatically as for instance in chorusing insects or frogs [40], but show several features of cooperative vocal exchanges [41].

During such extended sequences of vocal turn-taking, marmoset dyads manifest the dynamics characteristic of coupled oscillators, similar to what has been proposed for human conversational turn-taking [41]. Moreover, when a partner is ‘far away’ rather than ‘nearby’, individuals increase the volume, thus increasing the likelihood that the conspecific will hear their call [42]. Turn-taking conversations can also include more than two individuals. In trios of captive marmosets, individuals called more frequently than expected by chance when it was their turn, and the probability of calling twice before the others in the trio had replied was lower than chance [43]. Additional behavioural evidence shows that when multiple individuals are involved in antiphonal conversations, marmosets acutely attend to the calls of each individual separately [44]. In a recent experiment, Jovanovic & Miller [45] exposed marmosets to five playbacks of marmosets emitting phee calls simultaneously. The temporal dynamic of the calls of two pairs of these virtual monkeys suggested engagement in dyadic turn-taking conversations, whereas the fifth was vocally interacting with the focal animal. The focal animal apparently picked up these dynamics and contingently responded to the fifth monkey, thereby effectively solving this cocktail party problem. Turn-taking in independently breeding primate species appears less sophisticated and cooperative, if it occurs at all, in that it is less prevalent and sequences are shorter and thus cannot show this cooperative entrainment [41] or do not feature egalitarian role-reversibility (reviewed in [20]).

Egalitarian reversibility of the speaker/hearer roles has been highlighted as essential for linguistic turn-taking and can occur in situations without rigid ascription of interactional roles. This contrasts for instance with situations of vocal exchanges between dominants and subordinates where the roles are clearly predefined by dominance status, and also events of gestural communication in apes when carrying is solicited by immatures in bonobos and correspondingly answered by mothers [46], or invitations to leave tend to be initiated by the mother [47]. To what extent this apparent lack of egalitarian role reversibility in great ape gestural communication is a general feature of their way of communicating with each other, or whether it rather reflects a lack of studies in contexts that are more likely to elicit it, remains to be established in future research.

The cocktail party problem studies suggest marmosets are highly attentive to the conversations of their conspecifics, which is consistent with a recent thermography study [48]. Marmosets were allowed to listen to playbacks of vocal interactions between conspecifics, representing either a cooperative or agonistic vocal interaction. Subtle changes in emotional arousal, measured with thermography, revealed that the marmosets perceived these vocal interactions between conspecifics holistically as conversations, rather than as a simple concatenation of independent calls. Moreover, after witnessing such conversations, marmosets showed a stronger motivation to get close to a stranger that was involved in a cooperative than an agonistic vocal interaction with a third individual, consistent with third-party social evaluation [48]. Whereas sophisticated perception of social vocal interactions, in particular in competitive contexts, is likely widespread among primates [49], such a positive social evaluation of cooperative third-party conspecific interactions is remarkable.

Turn-taking was recently described in a different context as well, namely vigilance. Pairs of marmosets were provided with mashed food in opaque food bowls, such that a feeding marmoset could not see its surroundings. Not only did their partners increase vigilance when the mate was in this vulnerable position, but the pairs also engaged in turn-taking between feeding and vigilance [5] (see electronic supplementary material, video S1).

Overall, marmosets thus frequently engage in turn-taking. These interactions represent cooperative vocal exchanges rather than stereotypically triggered answers, can be reasonably fast paced and, in contrast to great apes, also satisfy the criterion of egalitarian role-reversibility [20]. Moreover, turn-taking develops during ontogeny, partially scaffolded by contingent responses of the adult group members (see below), and also emerges in domains other than antiphonal calling. Turn-taking is thus an important part of the marmoset interactional ethology.

(c) . Prosociality and a group-wide platform of trust

Human language is a form of low-cost honest signalling, which is notoriously difficult to evolve [50] because it requires some platform of trust, i.e. ‘a social niche in which cheap but honest communication with non-kin is possible, because messages tend to be trusted as a default’ [20] .

Callitrichid monkeys show high degrees of proactive prosociality among group members, reflecting a psychological mechanism that generates spontaneous, unsolicited prosocial behaviour [51]. This prosociality has been extensively studied in the foraging context. Here, proactive prosociality takes the form of offering food to others without the recipients begging for it, which can be readily observed in the wild [8] but also in captive experiments (e.g. [52,53]). Most importantly, a direct comparison of a large number of primate species revealed that callitrichids are the most likely to provide food to their group members in a group service paradigm [54]. This readiness to share, which is a proactively prosocial disposition, applies not only to food but also to information, as elaborated in detail elsewhere [13]. In a nutshell, compared with independently breeding primates, including great apes, callitrichids are much more likely to vocally advertise information that is beneficial to the partners rather than to themselves, and have a much stronger inclination towards teaching.

Proactive prosociality can have broader consequences for communicative complexity, provided at least two conditions are fulfilled. First, vocalizations should be under some intentional control rather than automatically triggered and should not be restricted to close kin or specific partners only. To scrutinize the intentionality condition, we built on the vast literature of primate intentional communication [55] and applied the criteria developed there to marmoset prosocial food provisioning [56]. We found that marmoset prosociality is intentional, because (i) donors can use multiple means to reach their prosocial goal (offer food to the recipient), and adjust them to specific conditions, (ii) it is audience-sensitive, e.g. helpers are particularly generous if no-one else is around to take care of immatures [57] and (iii) it is goal-directed, in that it continues exactly until the putative goal is reached and individuals check back visually whenever their actions do not lead to the transfer of food. Finally, they also take the need for help into account: food is offered readily and spontaneously to young immatures incapable of finding food themselves, [9], but with older immatures, adults are more likely to share food that immatures cannot extract themselves than food that can easily be obtained [58–61]. Food offerings to adults occur predominantly in situations where group members cannot reach the food without help, as in the group service paradigm [54] or in dyadic prosociality tasks, also between non-related strangers [52].

The second condition is that the targets of prosociality should not be exclusively selected partners, or kin only. Living in a cooperatively breeding system, adult helpers are often older siblings of the infants, and relatedness is thus high. However, in the wild, immigration events are frequent (2–3 migration events per group and year [62]) and genetic relatedness thus clearly lower than if only full siblings were present [4]. Nonetheless, both in the wild and in captivity, all group members participate in provisioning immatures [9,61], and individuals also behave prosocially towards non-related strangers in experimental dyadic prosociality tasks [52].

Overall, callitrichid groups are tolerant and peaceful [63], with prosocial tendencies expressed towards all group members and potential group members [51]. However, competition with other groups can be fierce, and escalated contests within groups occur too, typically when breeding positions become available, in particular among females [64]. Nevertheless, their everyday life is mostly characterized by high, group-wide tolerance, especially when mutual interdependence is high [65]. Importantly, high degrees of proactive prosociality among adult marmoset group members together with the habitual entrusting of infants to other group members, are a stable basis for a group-wide platform of trust. The comparisons reported above indicate that callitrichids are unique in the level of prosociality among the other primates, where unsolicited food sharing is quite rare [8].

(d) . Tuning in with others

Marmoset interactional ethology is accompanied by a tendency to tune in with social partners on a variety of levels, from hormonal synchronization to vocal accommodation and matching complex behaviours. Hormonally, closely bonded individuals show synchronized fluctuation in oxytocin (OT) over time (days and weeks) [66], and highly synchronized dyads show more prosociality in dyadic prosociality tasks [67]. Moreover, among breeders, pairs with a particularly strong social bond and high OT synchronicity tend to care better for their infants [68]. Vocal tuning in with others is apparent when individuals with a different dialect are paired: they quickly adjust their vocalizations and become more similar to each other (they develop a vocal pair signature [69]; see also below).

At the behavioural level, marmosets even show group-level similarity in personality: individuals from the same group are more similar to each other than to individuals from another group, independent of genetic relatedness [70]. In chimpanzees, which are more closely related to humans but are independent breeders, this was not the case [71]. This group-level similarity may reflect some basic homophily at the group level, and can facilitate cooperative problem solving and communication, but also social learning [72]. For instance, during social learning marmosets readily match the behaviours of others (behaviour copying and imitation; [73,74]). Behaviour matching has also been observed for contagion of emotional arousal [75,76], and perhaps consolation [61,77]. Together, this evidence suggests that similarly to humans, bio-behavioural synchrony [78] and tuning in plays an important organizing role for socio-cognitive abilities in marmosets—they clearly are in synch with each other. Again, evidence for primates other than callitrichids and humans is less abundant, and systematic comparisons remain to be done.

3. Vocal communication among marmosets and other callitrichids

We found that marmosets are remarkably proficient in several contexts typically attributed to the human interaction engine: they show low levels of gaze aversion, frequently engage in mutual gaze and turn-taking, have strong prosocial tendencies that are under some intentional control, and a general proclivity to tune in with their group members. If such an interaction engine does indeed give rise to more elaborate communication, as proposed for human language evolution, this pattern predicts that marmosets and other callitrichids should also have highly elaborate communicative skills. To examine this prediction, we now review the vast literature on marmoset vocal communication.

(a) . General volubility and vocal flexibility

Callitrichids are highly vocal primates with large call repertoires [79–83]. These monkeys encode a variety of information in their calls, such as individual identity, sex, breeding status, group and population identity (e.g. [84–93]). Their calls are complex: they have multiple frequency components [94] and separate calls are often linked together to form call combinations. Cotton-top tamarins (Saguinus oedipus), for instance, concatenate up to four or five calls into strings [81], although only a subset of the whole repertoire is used for this purpose. Common marmosets (Callithrix jacchus) habitually combine up to nine calls together [83], with trigrams being the most common call combination but quadgrams and longer combinations also often present [95]. Similar to human speech, these call bouts consist of sequential units that are interrupted only at specific points [96].

Some calls are functionally referential, for instance, alarm calls (as in other primate species [97,98]). Functional referentiality appears to also be present in food calls, when marmosets encode different food types (fruits versus insects) and recipients use this information accordingly [99,100]. During call production, marmosets show a coupling of the phonatory and articulatory system resulting in a speech-like bi-motor rhythmicity with oscillations that are synchronized and phase-locked at a 3–8 Hz theta rhythm [101]. Other non-human primates typically produce sound modulations that lack concomitant articulatory movements [102–104]. These findings [101] therefore challenge the view that the phono-articulatory rhythmicity that characterizes human speech is absent in non-human primate vocalizations.

When engaging in phee call turn-taking (phee calls are long-distance contact calls; see also above), turns tend to be rather long (68% of all answers occurred in the 5 s after the partners' call, and 78% after 7 s) [105]. However, marmosets can also speed up for turn-taking. When merely separated from their partner by a divider, they give trills (short-distance contact calls) contingently on the partner's trills (and not on the trills of other conspecifics that are likewise present in the colony and audible but not visible). In these trill exchanges, turns are remarkably shorter, with a median latency of 630 ms [106].

Acoustic call structure can be adjusted according to social and environmental factors, a form of vocal learning called vocal accommodation [107]. For example, Sobroza et al. [108] showed that the convergence of calls of two tamarin species (Saguinus midas and Saguinus bicolor) was dependent on both habitat type and territorial overlap. During social vocal accommodation, individuals adjust call structure to optimize call transmission not in a given physical environment but to the call of a social partner, which has been found in multiple callitrichid and other primate species and functions to signal social closeness [107]. In marmosets, translocation experiments revealed that social accommodation spreads through groups and gives rise to different dialects [69,85,109], and if animals were moved between colonies with different dialects they adjusted to the new dialects [110]. Vocal accommodation can also be observed in newly formed couples of marmosets [111,112] where, intriguingly, social accommodation appears to differ between call types. The short-distance trill contact calls accommodated most, to the point of losing some individual recognizability, whereas the long-distance phee calls accommodated less and individual recognizability was not affected [112]. This is biologically meaningful because in the short-distance calls, acoustic individuality is redundant since these calls are typically given when identity can readily be detected visually. Studies in the wild will have to further examine the associated fitness costs.

(b) . Development: vocal learning, scaffolding by adults and babbling

In callitrichids, it is not only the adults that vocalize. Infants are highly vocal as well and produce calls in a variety of contexts. However, infant vocalizations differ from those of adults. Young callitrichids undergo rapid vocal changes in the first months of life before obtaining a fully shaped, functional adult vocal repertoire [113,114]. Various parameters, including the duration, dominant frequency, minimum and maximum frequencies, bandwidth, amplitude modulation frequency and Weiner entropy, undergo changes during vocal maturation [115–119]. Moreover, the frequency of call production and variability of calls decrease over time [115,120,121]. Although these changes can partly be attributed to physical changes related to growth and physiological maturation, social factors play an important role as well, in particular, feedback from adults directly contingent on the infants' calls [115].

Adults frequently respond to vocalizing individuals with their own vocalizations. In general, more contingent adult feedback results in a faster transition to producing fully formed adult calls [115,122,123]. Marmosets that were exposed to much reduced numbers of contingent vocal responses by parents showed slower vocal development [123], confirming social vocal learning processes in infant callitrichids (as opposed to parents simply being more likely to respond to infants that produce more mature calls [114]). Offspring separated from their parents, which thus received limited vocal feedback, produced infant-specific calls for longer and showed a delay in the development of the adult vocal repertoire [124]. They also produced noisier (less mature) calls, differed in the usage of calls, and did not show some typical call combinations [125,126].

A particularly intriguing feature of marmoset vocal ontogeny is that adult group members are actively involved and scaffold development. Their feedback speeds up vocal development, but only if it is provided contingently to the infant vocalizations [123]. For instance, Chow et al. [127] tested adult–juvenile pairs in turn-taking situations. The immatures clearly have to learn the temporal dynamics of proper turn-taking, and they make characteristic mistakes in the beginning. For instance, they may respond to the adult call with latencies that are too short, or they may choose a wrong call to answer. When juveniles called too early, the adults added an extra break before responding to them, and when they chose the wrong call to answer, the adults themselves would interrupt the juveniles with the correct call. Together, this strongly resembles correcting behaviour.

Callitrichid infants often produce long strings of calls (up to several minutes) containing a mix of infant-specific calls such as cries, phee-cries, subharmonic phees and (elements of) adult-like calls, which has been referred to as babbling [113,128]. Babbling has been documented in both the common and pygmy marmosets and follows a similar pattern in both species [121]. Infants start vocalizing in the first week after birth and will babble until they are juveniles (e.g. the phee structure stabilizes when the immatures are about two months of age [113]). Adult marmosets have been shown to interact more with infants when they are babbling [128] and since higher rates of contingent feedback from parents are linked to earlier formation of adult calls [115,120,122], this further points towards socially mediated vocal development.

In sum, even though like in other primates, the call repertoire per se does not change owing to ontogenetic influences, callitrichid infants can be considered vocal learners [129], depending on the definition [130]. Similar to humans and songbirds, the transformation in their calls involves both change in usage and the transformation in the structure of the calls from immature to mature vocalizations [113,114] and is scaffolded by contingent social feedback [15]. This process has not been described in other primates.

(c) . Cognitive components: voluntary vocal control

The null model of non-human primate communication has long been that primate vocalizations are direct expressions of their internal states such as emotional arousal rather than under voluntary control (e.g. [114,131]. Callitrichids deviate from this null model in several ways, which we consider indicative of voluntary vocal control.

As reviewed above, callitrichid turn-taking can be remarkably flexible in various ways. For instance, marmosets discern their partners against a background of other individuals [45,106], or alternate between different affiliative calls depending on the distance from each other [132]. In a follow up study on these findings from the wild, Liao et al. [117] could show that directly measured individual arousal levels cannot fully explain when they switch between calls. Moreover, turn-taking typically involves the exchange of contact calls (e.g. phee calls or trills), and thus transmits rather limited information about the location of the individuals. Under some conditions, however, phee calls can be answered more flexibly. When foraging on a resource-rich spot and hearing a group member giving a phee call from far away, the foraging group members may answer with a chorus of very loud food calls rather than with phee calls [62].

The ability to inhibit automatic call production likewise requires considerable vocal control. Captive tamarins have been reported to engage in whisper-like calling, i.e. reducing the volume of their vocalizations, when the vet (a potential threat) was in the colony [133]. Likewise, there is only a weak correlation between food preference (arousal) and the emission of food calls, among others, because marmosets inhibit food calls towards live prey (mealworms [134]). As mentioned above, marmosets can adjust their sound level to environmental noise (i.e. the Lombard effect), and intriguingly, they are capable of modulating calls accordingly even if the environmental disruption starts after call initiation [135]. This suggests that these animals can counteract the involuntary Lombard effect, and the authors (Pomberger et al.) argue that this may be the result of a volitional articulatory motor network originating in the prefrontal cortex, which cognitively controls the vocal output of the phylogenetically old and widespread primary vocal motor network which predominantly consists of subcortical neuronal networks. Perhaps most impressive with regard to call inhibition and planning are the results of an experiment in which marmosets, engaged in phee call turn-taking, were disturbed by white noise in alternating, predictable intervals of 8 s of noise versus silence. The marmosets adjusted their turn-taking dynamics accordingly: they would either shorten their intervals, such that both the call and answer would fit in the same 8 s of silence, or would increase it, so that after a call they would wait for the noise period to end and answer only afterwards, during the subsequent silent interval [136].

Intentionality in calls is often inferred based on the instrumental and deceptive use of calls [134] and audience effects on call production [55]. Babbling is a typical behaviour of immatures, but sometimes is used by adult individuals to appease more dominant ones around clumped resources. Snowdon [134] shows that this instrumental use of babbling is learned during ontogeny. In our colony, we have observed marmosets giving false aerial predator alarm calls to get access to highly preferred clumped food, or that a young female that was denied access to the infants of the group by its older siblings started to emit food calls without having food at all, which resulted in the infants running towards her so she could pick them up and carry them (J.M.B. and R.K.B. 2018, personal observations).

Audience effects, finally, have been reported systematically in several callitrichid species, especially in the context of food calls. They give more food calls when they are alone than when group members are present, which is consistent with the function of these calls to inform others about the presence of food, to co-feed and eventually share actively with immatures [99,137,138]. Proactive food sharing, i.e. offering food to others that are not begging [9], is a behaviour that is habitual in callitrichids but virtually absent in other primates [8] and typically initiated with food calls. Golden lion tamarins produce specific food-offering calls, and in the wild, they use these calls to offer already extracted food to immatures. For older juveniles, however, they use these calls to attract them to the location where food still has to be extracted, such as a crevice containing a larva [60]. Follow-up studies in captivity could show that juveniles indeed learn from such food-offering calls to expect food in specific substrates [139]. Consistent with these results, several studies in captivity show that callitrichids, when deciding whether to share food with offspring or not, take the skill level and needs of the immatures into account, independently of their age. Accordingly, regardless of the immature age, they share more and for longer with immatures that are not yet able to extract the food themselves, e.g. because it is too difficult for them to open an apparatus that contains food (cotton-top tamarins [59,140]; lion tamarins [141]; common marmosets [58,142]). Consistent with these captive findings, a longitudinal study of wild marmosets revealed that they also share more with the immatures when food in the forest is scarce rather than abundant [61].

In sum, callitrichids are highly vocal monkeys. They stand out among primates, in particular by the sheer amount of vocalizing in cooperative contexts, the fact that the ontogeny of vocal communication is scaffolded by contingent infant–adult interactions, and a considerable amount of vocal control. Although more research is required to pin down the differences between species at a more fine-grained level, the callitrichid vocal communication reviewed here provides support for what had already started to become apparent two decades ago, namely that they have 'a degree of plasticity in vocal signals that is rare among nonhuman primates' [143].

4. Pathways to vocal complexity in cooperative breeders

(a) . The adult perspective

We have shown that marmosets possess critical elements of the human interaction engine, and also excel in vocal communication, which suggests that the former may indeed be driving the latter. Whereas so far it has not been possible to show that this is indeed a causal link, we can explore this possibility by detailing how the specific behavioural affordances of a cooperative breeding system (figure 1) may facilitate vocal communication, from the perspective of both adults and immatures. Here, we provide a short overview of such potential links, which are developed in detail for adults in [13] and for immatures in [14].

For adults to make their cooperative breeding system work smoothly, a prosocial psychology that enables them to freely share food with others is crucial to ensure sufficient provisioning of immatures. The same prosocial mindset may also be the key factor facilitating the propensity to share information, reflected in the high volubility and information sharing via food calls. Practising information sharing on a regular basis will also pave the way for increased cognitive control over it, because if food provisioning can be tailored according to infant needs, tailoring the emission of vocalizations according to others' needs is no longer a big leap.

Living in a cooperative breeding system also brings about an increased need for coordination among group members, which is facilitated if all group members are aware of the location of others. The extensive engagement in turn-taking provides exactly this, and while doing so, all individuals are highly responsive to other group members. Although the full meaning of the numerous calls, often produced in sequences, is still poorly understood, we know that they predominantly occur in affiliative and cooperative contexts, suggesting that they contribute to within-group coordination.

(b) . The immature perspective

Because immature cooperative breeders are primarily recipients of care rather than providers, and must ensure sufficient support, this puts them under a different set of selective pressures early in life (as detailed in [14]).

Among cooperative breeders, mothers are virtually unable to raise an infant without help. As a consequence, they will adjust their investment if they perceive a lack of help in their social environment. They also have an overall higher threshold to keep investing in the offspring, in callitrichids [144,145] and humans [3] alike. Infants therefore need to advertise their vigour to their mothers, which may bias care in their favour when helpers are scarce. Moreover, being cared for by several caregivers means that an infant is not always close to the most motivated helper. Therefore, being able to attract caregivers, or later, when more mobile themselves, being able to discern the one that is currently most motivated to provide care and selectively approach this one would be highly advantageous.

The conspicuous babbling of marmosets may contribute to vocal learning, but it clearly also serves to advertise vigour and attract caregivers. First, if babbling was for vocal learning only, there would be no need for it to be so loud and demanding—it would be dangerous considering the predation risk resulting from the infants' small size [134]. Second, it has been repeatedly shown that babbling infant marmosets are attractive to adults, which are more likely to approach and engage in affiliative contact with an infant that is babbling [128].

Infants of cooperative breeders, then, are growing up in a cooperative, highly interactional environment where adults are busy caring for them and coordinating various activities. The adults appear to take the needs of infants into account to some extent and are eager to provide immediate, contingent care. At the same time, the infants have to ensure they look vigorous enough to be worth investing in by their mothers and make themselves conspicuous to not be overlooked when several immatures compete for care (marmosets usually give birth to twins, and, like in humans, the next set of offspring is born when the previous ones are still far from being independent). Their babbling appears to play a central role in this and when learning to use it in more strategic ways, they also train over long periods of time during their ontogeny to reach vocal control over this vocal behaviour, thus paving the way for vocal control more generally. In short, this is the environment of their cognitive development, in which they will build all the cognitive gadgets [146] that will later constitute their marmoset interaction engine. If instead of a small-brained marmoset infant, a large-brained hominin baby is exposed to such a rich, contingent social environment from birth, this must have more far-reaching consequences, potentially having led to the evolution of our rich socio-cognitive repertoires and eventually language [1,14,16,17,147].

5. Conclusions

(a) . A convergent interaction engine?

The goal of this contribution was to explore the evolutionary origin of the human interaction engine, in particular the idea that the cooperatively breeding callitrichid monkeys may have convergently evolved a set of motivations and socio-cognitive skills considered to be critical components of the human interaction engine. We found that callitrichid monkeys share several of these components, in particular the role of mutual gaze to coordinate joint actions, cooperative turn-taking, prosociality and a group-wide platform of trust, as well as a general tendency to tune in with others, from physiology to behaviour. If such an interaction engine indeed facilitates the emergence of more complex communication, marmosets should also excel in vocal communication, and a comprehensive review of the literature suggests they do indeed, consistent with the cooperative breeding hypothesis [1,3,12,14]. Unfortunately, there has been little or no work on marmoset gestural communication and its integration with vocalizations. Since other, typically less vocal primates often make extensive and flexible use of gestural and facial signals (in particular great apes, see [148]), a more comprehensive comparison should take this multimodality into account to fully judge species differences in expressiveness. Generally, it would be desirable to have a higher number of directly comparable studies from other species that would allow us to pinpoint what is specific to cooperative breeders.

A broader goal should be to empirically define interaction engines specific for every social species, as we can expect some mechanisms that support a specific social lifestyle to exist for all individuals that interact with others. In the simplest case, where social life is not much more than an anonymous aggregation of conspecifics, this must minimally include a preference for being close to others rather than alone, and in some species a mechanism to prevent cannibalism.

The cooperative breeding hypothesis predicts that the interaction engines of different cooperative breeders show systematic overlap. This is clearly an empirical question and direct comparisons between species are notoriously difficult. Nevertheless, evidence for communicative sophistication in cooperative breeders is also accumulating in lineages other than primates, as for instance in birds, where phylogenetically controlled analyses over a large number of species demonstrate a link between cooperative breeding and vocal repertoire [149], as well as between cooperative breeding and experimentally assessed proactive prosociality [150]. In general, babbling appears most prevalent in animals with vocal learning and/or allomaternal care [121], such as the cooperatively breeding giant otters with their large vocal repertoires and immature babbling. Nevertheless, more research is clearly needed, and lineages such as mongoose species [151] or bats [152] that show a broad variation in rearing systems and communicative flexibility are particularly intriguing cases. Among bats, for instance, information sharing was more prevalent in species with allomaternal care [152].

(b) . Our double legacy: dissecting the primate origins of the human interaction engine

From a fully comparative perspective, we can thus not identify ‘the’ evolutionary origin of the interaction engine. The approach we would like to put forward is to ask whether we can trace back the evolutionary origin of the different parts of the human-specific interaction engine. Some of them may be shared with the other great apes (marked C in figure 2), others may be part of our general primate background (B), whereas others may have evolved convergently, for instance as a result of cooperative breeding (A). This approach would also automatically reveal the unique parts of the human interaction engine (D).

Figure 2.

Dissecting the primate origins of the human interaction engine. The human interaction engine can be viewed as composed of elements that are uniquely derived in humans (D) and elements that are shared with our closest living relatives (C), which are presumably homologous, with cooperatively breeding primates (A) which have presumably evolved convergently, and with other primates or non-primates (B). Since each species has a unique interactional ethology that overlaps more or less with that of humans, the same holds for the underlying interaction engines. The focus of this contribution was on elements that humans share with callitrichid monkeys and other cooperative breeders (A). These elements appear to be crucially involved in favouring communicative flexibility as reviewed above and, in the case of humans, the emergence of language. (Online version in colour.)

Our focus was on those elements that we share convergently with other cooperatively breeding primates, and perhaps also cooperative breeders from other lineages, namely A in figure 2. These elements appear to include a prominent role of mutual gaze to coordinate joint actions, cooperative turn-taking, prosociality and a group-wide platform of trust, as well as a general tendency to tune in with others, from physiology to behaviour. We thus explored the cooperative breeding model of human evolution [1,3] from the perspective of the interaction engine approach [18,19].

A caveat in this is that we currently still lack a comprehensive list of what is considered part of the human interaction engine, which makes it also more difficult to assess which elements may be unique to the human interaction engine (D in figure 2). Such elements are most likely related to the ability to recognize and establish common ground, which arguably requires considerable causal understanding and may or may not be based on a mature Theory of Mind and Gricean intentions, and are related to concepts such as shared intentionality [153] that are notoriously difficult to operationalize in a way that allows fair species comparisons [154]. The human interaction engine includes more cognitive elements than the marmoset interaction engine, such as grasping common ground in more complex situations, which requires more thorough causal understanding and better Theory of Mind abilities. These purely cognitive elements are arguably more likely to be shared with other great apes (C in figure 2) than callitrichids because they tend to be linked to brain size, and because the great apes' interaction engine can be expected to be more geared towards strategic cooperation and manipulation [155–157]. Finally, it may turn out that some of the unique elements proposed to be part of the human interaction engine are in fact the consequence of language evolution, rather than being part of a cognitive infrastructure that predated it and enabled its emergence. To address this possibility, cross-cultural studies targeting language universals as well as longitudinal ontogenetic studies are necessary.

Importantly, our approach of dissecting uniquely human achievements into their constituent elements and investigating the evolutionary origin of these elements separately is a very promising one that has also been applied to other uniquely human achievements, such as culture [131,158], general intelligence [159], morality [13] or language itself. For instance, we have known for decades that great apes with their huge and powerful brains are able to acquire quite some mastery of human language systems [160]. However, great apes usually do not offer food and are also less prone to sharing information with others. It is therefore not a big surprise that, even though capable of doing so, language-trained apes hardly ever freely share information. Rather, they use these acquired skills mostly to request things, e.g. food or treats. Marmosets on the other hand are prone to share food and information even though because of their limited technical skills there is not that much information beyond locations, foods and predators that could in fact be shared. It thus appears to be a good working model to think about uniquely human skills as a double legacy, on the one hand of our powerful cognitive apparatus that we have inherited from our great-ape-like ancestors, and on the other hand of novel motivational components that had been added to it as a result of convergent evolution, such as the prosociality linked to cooperative breeding. The interaction between these components, in particular during ontogeny, must have had cascading effects, eventually resulting in our unique set of socio-cognitive skills (1,14).

Data accessibility

This article has no additional data.

Authors' contributions

J.M.B.: conceptualization, writing—original draft, writing—review and editing; J.E.C.A.: writing—review and editing; R.K.B.: writing—review and editing; F.M.M.: writing—review and editing; K.W.: writing—review and editing; C.P.v.S.: writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed herein.

Conflict of interest declaration

We declare no competing interests.

Funding

This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme grant agreement no. 101001295, as well as the NCCR Evolving Language, Swiss National Science Foundation agreement no. 51NF40_180888 and the SNF project 31003A_172979.