Linked auditory and motor patterns in the improvisation vocabulary of an artist-level jazz pianist.

Improvising musicians possess a stored library of musical patterns forming the basis for their improvisations. According to a prominent theoretical framework by Pressing (1988), this library includes linked auditory and motor information. Though examples of libraries of melodic patterns have been shown in extant recordings by some improvising musicians, the underlying motor component has not been experimentally investigated nor related to its auditory counterparts. Here we analyzed a large corpus of ∼100,000 notes from improvisations by one artist-level jazz pianist recorded during 11 live performances with audience. We compared the library identified from these recordings to a control corpus consisting of improvisations by 24 different advanced jazz pianists. In addition to pitch, our recordings included accurate micro-timing and key velocity (i.e., force) data. Following a previously validated procedure, this information was used to identify the underlying motor patterns through correlations between relative timing and velocity between notes in different iterations of the same pitch pattern. A computational model was, furthermore, used to estimate the information content and generated entropy exhibited by recurring pitch patterns with high and low timing and velocity correlations as perceived by a stylistically enculturated expert listener. Though both corpora contained a large number of recurring patterns, the single-player corpus showed stronger evidence that pitch patterns were linked to motor programs in that within-pattern timing and velocity correlations were significantly higher compared to the control corpus. Even when controlling for potentially greater baseline levels of motor self-consistency in the single-player corpus, this effect remained significant for velocity correlations. Amongst recurring 5-tone pitch patterns, those exhibiting more consistent motor schema also used less idiomatic pitch transitions that were both more unexpected and generated more uncertain expectations in enculturated experts than less consistently repeated patterns. Interestingly, we only found partial evidence for fixed pattern boundaries as predicted by the Pressing model and therefore suggest an expanded view in which the beginning and ends of idiomatic audio-motor patterns are not always clear-cut. Our results indicate that the library of melodic patterns may be idiosyncratic to the individual improviser and relies both on motor programming and predictive processing to promote stylistic distinctiveness.