ABSTRACT
The hypothesis is proposed that auditory events are perceived to be partitioned according to boundaries constructed at times of maximum surprise. One way of quantifying surprise is via information theoretic predictions. The results of two experiments are presented that test the plausibility of this hypothesis using simple repeating auditory rhythmic sequences. Local estimators of information content within an auditory sequence are used to construct predictors of perceived segmentation. These predictors are fit to results of the experiment by using a structural equation model and are compared with Garner's Run-Gap model (Garner, 1974). The information theoretic model is found to be a significantly better predictor of the experimental results than the Run-Gap model.
