Learning in the social context : inference, exploration and evaluation in early childhood

Abstract

Some of the biggest achievements in our lives are made even before we learn to tie our shoes. Within a few years of life, we master a language, acquire cultural norms, and develop naïve, yet rich, abstract, coherent theories about how the world works. How do young learners achieve such a feat? The goal of my thesis is to lay the groundwork for a unified account of a rational inference mechanism that underlies this remarkable human faculty to learn so much, so fast, from so little. The first study (Chapter 2) provides evidence that 16-month-old infants can use co-variation information among agents and objects to infer the cause of their failed actions; depending on their attribution, infants either approached another agent or another object. The second study (Chapter 3) shows that 15-month-old infants consider both the sample and the sampling process to rationally generalize properties of novel objects in the absence of behavioral cues. The results are consistent with the quantitative predictions of a Bayesian model, and suggest that infants' inferences are graded with respect to the probability of the sample. Finally, the third study (Chapter 4) shows that older children make sophisticated inferences about properties of agents; children evaluated an informant based on information he provided, and such evaluations affected how children learned from that informant. These studies provide evidence for rational, probabilistic, domain-general inference mechanisms in preverbal infants, and demonstrate how young learners seamlessly integrate data from different sources in ways that affect their exploration, generalization, and evaluation of both the physical and the social world.