Modelling improvisatory and compositional processes

James Kippen & Bernard Bel

In Denise Penrose & Ray Lauzanna (eds.) Languages of Design, 1 (1992). Elsevier Science Publishers, Amsterdam: 11-26.

Abstract

An appli­ca­tion of for­mal lan­guages to the rep­re­sen­ta­tion of musi­cal process­es is intro­duced. Initial inter­est was the struc­ture of impro­vi­sa­tion in North Indian tabla drum music, for which exper­i­ments have been con­duct­ed in the field as far back as 1983 with an expert sys­tem called the Bol Processor, BP1. The com­put­er was used to gen­er­ate and ana­lyze drum­ming pat­terns rep­re­sent­ed as strings of ono­matope­ic syl­la­bles, bols, by manip­u­lat­ing for­mal gram­mars. Material was then sub­mit­ted to musi­cians who assessed its accu­ra­cy and increas­ing­ly more elab­o­rate and sophis­ti­cat­ed rule bases emerged to rep­re­sent the musi­cal idiom.

Since sev­er­al method­olog­i­cal pit­falls were encoun­tered in trans­fer­ring knowl­edge from musi­cian to machine, a new device, named QAVAID, was designed with the capa­bil­i­ty of learn­ing from a sam­ple set of impro­vised vari­a­tions sup­plied by a musi­cian. A new ver­sion of Bol Processor, BP2, has been imple­ment­ed in a MIDI stu­dio envi­ron­ment to serve as a aid to rule-based com­po­si­tion in con­tem­po­rary music. Extensions of the syn­tac­tic mod­el, such as sub­sti­tu­tions, metavari­ables, and remote con­texts, are briefly introduced.

Excerpts of an AI review of this paper (Academia, June 2025)

Summary of the Work

The man­u­script presents a for­mal language-based approach to mod­el­ing impro­visato­ry and com­po­si­tion­al process­es in music — par­tic­u­lar­ly North Indian tabla drum­ming — and high­lights the devel­op­ment of the Bol Processor soft­ware. The arti­cle effec­tive­ly demon­strates how gen­er­a­tive gram­mars, pat­tern lan­guages, and relat­ed exten­sions (e.g., pat­tern rules, homo­mor­phisms, neg­a­tive con­texts, and remote con­texts) can cap­ture the com­plex­i­ty and flex­i­bil­i­ty of musi­cal form, espe­cial­ly with regard to impro­vi­sa­tion. The authors also illus­trate the tran­si­tion from an ini­tial musi­co­log­i­cal focus in the ear­li­er Bol Processor (BP1) to the more gen­er­al­ized and com­po­si­tion­al ori­en­ta­tion of BP2, high­light­ing the system’s role in computer-assisted com­po­si­tion. Throughout, there is a strong empha­sis on inte­grat­ing eth­no­mu­si­co­log­i­cal insights, com­pu­ta­tion­al meth­ods, and prac­ti­cal soft­ware design.

Strengths

Comprehensive Description of Formal Methods

The paper pro­vides a thor­ough expla­na­tion of how dif­fer­ent gram­mar types — finite-state automa­ta, context-free gram­mars, type-0 gram­mars — are rel­e­vant for describ­ing musi­cal forms. This is com­ple­ment­ed by exam­ples that clar­i­fy the the­o­ret­i­cal con­cepts (e.g., the use of pat­tern rules to han­dle repeat­ed sec­tions or “voiced/unvoiced” trans­for­ma­tions). These sec­tions effec­tive­ly detail both tra­di­tion­al and inno­v­a­tive mod­el expansions.

Contextual Application in Ethnomusicology

By apply­ing gen­er­a­tive mod­els to North Indian tabla’s “theme-and-variations” struc­tures (qa‘idas), the paper demon­strates how such com­pu­ta­tion­al approach­es can illu­mi­nate facets of an oral­ly trans­mit­ted musi­cal tra­di­tion. The depth of col­lab­o­ra­tion with expert musi­cians and the iter­a­tive process of eval­u­a­tion is indica­tive of a con­sci­en­tious ethno­graph­ic methodology.

Transition to a Generalized Composition Environment

BP2 goes beyond tra­di­tion­al eth­no­mu­si­co­log­i­cal appli­ca­tions and can be employed as a computer-aided com­po­si­tion tool for broad­er musi­cal con­texts. The con­cept of “sound-objects” and flex­i­ble time-handling approach­es (stri­at­ed vs. smooth time, time piv­ots, con­straint sat­is­fac­tion, etc.) make the work­flow appeal­ing to com­posers and researchers inter­est­ed in gen­er­a­tive music systems.

Balanced Presentation of Achievements and Challenges

The paper address­es prob­lems of knowl­edge trans­fer, data rep­re­sen­ta­tion, and the com­plex­i­ty of rule-based sys­tems, pro­vid­ing valu­able lessons for future com­pu­ta­tion­al musi­col­o­gy projects. It also describes how the QAVAID sub­sys­tem attempts to auto­mate knowl­edge acqui­si­tion and infer­ence, there­by remov­ing some of the bot­tle­necks encoun­tered when man­u­al­ly con­struct­ing grammars.

Extensive Reference to Prior Research

A sub­stan­tial and well-organized ref­er­ence sec­tion places the work in con­ver­sa­tion with rel­e­vant lit­er­a­ture from for­mal lan­guage the­o­ry, musi­co­log­i­cal stud­ies, and AI-based com­po­si­tion sys­tems. The inclu­sion of the­o­ret­i­cal sources (e.g., cita­tions to Chomsky’s hier­ar­chy) and prac­ti­cal ref­er­ences (e.g., man­u­al or share­ware avail­abil­i­ty) under­scores the paper’s aca­d­e­m­ic depth and real-world applicability.

Overall Appraisal

This study offers a detailed and inno­v­a­tive account of how for­mal gen­er­a­tive tools can be har­nessed to describe, ana­lyze, and pro­duce com­plex musi­cal struc­tures. By bridg­ing the­o­ret­i­cal com­put­er sci­ence, eth­no­mu­si­co­log­i­cal field­work, and com­po­si­tion­al prac­tice, the man­u­script opens pos­si­bil­i­ties for deep­er explo­ration of music’s syn­tac­tic fea­tures and cre­ative appli­ca­tions in con­tem­po­rary com­po­si­tion. Its empha­sis on prac­ti­cal soft­ware devel­op­ment, eval­u­at­ed in tan­dem with expert cri­tiques, posi­tions it as a note­wor­thy resource for researchers and prac­ti­tion­ers at the inter­sec­tion of musi­col­o­gy, AI, and dig­i­tal arts.

Download this paper