Melodic expectations are important in creating patterns of tension and relaxation. The way expectations are formed has been modeled with tonal hierarchies (Krumhansl &Kessler, 1982)and realization of implied intervals (Narmour, 1990). These models have been extensively investigated by asking listeners to rate different possible melodic continuations. However, these models and experimental methods utilize a limited musical context and probe the melody at only one point in time. To clarify the role of a more extended musical context in expectation formation, a more dynamic approach is taken in this study. The aim was to collect continuous ratings of melodic predictability from listeners using monophonic melodies. The dynamics of the continuous ratings were assessed by various models. Listeners rated the predictability of melodies simultaneously while the melody is playing. Two sets of melodies were used:Experiment 1 contained 40 folk melodies with varying degrees of predictability whereas Experiment 2 consisted of 27 artificial isochronous sequences.
Listeners ’predictability ratings of melodies were predicted with three models. Model 1 consisted of Narmour ’s implication-realization principles supplemented with tonal stability and melodic anchoring. Model 2 was based on recursive probability distributions of two-tone continuations where the initial probabilities were derived from a large corpus of folk melodies. Model 3 comprised the entropy of the distribution of the tones within a moving window. Additionally, the applicability of an initial key-finding algorithm in the models was investigated. It was found that continuous rating of melodies provides relevant information of dynamic musical processes. Dynamic models supplement the existing models of melodic expectations by extending their context sensitivity. The limitations and advantages of the continuous rating methodology and dynamic models are discussed.
Keywords. predictability, implication-realization, tonality, continuous, dynamic