In theories of auditory scene analysis and melodic implication/realization, melodic expectation results from an interaction between top-down processes (assumed to be learned and schema-based) and bottom-up processes (assumed innate, based on Gestalt principles). If principles of melodic expectation are partly acquired, it should be possible to manipulate them – to condition listeners' expectations. In this study, the resistance of three bottom-up expectation principles to learning was tested experimentally. In Experiment 1, expectations for stepwise motion (pitch proximity) were manipulated by conditioning listeners to large melodic leaps; preference for small intervals was reduced after a brief exposure. In Experiment 2, expectations for leaps to rise and steps to fall (step declination) were manipulated by exposing listeners to melodies comprising rising steps and falling leaps; this reduced preferences for descending seconds and thirds. Experiment 3 did not find and hence failed to alter the expectation for small intervals to be followed by an interval in the same direction (step inertia). The results support the theory that bottom-up principles of melodic perception are partly learned from exposure to pitch patterns in music. The long-term learning process could be reinforced by exposure to speech based on similar organization principles.
Underconstrained stochastic representations for top-down computational auditory scene analysis