Emotion Painting: Lyric, Affect, and Musical Relationships in a Large Lead-Sheet Corpus

Abstract

How are lyrical emotions expressed in music? This paper explores the correlation between affect-carrying lyrics and musical features such as beat strength, duration, pitch height, consonance, and mode. Using computer-aided musicology software music21 and the NRC emotion lexicon, we conduct a corpus study on 1,895 folk and popular song lead-sheets encoded as MusicXML. The study reveals that metrical strength and note lengths are highly correlated with affects, while correlations of pitch height, consonance, and mode are in general less significant, at times contradicting previous research. Measurements of minor vs. major chordal context and tonal certainty, however, reveal certain previously unknown differences among emotional states. The paper uses a larger dataset of observations and gives greater values of significance than has appeared in symbolic corpus analysis of emotions in the past, and includes general discussions and directions for future work.