Fig. 2.

Model selection applied to avian visual pattern learning. The depicted perception, generalization and abstraction steps are crucial in pattern learning experiments in humans and other species. Our model selection approach “reverse-engineers” the outcome of these cognitive processes by comparing the likelihood of many possible hypotheses on how pattern learning occurs in different individuals. (See main text for details.) The generalization phase (a) gives rise to a number of hypothetical alternative strategies (b and c), which are compared using relative likelihood (d).