Figure 1.

Schematic showing an example sound sequence and how the idealized learning model combines heard sounds with a simple prior to form expectations about upcoming sound events (the “updated belief” above). The next sound then conveys some amount of complexity according to these probabilistic expectations of the updated belief. The “Goldilocks” hypothesis holds that infants will be most likely to terminate their attention to the sequence at sounds that are either overly simple (predictable) or that are overly complex (unexpected), according to the model. Thus, sounds to which the updated belief assigns either a very high probability (e.g., sound A) or a very low probability (e.g., sound C) would be expected to be more likely to generate attentional termination (look-aways) than those to which it assigns an intermediate probability (e.g., sound B).