Modelling social learning in monkeys.

Abstract

Quantifying Social Learning Processes In the laboratory, paired demonstrator-observer designs are used primarily to investigate whether animals are capable of specific forms of social learning (Galef 1988). Laboratory experimental have also been used to investigate the diffusion dynamics of learned behaviour through populations in controlled conditions (Lefebvre & Palameta 1988; Whiten et al. 2005). The results of such experiments do not tell us much about social learning in the wild and often lack ecolo gical validity in terms of the behaviour being learned. They have the potential, however, to provide quantitative data that can be used in the parameters for mathematical models of social learning. For instance, Kendal et al. (2007) presented captive groups of callitrichid monkeys with novel extractive foraging tasks. By measuring the proximity of each monkey to the task and noting any food extractions, they quantified the effects of two social learning processes, ‘stimulus enhancement’ and ‘observational learning’, and two asocial processes, ‘intrinsic movement to the task’ and ‘asocial learning of the task’ on the adoption of a novel foraging behavior. The values of these processes from the observed data were fed into a set of models for the spread of a novel behaviour. Model selection was used to discern that the model best-fit to the monkey diffusion data only required asocial processes. Nonetheless, quantification of the processes provided statistical evidence for a small positive effect of stimulus enhancement, where a demonstrator manipulating the task attracts an observer to move to the task, but not for observational learning at the task. Derived parameter values can also be used in competing models to predict the shapes of diffusion curves. Theoretical models predict that the diffusion of cultural traits will typically exhibit a sigmoidal (‘S’ shaped) pattern over time (Boyd & Richerson 1985; Cavalli-Sforza & Feldman 1981), while asocial learning has been expected to result in a linear, non-acceleratory, or at least non-sigmoidal increase in frequency (Lefebvre 1995). In the absence of parameter values to feed into diffusion models, the observed shape of a diffusion curve is unlikely to be reliable, as more recent models predict that under certain conditions asocial learning can generate acceleratory curves while social learning can generate deceleratory curves (Laland & Kendal 2003; Reader 2004). Many competing sets of assumptions can generate very similar diffusion curves using different parameter values. Thus, if parameter values can be estimated, it may be possible to select between competing hypotheses. Population-Level Homogeneity of Behaviour While social learning experiments in the laboratory may provide estimates of the limitations of social learning behaviour, an understanding of social learning will be incomplete without analysis of natural populations. Field translocation experiments have been performed to identify traditions in the wild. For instance, Helfman and Shultz (1984) translocated French grunts Haemulon flavolineatum between