Collocations in music? What systematic musicology can learn from corpus linguistics
Author: David R. W. Sears, College of Visual & Performing Arts, Texas Tech University, USA
Monday 12:00, March 11, 2019
NLP lab, room B203
Like language, much of the world’s music exhibits certain design features—namely, recurrence, syntax, and recursion—that both exploit and reflect the psychological mechanisms by which listeners organize sensory stimuli (Fitch 2006). As a result, allusions to principles of linguistic organization abound in music research (e.g., Lerdahl & Jackendoff 1983; Rohrmeier 2011). Yet despite recent strides by the linguistics community to discover potentially analogous organizational principles of natural languages using data-driven methods, applications of statistical modeling procedures have yet to gain sufficient traction in music research.
To resolve this issue, this paper considers how string-based methods for the discovery of collocations in natural language corpora might generalize to recurrent chord progressions in symbolic music corpora. To that end, I present a modeling pipeline that (1) selects an appropriate representation scheme for the symbolic encoding of chords; (2) applies the skip-gram method to identify 3- and 4-gram types consisting of potentially non-contiguous members; (3) excludes types reflecting “parts of music” (POM) that are rarely associated with interesting musical expressions (Manning & Schutze, 1999); and (4) calculates contingency tables and extended association measures that rank each type according to the statistical attraction between its members (Kilgarriff et al., 2012; Petrovic et al., 2010). In short, this pipeline produces convincing n-best lists for the discovery of meaningful harmonic progressions, though evaluating these lists using annotated corpora has yet to be conducted in the musicology community. I conclude by discussing possible limitations and future directions associated with the language metaphor in systematic musicology.
David Sears is Assistant Professor of Interdisciplinary Arts and Director of the Performing Arts Research Lab at Texas Tech University. Upon completing his PhD in music theory at McGill University in Montreal, Canada, he held a post-doctoral research position in the Institute of Computational Perception at Johannes Kepler University in Austria. His research interests include music perception and cognition, computational approaches to music theory and analysis, emotion and psychophysiology, and sensorimotor synchronization. Recent publications have appeared in the Quarterly Journal of Experimental Psychology, Music Perception, the Psychology of Music, the Journal of New Music Research, and the International Journal of Psychophysiology.