| dc.description.abstract | Large-scale thematic structure—the organisation of material within a musical composition—holds an important position in the Western classical music tradition and has subsequently been incorporated into many influential models of music cognition. Whether, and if so, how, these structures may be perceived provides an interesting psychological problem, combining many aspects of memory, pattern recognition, and similarity judgement. However, strong experimental evidence supporting the perception of large-scale thematic structures remains limited, often arising from difficulties in measuring and disrupting their perception. To provide a basis for experimental research, this thesis develops a probabilistic computational model that characterises the possible cognitive processes underlying the perception of thematic structure. This modelling is founded on the hypothesis that thematic structures are perceptible through the statistical regularities they form, arising from the repetition and learning of material. Through the formalisation of this hypothesis, features were generated characterising compositions’ intra-opus predictability, stylistic predictability, and the amounts of repetition and variation of identified thematic material in both pitch and rhythmic domains. A series of behavioural experiments examined the ability of these modelled features to predict participant responses to important indicators of thematic structure. Namely, similarity between thematic elements, identification of large-scale repetitions, perceived structural unity, sensitivity to thematic continuation, and large-scale ordering. Taken together, the results of these experiments provide converging evidence that the perception of large-scale thematic structures can be accounted for by the dynamic learning of statistical regularities within musical compositions. | en_US |
