Author: Bernard Bel

Bernard Bel

Knowledge acquisition and representation in music

Thèse de doctorat en sciences. Université de droit, d'économie et des sciences - Aix-Marseille III
https://theses.hal.science/tel-00009692

Résumé

Cette étude traite de la représentation informatique de connaissances en musique, abordée à partir de deux expériences en grandeur réelle. La première est une méthode d'acquisition de connaissances en ethnographie mettant en interaction un expert (le musicien), un analyste (le musicologue) et une machine dans une situation d'apprentissage. Les schémas d'improvisation des musiciens sont identifiés et exprimés à l'aide de règles de production dans un formalisme dérivé des grammaires génératives et des langages de formes. Un algorithme déterministe de test d'appartenance de chaînes arbitraires au langage défini par une grammaire (sensible au contexte) est présenté, ainsi qu'une technique d'inférence inductive de langages réguliers permettant l'acquisition automatique de connaissances lexicales et syntaxiques. La seconde expérience s'insère dans l'élaboration d'un environnement de composition musicale assistée par ordinateur. Le problème est ici la représentation du temps dans une structure discrète d'“objets temporels”, et plus généralement la synchronisation de processus parallèles. Une méthode est proposée pour la détermination d'une structure à partir de données incomplètes sur la synchronisation des objets. La notion d'“objet sonore” est ensuite explicitée formellement. Un algorithme efficace permet l'instanciation des objets sonores affectés à une structure en tenant compte des contraintes liées à leurs propriétés métriques et topologiques.

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

Summary

This 1990 doctoral thesis presents an innovative approach to computer representation of musical knowledge through two major experimental frameworks. The work bridges artificial intelligence, formal language theory, and musicology, addressing fundamental problems in knowledge acquisition and time representation in musical structures.

First Experiment: Knowledge Acquisition in Ethnography

The first part introduces a novel methodology for acquiring musical knowledge through a "dialectical anthropology" approach involving three actors: an expert musician, a musicologist-analyst, and a computer system. This methodology was applied to studying improvisation patterns (qa'ida) of North Indian tabla percussion music.

Key Contributions

• Development of BP (Bol Processor) grammars, an extension of formal grammars capable of representing repetition patterns and homomorphic transformations
• A deterministic membership test algorithm for a subclass of context-sensitive grammars
• An inductive inference technique for regular languages that simultaneously acquires lexical and syntactic knowledge
• Introduction of negative context rules and stochastic control mechanisms

The system enables musicians to validate machine-generated musical phrases, creating a feedback loop that refines grammatical models of improvisation schemas. The approach moves beyond traditional ethnographic collection by making the computer an active partner in knowledge acquisition.

Second Experiment: Computer-Assisted Musical Composition

The second part addresses time representation and synchronization in discrete musical structures, developing the BP2 environment for musical composition.

Key Innovations

• Introduction of "symbolic time" as distinct from physical time
• Development of polymetric formulas for representing parallel musical sequences
• Formal treatment of "time objects" and "out-time objects" (temporal vs. atemporal)
• An efficient algorithm for synchronizing sequences with incomplete temporal information
• A constraint satisfaction approach to "sound object" instantiation considering metric and topological properties

Theoretical Framework

The work establishes connections between musical structures and formal language theory, introducing concepts such as:

• Pattern grammars and restricted pattern languages (RPL)
• Transformational grammars adapted to musical contexts
• Event universes structured by simultaneity, precedence, and sequentiality relations
• Trace theory applications to polymetric structures

Strengths

• Methodological Innovation: The dialectical anthropology approach represents a significant advancement in ethnomusicological methodology. By placing the computer as an active participant rather than a passive tool, Bel creates a genuinely interactive knowledge acquisition system that respects both the expertise of traditional musicians and the rigor of formal methods.
• Theoretical Rigor: The mathematical formalization is sophisticated yet practical. The extension of formal grammars to handle musical repetition patterns and homomorphic transformations addresses real needs in musical representation that existing computational linguistics tools couldn't meet.
• Cross-Cultural Sensitivity: The work demonstrates genuine respect for non-Western musical traditions while avoiding ethnocentric biases common in computational musicology. The choice to work with North Indian tabla music, with its oral transmission tradition, was particularly appropriate for testing knowledge acquisition methodologies.
• Practical Implementation: Both BP1 and BP2 systems were actually implemented and tested, demonstrating the feasibility of the theoretical proposals. The complexity analysis (showing polynomial-time algorithms) indicates practical applicability.
• Time Representation Innovation: The distinction between symbolic and physical time, along with the concept of "time structure," provides a flexible framework that can accommodate different musical temporalities beyond Western metronomic time.

Significance and Impact

This work was pioneering in several respects:

• Methodology: The interactive, computer-mediated approach to ethnographic knowledge acquisition anticipated later developments in digital humanities and computational ethnomusicology.
• Formal Methods: The extension of formal language theory to musical applications provided tools that were later adopted and extended by other researchers in computational musicology.
• Cultural Perspective: The work demonstrated how computational methods could be applied respectfully to non-Western musical traditions, opening paths for more inclusive computational musicology.
• Temporal Modeling: The sophisticated treatment of musical time influenced subsequent work in computer music and musical information retrieval.

Contemporary Relevance

Thirty years later, this work remains relevant to several current research areas:

• Machine learning applications to music (the inductive inference techniques anticipate modern approaches)
• Digital humanities methodologies
• Music information retrieval systems
• Computer-assisted composition environments
• Cross-cultural computational musicology

The emphasis on expert-machine collaboration prefigures current interest in human-AI collaboration in creative domains.

Conclusion

Bel's thesis represents a significant contribution to computational musicology, combining theoretical innovation with practical implementation and cultural sensitivity. While some limitations exist, particularly regarding scope and evaluation, the work established important foundations for computer-mediated musical knowledge representation. The dialectical methodology for knowledge acquisition and the sophisticated treatment of musical time remain valuable contributions to the field. The work demonstrates how formal computational methods can be applied to musical knowledge while respecting the complexity and cultural specificity of musical traditions.

Download this thesis

Bernard Bel

Computer Music Journal (1998), Vol. 22, 2: 56-64

Abstract

The Bol Processor is the outcome of a migratory process, its design having been carried over in three phases and places: in collaboration with traditional North Indian musicians (1980-85), Western musicians in Europe (1985-93) and back in India with Carnatic musicians (1995-97).The theoretical framework of the underlying research project also evolved in three stages, taking inspiration from expert systems in the early 1980s, symbolic-numeric machine-learning in the end of the decade, and composition theory in the 1990s.Throughout this process, the designer has been faced with the challenge of blending software with "mindware", here taken to mean musicians' striving for "[…] tools enabling them to manipulate objects so as to imbue them with 'soul' or experiential value […]" (Laske 1996). In a cross-cultural approach this led to modelling descriptions of music and compositional processes at a level of abstraction sufficiently high to encompass "local" musical concepts without getting too abstruse.

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

Summary

Bernard Bel’s manuscript provides an extensive overview of the Bol Processor (BP), a music software environment that evolved through collaborations with Indian and Western musicians. The software addresses rhythmic, melodic, and compositional tasks using a grammars-based approach, symbolic time notation, and sophisticated algorithms (e.g., polymetric expansion, quantization). The review traces the history and theoretical underpinnings of this system, moving through the project’s three phases of development and culminating in a flexible interface linking BP to Csound. By emphasizing text-based data representations, BP aims to foster deeper thinking about music structure and compositional processes, accommodating contexts that fall outside purely Western traditions.

Strengths

Historical and Contextual Depth: The author clearly articulates how the project emerged from ethnomusicological research in Indian music, then expanded to address Western compositional techniques. This historical framing highlights how the software bridges cultural contexts while retaining a consistently powerful grammar-based approach.

Clear Explanations of Core Algorithms: The detailed descriptions of polymetric expansion, time-stretching, quantization, and grammar-based transformations are a valuable technical resource. The author provides examples, diagrams, and references that elucidate these concepts for readers unfamiliar with formal language methods applied to music.

Innovative Methodology: The software design seamlessly integrates symbolic approaches (formal grammars, integer-based time units, constraint satisfaction) with physical execution in MIDI and Csound domains. This duality allows composers and researchers to treat time both discretely and continuously as required.

Attention to Cross-Cultural Music Needs: BP’s adaptability to various musical notations, including Indian and Western pitch/note representations, is a strong signature of this work, reflecting deep respect for non-Western musical concepts and performance practices.

Comprehensive References and Examples: The supporting references shed light on related research in composition theory, music cognition, and computational models. Numerous code-like illustrations and figures demonstrate how actual musical tasks are implemented, assisting practitioners who might want to replicate or modify these ideas.

Overall Impressions

The manuscript provides an in-depth look at a uniquely versatile system for musical analysis and composition. The author’s commitment to text-based interaction, coupled with robust support for real-time operations and cross-cultural music concepts, stands out as a mature, flexible framework. The paper offers a thoughtful presentation of both underlying theoretical foundations and practical implementations, making it relevant for those interested in computer-aided composition, musicological research, and interdisciplinary approaches bridging Indian and Western musical forms.

Download this paper

James Kippen & Bernard Bel

Anthropological Quarterly (1989), 62, 3: 131-144

Abstract

A major problem in ethnographic description may be summed up as the search for ways to disentangle folk from analytical models. Knowledge-based systems have contributed to the development of formal structures for the manipulation of symbols associated with particular physical and conceptual phenomena. Importantly, their output can be interpreted by experts in the domain. This provides evaluation procedures for models elaborated jointly by analysts and informants. This paper describes an interactive system in which knowledge is represented as production rules in a format derived from the theory of formal languages. Modus ponens and modus tollens are explained and compared to derivation schemata in first-order predicate logic. The results of an application to the study of North Indian tabla drumming are assessed. We conclude that (1) knowledge represented at a low theoretical level fails to descriminate between the input from informants and the intuitive assumptions of analysts, (2) experimental procedures can be improved considerably if the system is designed to perform automated knowledge acquisition (using probabilistic grammars and inductive learning).

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

Summary of the Work

The article focuses on the use of computational approaches — particularly knowledge-based systems — to address the challenge of producing consistent and systematic ethnographic descriptions. The authors draw on their research with the Bol Processor (BP), a formal language-based system originally developed to represent and generate sequences of verbal drum syllables (bols) in North Indian tabla performance. They describe how the BP’s grammar formalism, inference engine, and membership tests facilitate iterative interaction between an analyst and informants (expert musicians). By demonstrating the system’s ability to produce, recognize, and evaluate permissible melodic or rhythmic variations, the authors aim to illuminate how computers can be more deeply integrated into anthropological and ethnomusicological research.

Key Contributions

Formal Grammar and Pattern Representation: The paper offers a clear explanation of how context-free and context-sensitive grammars can be used to specify culturally valid musical structures. In particular, the authors highlight the importance of representing patterns and constraints on musical improvisation in a generative format.

Interactive Methodology: A central point is the iterative feedback loop that allows informants to assess computer-generated sequences and provide corrections. This “apprentice-like” interaction underscores how computational tools can help researchers refine theoretical models of cultural knowledge in near-real-time.

Probabilistic Grammars: The authors incorporate a weighting system for production rules, thereby accounting for the relative likelihood of different musical derivations. This approach not only adds realism to the generative process but also tackles common issues with purely enumerative or random output.

Membership Testing: The membership test, which decides whether a newly proposed sequence belongs to a given grammar, is described as an efficient, deterministic bottom-up parser. This feature is significant because it allows informants to offer novel variations while enabling the system to quickly judge their adherence to the emergent rule set.

Reflections on Folk vs. Analytical Models: The article raises important anthropological questions about the boundary between informants’ internalized (folk) knowledge and the analyst’s formal (re)construction. The authors’ frank discussion of the challenges involved, such as the risk of overformalizing or “anticipating” informant knowledge, is methodologically relevant.

Strengths

• The paper emphasizes the value of combining fieldwork with computational experimentation. This dual approach—grounded in real ethnographic and musical practice—provides a compelling demonstration of how artificial intelligence techniques can advance the study of music and culture.
• Clear examples illustrate code structures, parsing mechanics, and the grammar design. These concrete details will help other researchers adapt or replicate the Bol Processor approach in different cultural or musical contexts.
• The account of iterative knowledge acquisition highlights a collaborative research dynamic, showing how insight flows back and forth between informants and computational models, rather than being a unidirectional exercise of extracting information.

Conclusion

This study offers a thoughtful and innovative approach to bridging ethnographic inquiry with computational models. By showing how arrangements of symbolic musical data can be systematically generated, tested, and refined in conjunction with human experts, the authors illustrate a novel and promising method for ethnographic description. The paper encourages a nuanced understanding of both the potential and limitations of using expert systems to capture the complexity of cultural knowledge.

Download this paper

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

James Kippen & Bernard Bel

In Mira Balaban, Otto Laske et Kemal Ebcioglu (eds.) Understanding Music with AI, American Association for Artificial Intelligence Press, Menlo Park CA (1992): 366-400.

Abstract

Bol Processor grammars are an extension of unrestricted generative grammars allowing a simple representation of string "patterns", here taken to mean repetitions and homomorphic transformations. These have been successfully applied to the simulation of improvisatory techniques in traditional drum music, using a production-rule system called "Bol Processor BP1". The basic concepts and parsing techniques of BP1 are presented.

A new version of Bol Processor, namely "BP2", has been designed to serve as a aid to rule-based composition in contemporary music. Extensions of the syntactic model, such as metavariables, remote contexts, substitutions and programmed grammars, are briefly introduced.

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

Overview and Summary

The authors propose a formalism, called "Bol Processor grammars," designed to capture and simulate performance and improvisatory behaviors in traditional drum music—particularly North Indian tabla. This work presents a detailed account of how their proposed grammar-based system (BP1 and subsequently BP2) manages musical elements such as generative rules, parsing procedures, and higher-order transformations. The authors draw on concepts from formal language theory, specifically incorporating string rewriting, generative grammars of varying types (from context-free to unrestricted), and pattern languages.

The monograph not only discusses theoretical frameworks but also provides implementations and examples relevant to composition, improvisation, and ethnomusicological analysis. By combining standard grammars with additional features (e.g., pattern rules, negative contexts, remote contexts, substitutions, homomorphisms, and a sophisticated weighting mechanism), the Bol Processor aims to model creative aspects of improvisatory traditions.

Contribution and Significance

• The paper bridges theoretical computer science (rewriting systems, generative grammars) with ethnomusicological inquiry. This interdisciplinary approach shows how language models can adapt to musical performance traditions, especially where oral transmission prevails.
• The authors introduce extensions to classical Chomsky hierarchies by incorporating string pattern languages and homomorphisms specifically tailored for music composition and analysis. This advancement is especially valuable to those researching computational musicology or algorithmic composition.
• By providing practical implementation details and guidelines (e.g., subgrammars, weighting rules, context-sensitive substitutions), the study conveys a clear path for others looking to model or simulate improvisational processes.

Strengths

• Clarity of Theoretical Underpinnings: The text carefully explains the fundamentals of generative grammars and pattern languages, ensuring that readers unfamiliar with formal language theory can still follow the rationale behind the Bol Processor model.
• Comprehensive Examples: The inclusion of worked-through grammar listings, detailed parsing traces, and real-world musical segments highlights an applied perspective. Readers can see exactly how the rules operate on concrete musical materials.
• Interdisciplinary Integration: The manuscript thoughtfully weaves together computational linguistics, ethnomusicology, and composition, offering a unique perspective to each discipline.
• Generative and Analytical Capacities: Emphasizing both the generation of new musical variations and the parsing of existing performances demonstrates the system’s two-fold utility: it supports creative composition while providing a framework for empirical analysis.

Areas for Further Development

• Handling of Larger-Scale Form: While the paper addresses theme-variation structures, the methodology could be extended to more extensive global forms or multi-sectional pieces. More elaboration on how the grammar might manage nested forms or very long structures would strengthen the approach.
• Quantitative Evaluation: The text provides evidence of successful modeling but could benefit from additional discussion of how correctness or “musical plausibility” is systematically tested, beyond anecdotal or interactive sessions with experts.
• Comparisons with Other Systems: A more in-depth comparison with existing computational music systems that also employ generative grammars (e.g., purely context-free or Markov-based approaches) might deepen an understanding of the Bol Processor’s unique contributions.

Readability and Presentation

• The writing is clear and consistently structured, especially around the discrete sections (introduction, pattern rules, parsing, and advanced features in BP2). Diagrams and grammar listings are helpful, though further clarifying annotations in some figures could assist readers less familiar with formal notation.
• The appendices excel in presenting extended examples and step-by-step parses, adding transparency to the methods. This style of presentation ensures reproducibility and offers insights into how to adapt or modify the system for other musical styles.

Potential Impact on the Field

The system has apparent implications not just for tabla and other percussion traditions but for any domain where complex variations can be expressed in a rule-based manner. Likewise, composers working with algorithmic or computer-aided composition may discover a robust set of techniques for shaping variation, texture, and form. Researchers in ethnomusicology might find new analytical tools for uncovering systematic elements in improvisational practices.

Overall, the work stands as a thorough exploration of generative and parsing approaches tailored to music, illustrating how computational models can deepen understanding of both fixed and improvised musical structures.

Download this paper