Corpus-based learning of tonal expectations with expectation networks

Abstract

ABSTRACTAlthough tonal expectations are an essential element of musical meaning in Western art music, it remains unclear whether the distinct melodic expectations associated with distinct scale degrees can be derived directly from musical corpora. Drawing on an analogy to neural learning, the current study outlines expectation networks, a novel and computationally simple method of measuring the expectation of encountering a given pitch followed in relatively close temporal proximity by another given pitch under a particular tonal system. Comparing expectation networks with listener ratings of asymmetrical pitch similarity and goodness of continuation between pitches reveals a large and highly significant effect, pointing to overlap between listeners’ judgments of tonal distance and corpus-derived expectations as to which pitches tend to be grouped close together in time.