Learning from the best, monkeywise

Nutcracking by capuchin monkeys has become the best non-hominoid example of tool use, and serves as a marker of the potential for anthropoids to develop and maintain cultures. This kind of tool use -- to extract hard-to-get food from mechanical defenses like shells or burrows -- is called extractive foraging.

For capuchins, this kind of extractive foraging allows them to exist in habitat where easier-to-get resources are limited for parts of the year. It is, in other words, a cultural tradition with immediate adaptive value. So exactly how capuchins learn to do this is pretty important.

Ottoni and colleagues (2005) observed the ways that young capuchins acquire information relevant to tool use. The abstract:

The present work is part of a decade-long study on the spontaneous use of stones for cracking hard-shelled nuts by a semi-free-ranging group of brown capuchin monkeys (Cebus apella). Nutcracking events are frequently watched by other individuals - usually younger, less proficient, and that are well tolerated to the point of some scrounging being allowed by the nutcracker. Here we report findings showing that the choice of observational targets is an active, non-random process, and that observers seem to have some understanding of the relative proficiency of their group mates, preferentially watching the more skilled nutcrackers, which enhances not only scrounging payoffs, but also social learning opportunities.

This might seem obvious -- that is, if you want to learn how to crack nuts, then you ought to watch somebody who really knows how to do it. But in terms of time investment it is a bit unusual -- these young capuchins spend a lot of time playing with other young individuals who aren't the best nutcrackers. The research found that scrounging for scraps was one good reason to watch a good nutcracker:

These results point to an active choice of observational targets, strongly suggesting that the nutcracking proficiency of the potential targets is taken into account by the observers. A motivation to learn from a more skilled conspecific does not necessarily have to be a causal factor in the active choice of the target: besides plain social interest and the attractive nature of a noisy activity, another, a more immediate motive -- scrounging -- can provide a parsimonious explanation for the proximate causation of selective observation of nutcrackers (although it can certainly contribute to learning as a side effect). But in this case, the proficiency of the target is even more clearly at issue, since the choice of the most proficient nutcrackers as observational targets would tend to yield, on average, the highest scrounging payoffs.

And it is significant for the transmission of tool use that proficient nutcrackers tolerate this kind of scrounging:

This capuchin group is, most of the time, a relatively relaxed society (Ferreira 2003), where younger individuals are highly tolerated. During the conspecific nutcracking observation and scrounging episodes reported here, no events of antagonism were registered.

So the basic idea is that young capuchins have an ability to "pick out" good toolmakers for observation. The authors relate this ability to "triadic awareness" -- the ability to understand features of a relationship among two other individuals:

Evidence of the sort of triadic awareness implied by an animals capacity of understanding some features of the relationship between two group mates has been experimentally provided by studies of Old World monkeys (Dasser 1988; Cheney and Seyfarth 1990; Silk 1999). Among New World species, Perry et al. (2004) showed that white-faced capuchins (Cebus capucinus) seemed to exhibit triadic awareness, since they tended to solicit coalition partners not only dominant to their opponents, but also with better relationships with themselves than with those opponents.
Our results strongly suggest that capuchin monkeys are capable of discriminating group mates according to their tool-using skills -- being more prone to watch more proficient nutcrackers. In this particular case, scrounging and social learning are not necessarily conflicting tactics, since both are best served by a watch-the-most-successful strategy (Laland 2004). These preferences result in a sort of prestige hierarchy -- distinct and independent of dominance -- such as proposed by Henrich and Gil-White (2001) as a key feature in the optimization of human cultural transmission. Whether the decision-making process involves an actual understanding of relative proficiency ranking or stems out of simple differential associative or reinforcement histories connected with each individual Target, remains to be verified -- but whatever be the underlying cognitive mechanism, this capability can play a decisive role in the establishment of tool use as a behavioural tradition.

Of course, this "triadic awareness" is precisely the kind of logical information that Bateson is attempting to characterize -- features of a relationship emerge from the interaction between two individuals and cannot necessarily be abstracted from knowledge of the individuals alone. (Ah-ha! Now, all this Bateson-quoting starts to make sense....)

A dominance hierarchy can spontaneously emerge just as a function of dyadic relations. No individual needs to know his or her exact place in such a hierarchy, he or she need only know the outcome of interactions with other individuals for the hierarchy to have force.

Triadic awareness allows more interesting strategizing, as pointed out above. It is certainly a desirable trait for individuals who can form coalitions, although it may not be necessary if coalitions are generally formed in opposition to top-ranking (alpha) individuals -- it shouldn't take triadic awareness for non-alpha individuals to coordinate their actions to depose an alpha.

But such awareness certainly should be necessary for a prestige hierarchy. At a minimum, observers of a prestige hierarchy must be able to assess the interactions of other individuals with objects -- exactly who is having some success cracking nuts, for example. That "awareness" consists of a prediction of the typical outcome of interactions -- that is, when you hear individual X cracking nuts, you know that a lot of nut fragments are going to be flying around him. But the prediction based on the interaction is attributed to the individual.

Attributing social qualities to individuals may be a bit more difficult than this -- because the interactions are more complex, and are products of two individuals instead of one -- but they may be of a similar logical type.


Ottoni EB, de Resende BD, Izar P. 2005. Watching the best nutcrackers: What capuchin monkeys (Cebus apella) know about others' tool-using abilities. Anim Cogn 24:215-219. DOI link