Background. According to Berlyne (1971), preference for stimuli is related to the complexity or unpredictability of the stimuli. Although this claim has been supported by a large number of studies in the field of music, adequate objective ways of measuring originality or surprisingness of music are scarce.
Aims. This paper presents a model of melodic complexity that is based on melodic expectancies and examines its predictions along with those of two other models. The three models are the expectancy-based model, tone-transition probability and information-theoretic models respectively. The expectancy-based model integrates research on music perception to provide a single measure of the extent to which listeners might be expected to find a given melody predictable. This model comprises of a series of contributory variables, drawn from empirical findings and theoretical models.
Method. The performance of the models was tested in an experiment where participants assessed the melodic complexity of a range of melodies. The participants were 56 psychology undergraduates (M = 18.80, SD = 0.93) who rated the complexity of 41 melodies on an 11-point scale. The melodies ranged from short, artificial pieces to more realistic tunes. These represented a wide variety of complexity and were ordered in five blocks. Participants' familiarity with the melodies and musical background and was also recorded.
Results. The expectancy-based model was the best predictor of participants' complexity ratings, explaining over 90% of the variance in these. The performance of tone transition probability model was inferior although it was consistently higher than the performance of the information-theoretic model. The key principles of the expectancy-based model such as tonality, proximity and pitch reversal were found to be most important components in the best model.
Conclusions. Findings suggest that the expectancy-based model provides an accurate estimate of listeners' melodic complexity judgements. The primary application of the model lies in using its automated measures of melodic complexity to predict listeners' preferences. This will be illustrated with some data on songs by the Beatles.
Key words. complexity, preference, expectations, predictability