Psychological constraints on string-based methods for pattern discovery in polyphonic corpora
This work addresses a gap in music information retrieval by applying psychological insights to harmonic analysis, though it is incremental as it builds on existing n-gram methods.
The study investigated whether psychologically-motivated weighting functions could improve harmonic pattern discovery algorithms in polyphonic music corpora, finding that these functions enhanced the identification of conventional chord progressions, with specific models achieving improved mean reciprocal ranks.
Researchers often divide symbolic music corpora into contiguous sequences of n events (called n-grams) for the purposes of pattern discovery, key finding, classification, and prediction. What is more, several studies have reported improved task performance when using psychologically motivated weighting functions, which adjust the count to privilege n-grams featuring more salient or memorable events (e.g., Krumhansl, 1990). However, these functions have yet to appear in harmonic pattern discovery algorithms, which attempt to discover the most recurrent chord progressions in complex polyphonic corpora. This study examines whether psychologically-motivated weighting functions can improve harmonic pattern discovery algorithms. Models using various n-gram selection methods, weighting functions, and ranking algorithms attempt to discover the most conventional closing harmonic progression in the common-practice period, ii6-"I64"-V7-I, with the progression's mean reciprocal rank serving as an evaluation metric for model comparison.