Melody in Poems and Songs: Fundamental Statistical Properties Predict Aesthetic Evaluation

Abstract

Since antiquity, the concept of melody has been related not only to music, but also to language, specifically poetic language. However, while melodic properties of music were already mathematically defined more than 2,000 years ago, melodic properties of poems have remained fairly elusive to date. Proceeding from the assumption that fundamental melodic properties should in both domains be measurable in terms of recurrent time series of pitch and duration values, we administered two statistical measures—(a) autocorrelation coefficients and (b) spectral exponents—that both allow to quantify the relevant degrees of recurrence for recited poems and their sung versions alike. In a naturalistic concert event, 11 poems were recited, their musical settings sung, and aesthetically evaluative as well as emotion ratings collected from the audience. Correlating the autocorrelation coefficients and the spectral exponents with the subjective ratings revealed that the two acoustic measures chosen are predictive, to different strengths and with regard to different aspects, of the subjectively perceived melodiousness for both songs and poems. This approach to melody in songs and poems may also open new venues for empirically comparing melodic properties of speech and music beyond poetry.