Segmentation of Hungarian folk songs using an entropy-based learning system

A memory-based maximum entropy model has been developed to simulate the learning in an oral musical tradition, with the aim to find the optimal segment streams in melody sections. The model operated on a representative set of 2323 Hungarian folk songs. The pentatonic motive boundaries became more and more preferred during convergence, which shows the self-supporting feature of pentatonality in certain melodic systems. A pronounced correlation between typical segment contours and complete section contours poses the existence of certain typical schemata at macro- and micro levels in the melodies.