Output details

Interrogating melodic similarity : a definitive phenomenon or the product of interpretation?

Melodic similarity is a key concept both for musicology and for contemporary media science. On the one hand, analyses of musical structure and accounts of musical meaning often depend on the identification of similarities and differences between melodies, while on the other searching in large databases of music depends on a modelling of similarity in order for inexact matches to be retrieved. Underlying most use of the concept is the unsound assumption that because two pieces of melody sound similar, that similarity must be discoverable in the two sequences of notes, whereas in fact many other factors of context come into play. This article examines the nature of claims for melodic similarity in greater depth than other published material, drawing on previous experimental data in a novel fashion. Monte Carlo modelling, a rigorous method of statistical hypothesis testing; is used for the first time on musical-similarity data, investigating the assumptions underlying previous interpretations of this data. On the basis of this and other methods of scrutinising the concept, a reorientation of research on melodic similarity is proposed. The article arises from a conference presentation at the Sound and Music Computing conference, University of Padua, 2011, and was published in a special issue of Journal of New Music Research for which all editorial and review procedures were handled by the guest editors, Federico Avanzini and Giovanni de Poli (University of Padua) and Davide Rocchesso (University of Venice).