A Supervised Learning Approach to the Behavior of 'Soul Dominants' in the McGill Billboard Corpus

Stanley Ralph Fink, Drake University

Spicer's "soul dominant" (2004, 2017) poses challenges to harmonic theories of popular music. This paper tests some prediction models designed to improve categorical outcome measures regarding whether a chord of the "soul dominant" quality will resolve to "tonic" (if only locally). I created a dataset (n = 1740) of soul dominants across 124 songs in the McGill Billboard Corpus transcriptions (Burgoyne 2011), classifying the chords by their bass scale degree within the tonal context given by the transcription. For each chord, I recorded the root and bass of the previous and following chords, and, where possible, the beat-class (Cohn 1992) of the chord's onset and offset within normative four-measure units of quadruple meter. This paper finds that soul dominants are more likely to resolve directly to tonic when preceded by bass motion of a P4 or P5; or, when the chord transition happens on a relatively strong beat.