2.1 Stance selection

Any observed behaviour that is not obviously contributing in a causally transparent way to a readily identifiable end goal may be interpreted through the lens of either the ritual or the instrumental stance, guided by salient social cues, such as comportment, gait, gesture, ostensive cueing, social synchrony, coordination, confidence, experience, seniority, or verbal cues suggesting normativity or expectations of conformism. To illustrate how difficult it can be to select a ritual or instrumental stance on observed behaviour, it is instructive to consider the example of “Sylvia's recipe” (Gergely & Csibra, Reference Gergely, Csibra, Enfield and Levenson2006; see also Whitehouse, Reference Whitehouse2011, Reference Whitehouse2021). Sylvia had a distinctive way of roasting ham. She cut off both ends of the joint before placing it in the oven. She had learned this technique at her mother's knee, never questioning it. Many years later she prepared this dish while her elderly mother was visiting. Observing Sylvia's technique in astonishment, the mother asked her what she was doing. It turned out that when Sylvia was a child the family roasting tin was too small to accommodate an average joint of ham and that is why her mother cut off the ends. The young Sylvia observing her mother's cooking methods never enquired as to the underlying rationale. She may have assumed that the removal of the ends of the joint had some instrumental explanation (e.g., to allow the juices of the meat to flow out) even if that purpose was not immediately obvious. One could equally imagine a cultural milieu in which Sylvia's recipe was accorded a supernatural function, for example to release the spirit of the animal that has given its flesh (quasi-instrumental – see sect. 2.4). But it is equally possible that Sylvia adopted a purely normative, non-instrumental understanding of the procedure, prioritizing its social meanings and functions. Accordingly, cutting off the ends could be a clue to the cook's ethnic origins, via traditions passed down by many generations of mothers before her. Moreover, whenever a normative perspective is activated there is also a further, if rather baffling, possibility: That nobody knows why meat should be prepared in this peculiar fashion, it simply should. Such a response to stipulated ritual procedures is particularly common when the procedures themselves have become familiar habits. In many such cases, scripts that are so familiar that they become sedimented in implicit procedural memory no longer require conscious effort to perform and are thus less likely to prompt conscious reflection on why we do them at all (Whitehouse, Reference Whitehouse2004). When we interpret behaviour in these ways – as a normative convention rather than as an instrumental technique – we adopt a ritual stance (McKay & Whitehouse, Reference McKay and Whitehouse2015; Whitehouse, Reference Whitehouse and Sun2012b).

The story of Sylvia's recipe is instructive because it shows clearly that many observed behaviours could activate either the ritual stance or an instrumental stance – social learning could potentially go either way. It might well be the case, that knowing instrumentally how best to release liquids from meat during the roasting process with the intention of maximizing the tenderness of the resulting dish, could confer prestige on the cook. Clearly, any prestige acquired in that way is a consequence of demonstrated competence at cooking rather than cooperative or affiliative commitment. Nevertheless, cutting off the ends of the meat could just as credibly be understood as a ritual action – that is, a convention that cannot be justified because it leads to an outcome via some knowable physical–causal chain. This would be the case if, for example, Sylvia's mother is thought to have removed the ends of the meat to make it recognizable as a dish associated with a particular cultural group. In some cases, the “done” or “proper” way of presenting food is also associated with an elite stratum in society (e.g., the serving of chilled oysters on the shell) and, because this is the “high class” or “posh” way of doing it, dutifully observing this aspect of the ritualized presentation of food may indeed confer prestige both on the person who has mastered the method and the discerning consumer of the dish.

BST proposes that when we learn or carry out a particular sequence of actions we “flexibly switch between the two based on relevant social cues” (Herrmann, Legare, Harris, & Whitehouse, Reference Herrmann, Legare, Harris and Whitehouse2013). But even though an action sequence may entail the activation of both ritual and instrumental stances at different moments, one or other of the two stances may tend to predominate in a particular action cluster, such that we are more inclined to view it through the lens of the ritual stance (e.g., when kneeling to pray in church, in conformity with the congregation) or the instrumental stance (e.g., when kneeling to fix a bicycle, to facilitate access to a low-lying sprocket). Nevertheless, there are also many action clusters that activate both stances in such rapid succession that it may be hard to decide whether to class them overall as ritual or instrumental. The noted anthropologist Edmund Leach, for example, describes the process of gardening among the Kachin of highland Burma as continually oscillating between practical instrumental elements necessary to ensure plant growth and “aesthetic frills” that are technically unnecessary but conventional aspects of horticulture, identifying the gardener as a bona fide member of the Kachin cultural group (Leach, Reference Leach1954). Much the same may be said of spectator sports such as football (Newson et al., Reference Newson, Shiramizu, Buhrmester, Hattori, Jong, Yamamoto and Whitehouse2020), which have many very ritualistic aspects (from “Mexican waves” to the wearing of special scarves) but also very causally transparent procedures for achieving end goals (such as the placement of a ball at the back of a net). We return briefly to this point in section 2.4.1. when discussing the affiliative aspects of ritual participation and also in connection with quasi-instrumental rituals.

The use of a “stance” metaphor for ritual and instrumental social learning differs from Dennett's (Reference Dennett1987) notion of an “intentional stance,” which suggests a mindset cognizant of the view that the actions of others are governed by beliefs and desires. As discussed in the next section, we do not assume that ritual and instrumental stances operate only in ways that are deliberative and accessible to conscious awareness. We propose instead that the stances are best understood as states of increased attentiveness towards either the link between conventionality cues (Fig. 1) and positive social outcome or the link between instrumentality cues (Fig. 1) and improved goal achievement. In other words, the learner's sensitivity to properties of actions and the models performing them, and propensity to associate these properties with either conventionality or instrumentality, triggers differential learning approaches. In the case of triggered conventionality, the learner is more attentive to action details (such as the chronological order of the action sequence) resulting in overall higher copying fidelity and greater longevity of the modelled action during transmission, while in a scenario where the action is processed as instrumental, attention shifts more towards the outcome, thus resulting in overall lower copying fidelity and higher rates of innovation.

Figure 1.

Schematic overview of the bifocal stance theory (BST).

Importantly, this does not mean that the instrumental stance is incapable of producing stable practices. On the contrary, techniques that have proved reliable in achieving a certain goal often become fixed, preventing regression to more primitive states (Tennie et al., Reference Tennie, Call and Tomasello2009). Conversely, innovation within the ritual domain occurs as well, and deviations from a script may even be encouraged, for example in the case of Congolese spirit play (Lewis, Reference Lewis2002) or status competition among the Tiwi (Hart, Pilling, & Goodale, Reference Hart, Pilling and Goodale1988). Indeed, given that the ritual and instrumental stances are intimately interwoven in many human pursuits, we anticipate that both forms of social learning overlap and influence each other. For example, when learning a particularly complex technical skill, it helps to mix in a little bit of ritual. While too much ritualization impairs learning, perhaps due to cognitive load, a small amount of ritualization may enhance it (Kapitány, Kavanagh, Whitehouse, & Nielsen, Reference Kapitány, Kavanagh, Whitehouse and Nielsen2018). Given examples of ritual innovation, technological stability, and stance admixture, the differences in transmission fidelity between the two stances are predominantly one of degree. All else being equal, practices that inspire copying via the ritual stance tend to be more accurately reproduced relative to those copied via the instrumental stance, making the former comparatively more resistant to change.

Overall, BST therefore seeks to explain the co-existence of adherence to convention and technological innovation in humans from an evolutionary perspective, where certain cues of the copied action influence the nature and degree of copying fidelity.

2.2 Stance deliberateness

A fundamental question prompted by the BST framework, largely unaddressed in previous literature (although see Whitehouse, Reference Whitehouse2002), is the degree of deliberateness with which instrumental and affiliative motivations operate. The stances could be automatic in the sense that they rely on psychological processes to which learners have little conscious access, or deliberative in that they rely on conscious cost–benefit calculations. More specifically, the stances could rely on type 1 or type 2 psychological processes (Evans & Stanovich, Reference Evans and Stanovich2013; Kahneman, Reference Kahneman2003; Norman & Shallice, Reference Norman, Shallice and Davidson1986). Type 1 psychological processes do not depend on working memory; they are typically automatic, fast, associative, effortless, and non-conscious. Type 2 processes make demands on working memory; they are typically deliberative, slow, rule-based, more effortful, and conscious.

It is easy to imagine the stances running on deliberative type 2 processes. In this case, a boy watching an adult cross herself in church would make conscious decisions about the purpose and resolvability of the crossing action. Using explicit mentalizing (also known as “mindreading” and “theory of mind”), he asks himself what the adult intends to achieve through her action, and what she wants to communicate to him (Csibra & Gergely, Reference Csibra and Gergely2009). Using causal reasoning, he asks whether the action generates an instrumental outcome and, if so, mentalizing again, whether anyone understands the action–outcome relationship. He may even engage in normative reasoning, asking whether the crossing action is required, encouraged, or prohibited for him by his social group. Based on these decisions – outputs of type 2 mentalizing, causal reasoning, and normative reasoning processes – the boy adopts a stance, attending to the details (ritual) or overall form (instrumental) of the action, and copies it with greater or lower fidelity with a specific motivation. For example, he may copy the crossing action with high fidelity expecting it to bring him divine grace, acceptance by his community, or the approval of the adult he is copying.

It is harder to imagine the stances running on automatic type 1 processes because these processes are not part of our intuitive psychology. They have been discovered by cognitive science rather than revealed by common sense. In the automatic case, a stance is triggered rather than chosen. Observable features of the action and context, the conventionality and instrumentality cues listed in Figure 1, send the cognitive system into one of the two stances via associative processes. For example, when the boy sees the adult performing an action where the start and end states are equivalent – her hands are joined in prayer when she begins to cross herself and come to rest in the same position – it activates the cluster of psychological processes that constitute the ritual stance, including attention to detail and a diffuse expectation of social rather than nonsocial rewards. He acts expecting to get something good from other people but does not know what it will be or have a theory about why he will get it. Like the associative links that mediate faster responding with the left hand to stimuli on the left of the visual field (spatial compatibility effects; Lu & Proctor, Reference Lu and Proctor1995) and faster reading of colour words printed in the colours they name (Stroop effects, MacLeod, Reference MacLeod1991) the associative links that mediate automatic activation of stances could be innate or learned. More precisely, they could be genetically or culturally inherited (see sect. 5).

Everyday experience suggests that stance switching is at least sometimes deliberative in older children and adults. Most of us can remember asking ourselves in a new social context how faithfully to copy the locals. As a guest at a formal college dinner, should I wear an academic gown or just dress smartly; should I dip my fingers in the finger bowl, or just moisten my napkin? However, automatic stance switching would have significant advantages because it is faster and less cognitively demanding than deliberation. It would allow adaptive modulation of social learning in infancy and early childhood, before type 2 executive processes have matured, and agile movement between stances throughout the lifetime. Therefore, it is likely that stance switching, like linguistic “code-switching,” is sometimes deliberative and sometimes automatic. In the case of code-switching, bilingual friends move back and forth between their languages, within a conversation or even within a sentence. Switching can be deliberate – for example, a speaker may decide consciously to switch to English for scientific terms which she knows are more familiar to the listener in that language – but switches can also be triggered automatically by the emotional tone of the conversation, or a quizzical look from the listener (Pietikäinen, Reference Pietikäinen2014).

Priorities for future research are to discover whether stances can be selected automatically and, if so, when stance switching is automatic and when it is deliberative. To find out, we need to ask people why they have copied with high or low fidelity, and to check whether their answers match the cue conditions that modulated their copying behaviour. A good match would suggest deliberation, and a poor match would suggest that stance selection occurred automatically, and their statements were rationalizations after the fact (see sect. 2.4.3 on the difference between attributed and evolved functions). Similarly, children and adults could be given tasks that demand working memory alongside social learning tasks that call for stance switching. If these concurrent tasks do not interfere with efficient stance switching, it would suggest automaticity. If efficiency declines as working memory demand increases, it would suggest that, in the tested population and circumstances, stance switching is deliberative. Based on studies of mentalizing (Apperly, Reference Apperly2010; Heyes & Frith, Reference Heyes and Frith2014), metacognition (Goupil & Kouider, Reference Goupil and Kouider2019; Heyes, Bang, Shea, Frith, & Fleming, Reference Heyes, Bang, Shea, Frith and Fleming2020), and other cognitive processes that come in automatic (implicit) and deliberative (explicit) forms, it is likely that stance switching is automatic in infancy and early childhood and then becomes more deliberative, but only in circumstances where there is minimal time pressure and significant risk that automatic processes will produce error.

2.3 Ritual stance, affiliation, and group identity

Why would the copying of irresolvably opaque behaviour be motivated by the desire to affiliate? Rituals serve many social functions, as exemplified by scarification procedures that signal mate quality (Singh & Bronstad, Reference Singh and Bronstad1997), ceremonies that commemorate and promote acts of heroism (Kertzer, Reference Kertzer1988), as well as resource distribution practices that buffer against inequalities due to exogenous factors (Woodburn, Reference Woodburn1982). Despite the multitude of different purposes rituals can serve, their common underlying thread is the regulation of group life – it is through affiliation that the individual gains access to the group as well as the benefits that come with it (resources, mates, cooperation, protection, etc.) (Durkheim, Reference Durkheim1965; Kertzer, Reference Kertzer1989; Khaldūn, Reference Khaldūn1958). We propose several reasons why ritual participation leads to group bonding. One is that arbitrary conventions serve as exclusive identity markers for members of the ingroup. Although (or indeed because) such behaviours may have precisely zero value from a technical–instrumental perspective, participation is an attractive prospect for anyone wishing to affiliate with the ritual community in question. This explains how rituals survive the selective environment of our cultural landscape; they are not replaceable by actions construed in instrumental terms because they fulfil an entirely different function, namely that of maintaining patterns of cooperation, ranging from group loyalty to obligations of reciprocity, conferring a selective advantage on the cultural groups adopting them (although we also discuss the case of magical procedures in section 2.4.2). As such, rituals establish discrete ingroups, in contrast with useful technologies that spread willy-nilly across groups (Whitehouse, Reference Whitehouse2012a).

Rituals also commonly generate social cohesion and thus cooperation via social synchrony, in the form of dancing in rhythm, choral singing, chanting, swaying, and marching, evincing well-documented effects on social bonding and cooperation (Catmur & Heyes, Reference Catmur and Heyes2013; Fessler & Holbrook, Reference Fessler and Holbrook2016; Hagen & Bryant, Reference Hagen and Bryant2003; Hove & Risen, Reference Hove and Risen2009; Mehr, Krasnow, Bryant, & Hagen, Reference Mehr, Krasnow, Bryant and Hagen2021; Wiltermuth & Heath, Reference Wiltermuth and Heath2009). Rituals typically require significant investment of labour and materials despite lacking any technical value or output and, as such, serve as costly signals of commitment to the group (Irons, Reference Irons2001; Sosis & Alcorta, Reference Sosis and Alcorta2003). Indeed, the more costly and fitness-decreasing rituals become, the harder it is to square participation with dislike for the group, enabling cognitive dissonance effects to motivate group love (Aronson & Mills, Reference Aronson and Mills1959). Moreover, modulating the frequency and emotional intensity of collective rituals also produces distinctive effects on the intensity and scale of group bonding: High-frequency, low-arousal rituals facilitate rapid spread of stable identity markers to large populations (such as doctrinal religions and nations) while low-frequency, high-arousal rituals generate highly cohesive localized groups (such as initiation cults, military brigades, and terrorist cells) (Whitehouse, Reference Whitehouse1995, Reference Whitehouse2000, Reference Whitehouse2004, Reference Whitehouse2018, Reference Whitehouse2021).

A growing body of empirical research suggests that the ritual stance is sensitive to social cues indicating that modelled behaviour is conventional, normative, and otherwise relevant to group alignment, and therefore that the motivation to copy such behaviour is affiliative in orientation (Clegg & Legare, Reference Clegg and Legare2016b; Herrmann et al., Reference Herrmann, Legare, Harris and Whitehouse2013). We therefore expect role models who are strongly associated with the group, for example because they are popular or exemplify group values, to trigger the ritual stance among would-be learners. By contrast, the instrumental stance is sensitive to social cues indicating that the behaviour is technically useful and copying is rewarded by the acquisition of new practical skills. Thus, we would expect role models who signal competence, experience, and skill to trigger the instrumental stance among would-be learners.

2.4 BST and quasi-instrumental rituals

Practices signalling conventionality but also promising to deliver desirable end states are frequently reported in the ethnographic record. For example, competitive sports are oriented to outcomes, such as scoring goals, but in conventional, causally opaque ways (e.g., playing a ball into the opposing team's net but without using one's hands). Quasi-instrumentality is also a feature of conventional solutions to coordination problems (e.g., you must only drive on the left to avoid crashing into other vehicles). But it applies also to magical rituals (e.g., you have to utter a particular spell in order to make it rain). In all such cases, part of the causal pathway from action to intended outcome is underspecified (causally opaque). Most importantly, nobody expects there to be a causally transparent reason why you use feet and not hands in one sport, drive on the left not the right in some countries, or repeat this incantation rather than that to make it rain. It is simply the correct way of doing things in this region or that cultural setting – the recognized convention. To the extent that such behaviours activate the ritual stance, they can certainly serve to communicate information about the group alignments of the person adopting them. They can tell us, for example, what sports people grew up with, what countries they live in, and what cultural traditions they uphold. The process of observing these conventions therefore has an affiliative aspect. Nevertheless, in all these cases the instrumental stance is also likely to be activated because there are very salient end goals in play – such as communicating information, placing a ball in a net, driving safely, or ending a drought. Accordingly, this subsection discusses common examples from everyday life (such as games and the acquisition of language), but also focuses on how cases of magic can be accounted by BST, because so many rituals throughout the world's cultures (past and present) are concerned with bringing about end goals.

3.1 Psychological evidence supporting BST

The two functions of social learning (to acquire instrumentally useful information and to affiliate with others) as well as their links to differential copying strategies have been discussed in developmental science for some time (Over & Carpenter, Reference Over and Carpenter2013; Uzgiris, Reference Uzgiris1981). Here we seek to extend awareness of bifocal functionality by putting these discussions of social learning into a context where documented differences in copying fidelity can be linked to the salience of social and action-based cues (see also Fig. 1) that promote either the wish to affiliate via a norm-focused ritual stance or to achieve an end state via a goal-focused instrumental stance (Herrmann et al., Reference Herrmann, Legare, Harris and Whitehouse2013; Legare et al., Reference Legare, Wen, Herrmann and Whitehouse2015; Watson-Jones et al., Reference Watson-Jones, Legare, Whitehouse and Clegg2014). Although BST is grounded in anthropological theories of social learning (Whitehouse, Reference Whitehouse2011), it has been the object of a series of empirical investigations aimed at disambiguating the causes and effects of ritual and instrumental stances respectively using carefully controlled experiments (Herrmann et al., Reference Herrmann, Legare, Harris and Whitehouse2013; Legare et al., Reference Legare, Wen, Herrmann and Whitehouse2015; Watson-Jones et al., Reference Watson-Jones, Legare, Whitehouse and Clegg2014, Reference Watson-Jones, Whitehouse and Legare2016). BST proposes that the differential functions of the two stances (ritual and instrumental; see Box 1) favour a psychological division of labour where differing attentional and motivational states guide social learning in optimized ways. The ritual stance is prompted by irresolvable causal opacity and the absence of a salient goal or goal demotion (Box 1). In contrast, through the lens of an instrumental stance, actions are viewed as having a rational causal structure, each step contributing to an end goal. Accordingly, to be an efficient social learner, selectivity in the degree of copying fidelity is required. When learning a novel behaviour via action copying, we may rely largely on verbal or contextual cues that decide whether an action is best interpreted as conventional or instrumental (Clegg & Legare, Reference Clegg and Legare2016a, Reference Clegg and Legare2016b; Herrmann et al., Reference Herrmann, Legare, Harris and Whitehouse2013; Legare et al., Reference Legare, Wen, Herrmann and Whitehouse2015). Nonetheless, explicit cues in the form of verbal instructions are not always available (e.g., in the absence of pedagogy), so the action type itself often constitutes a source of either instrumentality or conventionality (Watson-Jones et al., Reference Watson-Jones, Legare, Whitehouse and Clegg2014). If verbal or other extraneous contextual cues signal conventionality, then a learner should be sensitive to the action's social value and copy it with higher fidelity. Watson-Jones et al. demonstrate such a link by testing the psychological impact of witnessing third-party ostracism on copying fidelity in a sample of 3–6-year-olds when ritual versus instrumental stances were made differentially salient. Children were shown video primes consisting of moving geometric figures reacting in an exclusionary (ostracism condition) or inclusionary manner (control condition) when approached by a new shape and were then encouraged to copy actions shown by a model. The sequences were either conventional, exhibiting the same start and end states (Fig. 2), or instrumental, ending in a different end state than the one they began with. Copying fidelity was highest for conventional actions in the ostracism prime condition. As predicted by BST, the attentional advantage that causally opaque action holds over its transparent counterparts in social settings is also exacerbated by directly experienced ostracism, as shown by follow-up designs (e.g., Watson-Jones et al., Reference Watson-Jones, Whitehouse and Legare2016) that manipulated group affiliation (ingroup/outgroup) using the Cyberball paradigm – a virtual ball tossing game in which participants are either included (base condition) or excluded (ostracism condition). Children who were subject to exclusion from their ingroup showed the highest copying fidelity of a causally opaque action, surpassing the copying accuracy of those who were previously included by their ingroup as well as those who were rejected or welcomed by an outgroup. It appears that the anxieties associated with social exclusion, even in a very mild form, have potentially far-reaching consequences for how modelled action is processed and attended to by social learners.

Figure 2.

Start- and end-state equivalence, where item positions in the first step (first tile) of the sequence are identical to the end-state configuration (last tile), serves as conventional action condition (from Watson-Jones et al., Reference Watson-Jones, Legare, Whitehouse and Clegg2014).

The studies reviewed above show that either directly or indirectly experienced negative interactions with others can cue changes in young learner's copying behaviour, demonstrating sensitivity to links between causally opaque action and favourable social outcomes. Such results are potentially consistent with the ritual stance being innate, domain-specific, and rule guided (e.g., a biologically inherited module that is only active in social interactions and that follows a “copy when ostracized”-rule). Although such an account provides one possible framing (Whitehouse, Reference Whitehouse2021), we consider here the possibility that the observed data could just as plausibly be produced by domain-general processes such as associative learning, in which excitatory and inhibitory links between certain action types and social outcomes are forged during development (Heyes, Reference Heyes2012). For instance, the increased negative affect from social exclusion (rather than the exclusion itself) could result in additional efforts expended which, paired with the prediction errors that come from witnessed opacity, could result in overall higher fidelity scores for causally opaque action sequences but not causally transparent ones. Thus, an increase in non-specific arousal might have differential effects on ritual and instrumental copying, where heightened energetic investment in sequence copying could manifest in overall heightened fidelity, whereas putting more effort into endpoint copying does not result in as much accuracy gain given that the focus lies on the last step, the end goal. This potential alternative to a social reparatory account stresses the importance of efficient control conditions (e.g., general types of arousal as opposed to exclusion alone) in future BST designs to disentangle the potential specificity of stances from outputs that operate on mechanisms of general-purpose learning. If copying fidelity of opaque action is heightened in asocial conditions then domain-generality becomes a plausible interpretation.

Despite these outstanding questions that clearly show the importance of integrating BST into other fields of enquiry, the reviewed experiments are in line with much of the developmental literature and give rise to the reasonable assumption that at least a basic understanding of action–reward links precedes the fully fledged bifocal stance arrangement with which adults navigate society quite efficiently. Indeed, the literature documents a multitude of cases in which learning facilitates the acquisition and storage of social information such as norms, values, and beliefs (Legare, Evans, Rosengren, & Harris, Reference Legare, Evans, Rosengren and Harris2012; Whitehouse, Reference Whitehouse2011). Norms are characterized as mutual agreements on how members of a group ought to conduct themselves in various social contexts and their acquisition marks a key milestone in development (Chudek & Henrich, Reference Chudek and Henrich2011; Kenward, Reference Kenward2012; Kenward et al., Reference Kenward, Karlsson and Persson2011). Three-year olds intervene when witnessing norm violations through the use of normative language (Vaish, Missana, & Tomasello, Reference Vaish, Missana and Tomasello2011). Interestingly, children protest especially when an in-group member violates a norm but remain passive when witnessing transgressions committed by an unwitting outsider (Schmidt, Rakoczy, & Tomasello, Reference Schmidt, Rakoczy and Tomasello2012). Further, pretend play paradigms show how children are committed to arbitrary rules and the proper way to handle play scenarios (Nielsen, Reference Nielsen2012; Rakoczy et al., Reference Rakoczy, Warneken and Tomasello2008; Rakoczy, Brosche, Warneken, & Tomasello, Reference Rakoczy, Brosche, Warneken and Tomasello2009a, Reference Rakoczy, Warneken and Tomasello2009b). In Keupp et al. (Reference Keupp, Behne and Rakoczy2013), children protested more when a third party omitted irrelevant actions whenever those were labelled as conventional, pointing towards a normative interpretation of play behaviour. Similarly, Clegg and Legare (Reference Clegg and Legare2016a, Reference Clegg and Legare2016b) demonstrate that linguistic cues that are indicative of convention rather than instrumentality boost copying fidelity of 3- to 6-year-olds in a necklace-making activity, further reinforcing the notion that the interpretation of the context in which knowledge is acquired influences the way it is transmitted. Moreover, recent findings indicate that norm acquisition follows similar developmental trajectories across different societies (House et al., Reference House, Kanngiesser, Barrett, Broesch, Cebioglu, Crittenden and Yilmaz2020). Overall, the reviewed findings paint a coherent picture of children's tendencies to attribute normativity when prompted, but instances of self-inferred norm attribution, occurring in contexts where it is not explicitly prompted, are also documented (Schmidt et al., Reference Schmidt, Butler, Heinz and Tomasello2016). Again, it is tempting to resort to cognitively rich explanations that accord children a fully fledged understanding of norms, group affiliations, and social dynamics. These findings are nonetheless suggestive that an early emerging rudimentary capacity to associate certain types of actions with certain types of outcomes lays the foundations for an appreciation of the affiliative value of cues signalling non-instrumentality. Accordingly, the ritual stance is geared towards obtaining social rewards, while the focus on reaping inanimate benefits within the context of goal achievement marks the instrumental stance (Herrmann et al., Reference Herrmann, Legare, Harris and Whitehouse2013; Legare et al., Reference Legare, Wen, Herrmann and Whitehouse2015; Watson-Jones et al., Reference Watson-Jones, Legare, Whitehouse and Clegg2014).

In addition to studies tailored to test BST and normative protest, basic research on “overimitation” (Box 1) (Call, Carpenter, & Tomasello, Reference Call, Carpenter and Tomasello2005; Horner & Whiten, Reference Horner and Whiten2005; Nagell, Olguin, & Tomasello, Reference Nagell, Olguin and Tomasello1993; Whiten, Custance, Gomez, Teixidor, & Bard, Reference Whiten, Custance, Gomez, Teixidor and Bard1996) provides further support for our framework. While nonhuman primates sometime copy simple instrumental behaviours, giving rise to localized tool-using cultures in some natural settings (Boesch & Boesch-Achermann, Reference Boesch and Boesch-Achermann2000), humans copy not only technically useful behaviours, but also arbitrary social conventions to an extent unmatched by other primates (Horner & Whiten, Reference Horner and Whiten2005). Since at least the late Palaeolithic, we have produced and passed down increasingly complex cultural traditions, building cumulatively over generations (Hodder, Reference Hodder2012). This appears to have resulted from a subtle change in human learning strategies whereby instead of copying pragmatically any behaviour that produces desired outcomes, we also began to copy behaviour for which no causal rationale was available. Studies of the phenomenon have been carried out with western samples of both adults and children (Flynn & Smith, Reference Flynn and Smith2012; McGuigan, Gladstone, & Cook, Reference McGuigan, Gladstone and Cook2012; McGuigan, Makinson, & Whiten, Reference McGuigan, Makinson and Whiten2011; Whiten et al., Reference Whiten, Allan, Devlin, Kseib, Raw and McGuigan2016), as well as in a broad range of other cultural groups, from hunter–gatherers in Africa to swidden horticulturalists in Melanesia (Clegg & Legare, Reference Clegg and Legare2016a; Corriveau et al., Reference Corriveau, DiYanni, Clegg, Min, Chin and Nasrini2017; Nielsen & Tomaselli, Reference Nielsen and Tomaselli2010; Nielsen, Mushin, Tomaselli, & Whiten, Reference Nielsen, Mushin, Tomaselli and Whiten2014; Taniguchi & Sanefuji, Reference Taniguchi and Sanefuji2017).

Studies of overimitation often present research participants with tasks involving the extraction of a reward from a puzzle box following a demonstration of both causally transparent and causally opaque behaviours by the model (e.g., Horner & Whiten, Reference Horner and Whiten2005). In some cases, causal opacity is achieved by the insertion of behaviours that do not help with the extraction of the reward (e.g., “tapping the side of the jar with a feather and then unscrewing the lid”; Lyons et al., Reference Lyons, Young and Keil2007). In other experiments, such behaviour does contribute to the end goal but only inefficiently (e.g., pushing a light switch with one's forehead rather than using one's hands; Buttelmann, Carpenter, Call, & Tomasello, Reference Buttelmann, Carpenter, Call and Tomasello2007; Nagell et al., Reference Nagell, Olguin and Tomasello1993). And in yet other designs, a more efficient method is available but overlooked by the model (Corriveau et al., Reference Corriveau, DiYanni, Clegg, Min, Chin and Nasrini2017). In all such cases, human participants in the experiments copy the unnecessary or inefficient behaviour rather than prioritizing only the causally relevant aspects. Accordingly, BST proposes a much more pivotal role of non-instrumental imitative strategies, by drawing attention to its ability to act as signals of sharedness and aligned intention between actor and imitator (Over & Carpenter, Reference Over and Carpenter2012; Tomasello, Carpenter, Call, Behne, & Moll, Reference Tomasello, Carpenter, Call, Behne and Moll2005). For instance, using a puzzle box overimitation paradigm, Nielsen and Blank (Reference Nielsen and Blank2011) have shown that children tend to not copy the causally irrelevant parts of a solution when the demonstrator is absent during copying. Further, when previously primed by either a mimicry game or an instrumental game, 2-year-olds overimitate or engage in selective copying, respectively (Yu & Kushnir, Reference Yu and Kushnir2014). Most importantly, the effects of the mimicry prime only persist if the experimenter is the same across both phases (thus excluding simple motor priming as an explanation). Additionally, 2-year-olds were also found to make use of unnecessary tools only in conditions where the model is physically present as opposed to shown on a screen (Nielsen, Simcock, & Jenkins, Reference Nielsen, Simcock and Jenkins2008). When learners copy observed behaviour accurately, they temporally increase similarities between themselves and the model, which in turn increases their likelihood of being perceived favourably by the model as shown by research documenting that children preferentially help those that imitated them previously (Carpenter, Uebel, & Tomasello, Reference Carpenter, Uebel and Tomasello2013; Over, Carpenter, Spears, & Gattis, Reference Over, Carpenter, Spears and Gattis2013). Increased copying fidelity observed in scenarios of overimitation has the effect of conserving shared conventions in the cultural repertoire. This is in line with BST's core premise that the copying of causally irrelevant elements is because of the motivation to affiliate with the model. There are various potential points of proximate origin from which such motivations can spring. For instance, social learners might retrospectively pick up a link between opaque copying and the presence of social rewards, in a way that the behaviour becomes habitual and thus persists even in the absence of further incentivisation. A learning individual may also extrapolate an observed relationship between causally opaque elements and affiliative gains onto other exemplars of ritual behaviour (in a prospective fashion). Most likely both retrospective and prospective learning modes are involved in the ontogeny of BST, which is discussed in more detail in section 5.

We now turn to accounts of overimitation that potentially compete with BST. These suggest that non-instrumental copying is a mere by-product of our strong reliance on culture to generate technological solutions. Overall, BST provides an alternative view to such by-product explanations which attribute little to no importance to the type of action that is transmitted. Moreover, we argue that BST provides a more encompassing framework within which to understand the broad range of research findings on social learning.

3.2 By-product accounts of non-instrumental copying

One possible explanation for replication of instrumentally superfluous action might be the learner's over-estimation of its causal efficacy (Lyons et al., Reference Lyons, Young and Keil2007, Reference Lyons, Damrosch, Lin, Macris and Keil2011). When children copy causally opaque behaviour, it could be because they think the model is trying to help them to acquire useful information and skills, even if it is not immediately obvious how or why (Buchsbaum, Gopnik, Griffiths, & Shafto, Reference Buchsbaum, Gopnik, Griffiths and Shafto2011; Hoehl, Zettersten, Schleihauf, Grätz, & Pauen, Reference Hoehl, Zettersten, Schleihauf, Grätz and Pauen2014; Vredenburgh, Kushnir, & Casasola, Reference Vredenburgh, Kushnir and Casasola2015). For example, an influential early study on this topic revealed that infants were willing to copy unnecessarily tortuous methods of activating a light switch (using the forehead instead of the hands) after observing an adult model doing it that way (Gergely, Bekkering, & Király, Reference Gergely, Bekkering and Király2002). The infants in this study were, however, sensitive to clues as to why the adult was behaving so strangely. Thus, if the model was tightly wrapped in a blanket to constrain any arm movements, the infants inferred that this was why she used her forehead in preference to her hands to activate the switch and so they were more likely to use their hands to activate the light rather than to copy the more inefficient method modelled to them. This has led some psychologists to conclude that we only copy causally opaque behaviour on the assumption that it must have a sound rationale, even if that is not apparent to us right away. It has been suggested that such causal misattribution can be an implicit process, for instance when children given training to identify correctly the causally unnecessary elements in a modelled procedure still proceed to copy the irrelevant actions (Lyons et al., Reference Lyons, Damrosch, Lin, Macris and Keil2011). In these studies, children are actively encouraged to recognize that the actions are unnecessary and discouraged from reproducing them and yet they still engage in overimitation. A possible explanation for this is “automatic causal encoding” – that is, the copying occurs outside conscious control (Lyons et al., Reference Lyons, Young and Keil2007). In other words, when children copy the unnecessary actions, they are motivated by implicit expectations of causal structure and efficacy without being aware of it. These interpretations assume that non-instrumental copying is merely a by-product of instrumental copying – hyperactive causal attribution which automatically over-ascribes instrumental goals to all goal-demoted and causally opaque actions. These accounts assume that the costs of falsely ascribing instrumental efficiency to functionally useless actions are offset by instances where those actions turn out to be teleologically efficacious.

Misattribution accounts are not without their critics (Kenward, Reference Kenward2012; Kenward et al., Reference Kenward, Karlsson and Persson2011). In line with BST, Kenward et al. (Reference Kenward, Karlsson and Persson2011) propose that overimitation, rather than stemming from distorted causal beliefs, acts as facilitator in the development of norm acquisition. They found that when operating a puzzle box, children (aged 4–5) copy unnecessary actions despite verbally declaring their inconsequentiality to reward extraction. Answers that signal either unquestioned copying or deontic reasoning were given upon further prompting, suggesting a normative mindset that uses opacity to achieve social rewards.

Considering these findings, it may be tempting to regard these two interpretations of overimitation – automatic causal encoding versus affiliative motivation – as mutually exclusive alternatives. That is, the copying of causally opaque behaviour is either motivated by expectations of hidden causal structure based on the imputation of pedagogic motivations (the “instrumental stance”) or by expectations of convention learning based on imputation of normative motivations (the “ritual stance”). Instead, BST proposes that humans oscillate between stances during social learning, responding flexibly to the kinds of cues available to them when they observe modelled behaviour. Accordingly, “when the model is thought to have expertise in the performance of a task (signalled by cues like confidence, experience, success, and authority) one may anticipate an opportunity to learn something of practical use about the affordances of objects and so adopt an instrumental stance on the behaviour. But when the model is thought to be exemplifying a ‘proper’ or normative way of behaving (signalled by cues relevant to affiliation, conformism, or deference to tradition) it may be more appropriate to adopt a ritual stance on the behaviour and assume that it is simply the correct or ‘done’ way rather than the most causally efficacious way, of acting” (Whitehouse, Reference Whitehouse2021, p. 35).

By-product reasoning also features in the action parsing model as applied to ritual (Lienard & Boyer, Reference Lienard and Boyer2006; Liénard & Lawson, Reference Liénard and Lawson2008). Action parsing encompasses three levels on which observed actions are processed: the script level, which describes the overarching theme of the action sequence (e.g., tidying the kitchen), behaviour, which is a direct and end-focused description of the action (e.g., cleaning a glass), and gesture, which describes the discrete sub-components of the behaviour (e.g., raising the glass, rubbing it with a cloth, etc.; Kapitány & Nielsen, Reference Kapitány and Nielsen2017). Advocates of this approach argue that if instrumental outcomes of an action sequence are prevented by a missing causal link within the sequence (e.g., the glass and cloth are both raised but not brought into contact with each other during performance of a rubbing motion) then the attention of the observer shifts down to the gestural level. Accordingly, when witnessed, causal opacity and goal demotion cause a prediction error that comes from failing to parse the observed sequence on the level of behaviour, thus promoting “cognitive capture,” which is characterized as increased allocation of attentional resources towards gestural components (Kapitány & Nielsen, Reference Kapitány and Nielsen2017; Lienard & Boyer, Reference Lienard and Boyer2006). This is in line with “predictive coding” accounts, proposing that the discrepancies between top-down predictions and bottom-up sensory stimulation result in recruitment of additional neural resources geared towards updating the internal model of the agent and improving the accuracy of future estimates (Baldeweg, Reference Baldeweg2006; Friston & Kiebel, Reference Friston and Kiebel2009; Huang & Rao, Reference Huang and Rao2011). Predictions based on expected causality are therefore violated by goal absence, requiring ambiguity resolution. Indeed, the attention-grabbing nature of causal opacity improves recall (Kapitány et al., Reference Kapitány, Kavanagh, Whitehouse and Nielsen2018) and objects linked to opaque treatments are perceived as more special and desirable than occurring in causally transparent sequences (Kapitány & Nielsen, Reference Kapitány and Nielsen2015). Causally opaque actions may make particularly unique (and hence efficient) identity markers precisely because they allow much greater combinatorial freedom on the gestural level rather than being confined to the realm of goal-directed behaviour anticipated by the observer.

The proposition that “steps become goals,” that in causally opaque action sequences the absence of causal structure produces errors within the observer's mental prediction system that shifts attention away from the endpoint towards the sequence constituents, finds some limited support in developmental research. For instance, in one study (Carpenter, Call, & Tomasello, Reference Carpenter, Call and Tomasello2005), 12-month-olds observed pretend play scenarios performed by adults (making a toy mouse hop across a mat with sound effects) either with a salient endpoint (the mouse was placed in a toy house) or not (it remained on the mat). Copying of play sequences was found to be more accurate (hopping and sounds were reproduced) in the latter condition, suggesting that the absence of a goal prompted children to focus on the fine-grained level of gestural components. In similar designs, the level of fidelity with which young children copy decreases in conditions where they understand the purpose of the sequence (Williamson & Markman, Reference Williamson and Markman2006). Increased attention attributed to causally opaque action can therefore partially account for the heightened copying fidelity with which conventions and rituals are reportedly transmitted. Thus, the structure of an action sequence itself could drive observed differences in copying fidelity. As with explanations of casual misattribution given in the context of overimitation, the action parsing account would propose that transmission of causally irrelevant action is a by-product of error prediction systems that evolved for instrumental purposes. However, as argued by BST, which considers the tremendous social value of ritual participation (Whitehouse & Lanman, Reference Whitehouse and Lanman2014; Xygalatas et al., Reference Xygalatas, Mitkidis, Fischer, Reddish, Skewes, Geertz and Bulbulia2013), this is unlikely to be the end of the story. For instance, the action parsing model fails to predict differences in copying fidelity primed by ostracism as foreseen by BST (and as reviewed above). Nothing about the level at which an action is parsed and attended to indicate why it should be more suitable for affiliation than some other type of action. Ultimately, we need BST to explain this type of modulation, where copying fidelity is not exclusively driven by attentional processes on the level of parsing (the cognitively “parasitic” nature of opacity and goal demotion) but also by the learner's social concerns and motivations during the process of transmission. More generally, we argue that the value of BST lies in its ability to add explanatory depth and nuance to the existing literature, while remaining consistent with the accounts described above: Causal misattributions, prediction errors, and the level of parsing, are all likely contributors to the high-fidelity transmission of conventions and traditions but none of them is sufficient to explain observed differences between ritual and instrumental learning.

In addition to its relevance to various sub-fields of social learning and cultural evolution research, another hallmark feature of BST is its emphasis on the importance of including different action types in experimental paradigms of social learning in order to disentangle its underlying motivations. In what follows, we make the case for using fine-grained distinctions between action characteristics in future research, stemming from a new framework of cultural action.