The compositional hierarchical model is a deep architecture characterized by transparency, explicitness of learned concepts, and the ability to learn on small datasets. The model was tested on the task of melodic expectation with the prior knowledge of the music of different cultures, and its performance in terms of the correctness of predictions was compared with human performance. An experiment was conducted, assessing the ability of two groups of participants---European (Slovene) and Chinese---to predict the continuations of Western and Chinese musical excerpts. Familiarity with the musical style contributes to a lower perceived complexity, and we found that the same applies to the task of musical expectation: the participants were more successful in predicting the continuations in the music of their own culture than the foreign one. Furthermore, the model also adapted the pattern modelling method with regard to the different types of music learned, and in some aspects, it was even more successful than people.
|
