Bernard Bel <[email protected]>
In TABOR, Jerry (ed.) Otto Laske: Navigating Musical Horizons. (Contributions to the Study of Music and Dance; 53) Westport, Conn. : Greenwood Press. 1999, p. 81-91
C'est un extra-terrestre!
This was my wife's exclamation the first day she saw Otto Laske. After years of communication through the Internet, our short encounter in 1990 had turned to a kind of recognition exercise... gathering reminiscences of a common world. Ours was indeed the virtual one of "global culture", but we soon discovered a shared suspicion towards ethnocentrism, cultural prejudice and illusory speech knowledge.
Otto Laske was born in Olesnica (Silesia) and graduated in philosophy under Theodore Adorno in Frankfurt-am-Main (Germany). In the late 1960's he went to Boston (USA) to complete his training as a composer. Later he studied and worked in the USA, Canada, the Netherlands, Finland... He speaks fluent French and writes English poetry.
The man from another planet was bound to become the target of controversial debates often arising from single-sided views on his work. I find it very difficult to introduce such a multi-faceted intellectual démarche, ranging from philosophy, psychology, sociology, computer science, leave alone the practice of poetry and music. A conventional academic survey of Laske's writings -- focusing on products to the detriment of processes -- might betray his commitment to action science. This paper is therefore an attempt to trace the premises of his theoretical work in computational musicology. Being a subjective account, it may be taken as a reconstruction of Laske's favourite themes in the light of experience in the related field of music software engineering.
Early writings of Otto Laske should be reviewed in the context of the early 1970's, when uncompromising ideas about competence and performance were dominating the field of linguistics (Chomsky & Halle 1968). Unlike musicologists applying generative models to music, Laske did not attempt to formulate a model of musical competence (an intrinsic, tacit knowledge distillable into a grammar, 1992c:210) opposed to performance...
Performance, giving rise to individual compositions, and thus requiring, as its theoretical analog, a musical poetics, deals with musical activities as deviations from grammaticalness which result from mental and psychological factors external to the grammar, such as finiteness of memory, aesthetic intention, and pursuit of personal strategies. (op.cit.:214)
The musical notion of performance, in Laske's view, did not coincide with the Chomskian concept recalled here. In a musicological sense, it was that part of the musician's knowledge which pertains to his use of competence and of local knowledge under real-world conditions (Laske 1992a:10).
Competence itself needed to be reviewed in the light of an action theory. For, the idea of musical competence "out there" providing a logical explanation of musical activity (i.e. amenable to performance prediction, Laske 1992a:10) was confronted with several open questions:
In music, then, who was the native speaker? The composer, listener, instrumentalist, conductor, music-analyst --who? What did the tasks these different experts were pursuing have to do with each other, if anything? (op.cit.:217)
Laske anticipated an important conceptual change which was later enacted by (ethno)linguists gaining insights into the historical dimensions of languages. Studies on pidgin languages, notably, suggest that linguistic competence (here meaning domain-specific knowledge) may be "constructed" from performance. Constructed, indeed, and altogether maintained, since human languages are subject to growth, decline and even death (Hagège 1996:35). This is also the case with aurally transmitted musical forms which do not rely on permanent knowledge substrates. If so, envisaging musical competence as a cumulative process, rather than a rigid assemblage of universal and local axioms, may provide a new explanatory framework accounting for change in the so-called "immovable" traditions.
Besides Chomsky's generative linguistic model, Laske was influenced by the work of Michael Kassler (1963). However, he found limitations in Kassler's attempt to derive a theory of music systems from individual pieces of written music. At first, the theory seemed to deny the importance of problem solving: the procedural basis of any theory of music as something humans do, rather than just understand (Laske 1992c:211). In addition, Kassler's analytic approach was based on example-based composition, a theory of "existing musics" bypassing the problem of musicality, the domain of "possible musics":
In retrospect, [Kassler's] work is limited by the same assumptions that underly contemporary work in expert systems. Frequently, the resulting knowledge bases embody only those elements of an expert's knowledge that are pure competence, and public knowledge at that, leaving out the expert's performance in real time, and his idiosyncratic local knowledge of how his performance fits into a particular task environment. (Laske 1992c:211)
Unexpectedly, the issue of musicality raised by Otto Laske is of great relevance outside the western musical world it originated from. When dealing with improvisatory techniques in aurally transmitted musical systems, ethnomusicologists tend to minimise, if not refute, compositional (i.e. problem solving) processes. In the West, the status of a composer may only be granted to musicians claiming musical competence (assessed by experts and certified by academic records). In the broad view of social-cultural anthropology, this idea may be challenged as laden with eurocentric prejudice (Blacking 1989). It is also interesting to deal with musical creativity from a cognitive perspective, with the hypothesis that music improvisation might be tractable as "real-time composition".
During the early 1980's, Jim Kippen and I conducted field work in India, aiming at the development of a formal-language representation for tabla music and its implementation in a computer system named Bol Processor (Kippen & Bel 1989a). The methodological approach was one in which the computer could be incorporated into the process of data collection, representation and analysis, in order to facilitate an "apprentice-like" interaction between informants and analysts where priority may be given to the modelling of informants' and not analysts' views (ibid.). The Bol Processor was a sort of expert system in which a preliminary set of syntactic pattern rules (an extension of Chomsky grammars) was proposed by the analyst on the basis of hypotheses on compositional strategies. In a related experiment (Kippen & Bel 1989b), the initial model could be infered automatically from ordered sets of examples. The machine was then requested to derive sentences -- strings of terminal symbols, which were played or recited by the analyst so allowing expert musicians to assess their quality and accuracy. In addition, to prove that the current model was representative of the full scope of the compositional scheme under investigation, it was necessary to invoke a proof procedure (membership test) for any sentence proposed by the informant as a correct variation. Thus, grammars were not aimed at describing a particular musical item, nor a finite set of musical items (a corpus), but a composition scheme (named qaida in Hindi/Urdu) yielding a virtually infinite set of acceptable compositions.
The interaction between the expert, the machine, and the analyst in Bol
Processor experiments may be summarised in the following flowchart which Laske
(1990a:285) later extended to the western context:
Informants' analytical observations were thus incorporated into the model in order to correct the machine's inadequacies and to help in the formulation of increasingly valid hypotheses of musical structure.
Our work came in support of the view that this type of improvisation qualifies as (indeed, real-time) rule-based composition. This may apply to other improvisational practices where grammaticality (consistent decisions about well-formedness) is relevant, exemplifying cultural knowledge that is quantified (through encoding), non-hierarchical, reasonably coherent, consistent, and above all bounded (Kippen & Bel 1989a).
Inaccurate investigations of real-time cognitive processes mislead musicologists who reduce improvisation to a virtuoso combination of pre-composed fragments. This amounts to a conceptualisation of the underlying process as example-based rather than rule-based. Pre-composed fragments and "precompiled" strategic knowledge certainly play a role in tabla improvisation. But an exclusive example-based approach presupposes the existence of "models" (examples) and calls for a methodology in which these may be derived from existing musical pieces: analytical grammars illustrating (instead of explicating) music competence. As put by Laske (1992c:211), this is analogous to trying to derive a French grammar from individual poems by Stéphane Mallarmé.
The Bol Processor experimental set-up for improvisational knowledge acquisition may be compared with Laske (1973b,1992b) and Truax's OBSERVER programs in which musicians (notably children between ages five and twelve) were given simple compositional tasks for which traces of their mental processes could be analysed. The comparison raises the issue of procedural (action oriented) versus declarative (description oriented) knowledge, which has been the topic of passionate debates in the artificial intelligence community.
A pattern grammar in the Bol Processor resembles a Chomsky grammar in its attempt to describe a (virtually infinite) set of sentences (a formal language) acceptable by an idealised expert. In a formal grammar, rules are applied in an arbitrary order. Thus, the specification of the problem "generate a sentence" is clearly separated from the method of solution. This is the typical declarative approach on which logic programming languages like Prolog have been based (Kowalski 1985:82).
Contrasting to this, action protocols traced by OBSERVER are stepwise procedural descriptions of the compositional process. There was also procedural knowledge in early Bol Processor grammars, as rules were partially ordered thanks to the intuitive layering of grammars (Kippen & Bel 1992). Besides, new features in the "BP2" version (used as a composition assistant) include "flags" denoting states, global numeric variables, conditional jumps and real-time interaction with the MIDI environment (Kippen & Bel 1994). This amounts to using programmed grammars (Laske 1992c:224), i.e. introducing strategical or performance constraints monitoring derivations in the context of a specific real-time problem-solving task. Unlike abstract generative grammars, programmed grammars qualify as an integral component of a musical agent or robot functioning in a specific task environment (ibid.).
Current developments of knowledge models, notably object-oriented systems, point at a comprehensive encapsulation of combined declarative and procedural representations.
Modelling human activity presupposes a trade between generality, relevance and accuracy. OBSERVER protocols are extremely relevant and accurate, but they do not easily lend themselves to generalisation (inductive inference). On the other hand, formal grammars are accurate and easy to manipulate inductively (see Kippen & Bel 1989b), but their relevance to musical activities such as composition and listening may be questioned.
It may be clear from the above discussion that computational modelling attempts to elude pseudo-problems caused by natural-language aesthetic discourse, thereby challenging illusions fabricated by verbal knowledge. Uncritical reliance on speech definitions of artistic creativity, notably, undermines debates on "modernity" and "tradition" biased by culture-specific perceptions of change (Lath 1988, Zarrilli 1987)... Does artificial intelligence yield practical insights into such an interdisciplinary problem?
If you take a practitioner of some domain, your theory of knowledge in the domain, and attempt to embody a part of that knowledge in a computer program destined for interactive use by the practitioner, you have already decided that creativity is discipline-based. This may do you a lot of good in research regarding scientific creativity, but it does not bode well for studies, say, in poetic or musical creativity, where creativity may be unique or idiosyncratic before it becomes discipline-based (if it ever totally does), not to mention the fact that the competence in question may be largely "action knowledge," and thus essentially non-verbal [...], as in music or dance. In the described set-up, you have made the further determination that what interests you is how extant problems get solved, rather than how new products are forged and new problems get defined [...]. You are simply following a cultural script that tells you what is creative, and all you want to do is to find out how, in detail, processes deemed creative are being carried out in a step-by-step fashion. (In other words, you are exemplifying, but not explaining, creativity).
[...] the goal of defining a retrospective theory of creativity is perhaps a one-sided undertaking; [...] what may equally matter is an action science approach to creativity, where the primary problem is not "Where is creativity," but "How can we transport creativity from here to there?" Of course, these two pursuits are not divergent, but ultimately convergent, issues. (Laske 1992e)
This leads to a clarification of the distinction between "mainstream science" and an "action science":
A mainstream science gives retrospective accounts of creativity as documented or observed on the basis of traces of past processes deemed creative in light of cultural valuation [...]. By contrast, an action science "seeks to enact communities of inquiry in communities of social practice" (Argyris 1985:12). It is dealing with living practitioners, and consequently is based on concurrent reports and on negotiations, unable to detach its practice from that of living practitioners. An action scientist is a "creativity consultant." (Laske 1992e)
Focusing on musical activity (instead of theories of musical artifacts) provided a reinterpretation of the paradigm of "computation" underlying cognitive musicology, where [paradoxically] one attempts to formulate a theory or machine (thus an alloreferential construct) explicating a self-referential activity (Laske 1992a:13).
However, it took some time to realize that the Cartesian notion of knowledge (viz., the separation of knowledge from action and being-in-the-world) that underlies most artificial intelligence research [...] is largely inapplicable to music research. (Laske 1992a:5)
Otto Laske told me that his work had been considerably enriched by Simon's book, The Sciences of the Artificial. (1969), which he read non-stop the day he acquired it. To start with, Simon gave him a substantial theoretical base for a distinction between "sonic" and "sonological" concepts, which may be paralleled with the "etic" and "emic" components of speech.
In music, sonic entities are the elementary sound units isolated by acoustic analysis, while sonological components are equivalent classes constructed on these units by the cognitive-perceptive auditory system. In his introduction to a generative theory of music, Laske identified sonetics as a science of the real versus sonology as a science of the artificial (1975:35-52), the specific agenda of the latter being:
the design of sound artifacts [...] leading to the formulation of performance models of music: (1) Sound Engineering and Sonic Representation, (2) Pattern Recognition and Sonological Representation, (3) Musical Representation and Definition of Intelligent Musical Systems. (Laske 1973a: 354-5)
Sonology, both a grammatical and a strategical, that is, problem-solving theory of music listening (Laske 1992c:217) may be checked against Pierre Schaeffer's (1966) phenomenological analysis of listening.
Making the intention d'entendre the basis of listening, Schaeffer concentrated on the patterns and qualities of listened-to sound as a phenomenon in and by itself, unrelated to the musical grammar and sources of production (op.cit.:218).
According to Laske, Schaeffer's recognition procedure in which the listener makes a transition from a repertory of input sound objects (objets sonores) to a configuration of output musical objects (objets musicaux) seems to imply two different, though interrelated, recognition strategies: listening (associated with sonority) and understanding (associated with musicality) (op.cit.:219). In the domain of listening, Laske draws a parallel between these two strategies and the concepts of performance and competence, claiming that Schaeffer makes sonority, thus listening, a topic of a discipline in its own right, called sonology, while [...] accounting for musicality is a matter of formulating a musical syntax and semantics, in contrast to sonology (op.cit.:220). If so, Schaeffer's theory fails to account for the recognition of sound objects qua musical objects, i.e., how sound and sense come to be related in music.
In sum, a science of musical problem solving requires an intimate understanding of relations between the sonological component and the higher levels of sound organisation termed as "syntax" and "semantics" (Laske 1972:29).
Where sonology goes beyond both the grammatical and performance domain, it is a theory of auditory imagination forming part of epistemology and aesthetics. (Laske 1992c:217)
The coupling of syntax-semantical and sonological components in musical grammars is difficult to implement in computer composition programs. Many software environments exhibit advanced sound synthesis techniques with little scope for structural design. Conversely, compositional software often relies on the popular electronic instrumentation available in the MIDI studio, in which conventional macro events and the twelve-tone system are taken for granted.
When faced with the problem of developing a composition environment (Bol Processor BP2) expanding the boundaries of what is musically possible in both the Indian and western contexts (Kippen & Bel 1994), we opted for a two-fold definition of time-objects (or sound objects): (1) a sequence of elementary actions performed on a music instrument, e.g. a stroke on a drum or a stream of input/output MIDI messages, (2) a set of metric and topological properties relating to proximity constraints. These properties (e.g. "this object may be dilated up to 200% of its basic duration", or "this object shall never be overlapped by the preceding one...") embody time-objects with pertinent traits that make them well-formed in various musical contexts (this pertains to sonority, Laske 1992c:219). As to elementary actions, dealing with both "output" and "input" objects extended the time-object concept to programmed interactions with the outer environment (e.g. synchronisation).
The instantiation of time-objects (which turns them to musical objects) is the result of two processes: first, an arrangement of terminal symbols (a kind of "score") is produced by the grammar; then, a constraint satisfaction (time-setting) algorithm is invoked to shape and locate all time-objects mapped to the terminal symbols found in the score (Bel 1992). These processes combine several complementary aspects of computation: top-down (rule-driven) and bottom-up (data-driven), declarative and procedural, symbolic and numeric, stipulative and interactive, etc. Music software engineering, nowadays, has become more pragmatic -- and less innovative in some aspects. After discovering the Kyma music system, three years ago, Otto Laske told me that his whole theorisation of composition would have been different if he had imagined the environment offered by new technology.
Asserting that artificial intelligence (based on symbol manipulation, i.e. reasoning) may play an essential role in constructing music-understanding systems, Laske questioned the connectionist approach (based on networks of formal neurones) in the light of Schaeffer's paradigm:
Neither acoustics or psychoacoustics, sciences of the real, can explain the intention d'entendre of which Schaeffer, the phenomenologist, had rightly spoken, but for which he could not fully account. It remains to be seen whether connectionism, which construes the intention d'entendre as rooted in human biology, will fare better than Schaeffer in this regard. (Laske 1992c:220)
Historical anthropology of music investigates relations between musical works and the societies in which they were commissioned, produced, performed and appreciated. Its conclusions may often collide with the relatively romantic idea of individual artistic genius determining cultural processes. Laske's contemporary perspective reconciles both views by stating the importance of the task environment in which individual artists are operating. As put by Vecchione (1992:295),
[...] music has become something to produce in agreement with a certain number of protocols, programs, know-hows, questionings, problematics, instilled by designers into the devices they construct (hardware, tools and software). Thus, the task environment implies a certain type of questioning and the formalisation of compositional or musicological problems, sometimes leading to a dialogue between human and machine [...], the music-producing musician and the music being produced, or the machine and the musicologist envisaged as partners [...]
Further (Laske 1990a:55),
The task environment of a computer-assisted composer is more than a set of tools and materials. It is rather a habitat in which the composer lives, and its elements are partners in an ongoing conversation. Much of this environment consist of programs that act as external memories holding musical knowledge; the programs represent objectified thought, without which the composer cannot function. The task environment provides the point of origin of the composer's problem-solving.
As a consequence, a more accurate (and therefore) more abstract representation of music imposed itself. Beyond music as an emotional experience, the idea of music as a form of knowledge, or a form of thought (Vecchione, 1992:295), has progressively emerged:
The computer has changed the potential of music theory, since, for the first time, it has given composers a tool for capturing their processes, and for articulating a theory of music based on their knowledge of compositional planning and problem-solving. This is even the case for composers who never program themselves, but rather use existing computer programs, or no programs at all. (Laske 1989:46)
The structured musical task environments now in existence already form a repertory for musicological research. These systems are the embodiment of musical intelligence in many forms, and while the knowledge bases they incorporate are in most cases not very novel, the way in which they represent and distribute musical knowledge is revolutionary. (Laske 1992a:21)
Laske's motivation for seeing the computer as the artist's alter ego (1990a,1991) is rooted in the foundations of a cognitive anthropology expanding its scope beyond the behavioural study of mental mechanisms.
The program on which computers depend is an artifact that serves our goals; it is written in a language of our invention, and becomes embodied in a physical machine of our design. In an anthropological perspective, such a program is not different from a work of art.
[...]
In an epistemological perspective, of a science of the artificial, what we do thereby is to model ourselves, our knowledge, with the aid of an artifact conceived by us, called a logic machine' or computer.' Writing a program for a computer thus is a metaphorical expression for programming ourselves,' or a part of ourselves, and of our understandings. (Laske 1992d:241)
To the despair of reductionists, Laske addressed the question of mental processes taking psychological and sociological dimensions into account:
In a cognitive perspective, human artifacts are always grounded in an individual way of knowing that is simultaneously a social way. It is the task of the humanities to reconstruct that knowing, either in order to convince of the value of its products, or to better understand them, or both. [...] what engages us are the intentions, designs, images, goals, and strategies of the generative processes that yielded them, which have no natural or factual a priori necessity. (Laske 1992d:240-1)
It is, in fact, the relationship between the (individual and social) processes, on the one hand, and the artifacts, on the other, which is in question in the humanities, and not only the structure of the artifacts (as is believed in aesthetic disciplines), nor only the structure of the mental processes underlying artifacts (as is believed in artificial intelligence). Well understood and utilized, and supported by a reflected epistemological culture, the computing machine, or computer,' can be used as a research tool for validating that relationship which is of such capital importance to the humanities. (op.cit:244-5)
The picture is not complete. The reader should listen to Laske's music and appreciate his poetry. His essay on Charles Olson's projective verse (1990b) reveals a connection between the two sides of his creative genius.
Our common planet has also been occupied by dancers engaged in choreographic research. In a modern world dominated by the craze for "connectivity", it is increasingly compelling to care for innovative explorations of space and time.
Argyris, C. et al. (1985) Action Science. San Francisco, CA: Jossey-Bass.
Bel, Bernard (1992) Symbolic and Sonic Representations of Time-Object Structures. In (M. Balaban et al., eds.) Understanding Music With AI. Menlo Park: AAAI Press, pp.64-109.
Blacking, John (1989) Challenging the myth of 'ethnic' music: first performances of a new song in an African oral tradition, 1961. Yearbook for Traditional Music, 21. New York: International Council for Traditional Music, pp.17-24.
Chomsky, Noam, & Morris Halle (1968) Principles of Generative Phonology. New York: Harper & Row.
Hagège, Claude (1996) Introductory speech. Le Journal du CNRS, January, p.35.
Kassler, Michael (1963) Sketch of the use of formalized languages for the assertion of music. Perspectives of New Music, 1, 2.
Kippen, Jim, & Bernard Bel (1989a) Can the computer help resolve the problem of ethnographic description? Anthropological Quarterly, 62, 3, pp.131-40.
(1989b) The identification and modelling of a percussion "language", and the emergence of musical concepts in a machine-learning experimental set-up. Computers & Humanities, 23, 3, pp.199-214.
(1992) Modelling music with grammars: formal language representation in the Bol Processor. In A. Marsden & A. Pople (eds.) Computer Representations and Models in Music. London: Academic Press, pp.207-38.
(1994) Computers, Composition, and the Challenge of "New Music" in Modern India. Leonardo Music Journal, 4, pp.79-84.
Kowalski, Robert (1985) Logic as a computer language in education. In (L. Steels & J.A. Campbell, eds.)Progress in Artificial Intelligence. Ellis Horwood, pp.71-92.
Laske, Otto E. (1972) On problems of a performance model for music. Institute of Sonology, Utrecht State University (The Netherlands).
(1973a) In search of a generative grammar for music. In Perspectives of New Music, Fall-Winter 1973, Spring-Summer 1974 (double issue), pp.351-78.
(1973b) Toward a musical intelligence system: OBSERVER. Numus West, 4.
(1975) Introduction to a Generative Theory of Music. Sonological reports 1b, Institute of Sonology, Utrecht State University (The Netherlands).
(1977) Music, Memory and Thought: Explorations in Cognitive Musicology. Ph.D Carnegie Mellon Pittsburgh University. Ann Arbor (MI): University Research Press.
(1989) Composition Theory: an enrichment of Music Theory. In (M. Leman, ed.) Models of Musical Communication and Cognition, Interface, 18, 1-2, pp.45-59.
(1990a) The Computer As The Artist's Alter Ego. Leonardo, 23-1, pp.53-66.
(1990b) On the projective purpose of the act of verse: composition by field. ISTAR seminar on artistic creativity, Marseille (France), November.
(1991, guest ed.) Composition Theory. Interface, 20, 3-4.
(1992a) Artificial Intelligence and Music: A Cornerstone of Cognitive Musicology. In (M. Balaban et al., eds.) Understanding Music with AI. Menlo Park (CA): The AAAI Press, pp.3-28.
(1992b) The OBSERVER Tradition of Knowledge Acquisition. In (M. Balaban et al., eds.) Understanding Music With AI. Menlo Park (CA): The AAAI Press, pp.258-89.
(1992c) A Search of a Theory of Musicality. In (D. Penrose & R. Lauzzana, eds.) Languages of Design, 2. Amsterdam: Elsevier, pp.209-28.
(1992d) The Humanities as Sciences of the Artificial. In (B. Vecchione & B. Bel, guest eds.) Music and Sciences and Technologies: New Inquiries, New Connections. Interface, Journal of new Music Research, 21, 3-4, pp.239-55.
(1992e) Creativity -- Where should we look for it? Abstract submitted to the AAAI Spring Symposium on Artificial Intelligence and Creativity.
Lath, Mukund (1988) The 'Modern', the 'Traditional', and Criticism in the Indian Classical Tradition. Sangeet Natak, 89-90, pp.5-15.
Schaeffer, Pierre (1966) Traité des objets musicaux. Paris: Seuil.
Simon, Herbert A. (1969) The Sciences of the Artificial. Cambridge (MA): The MIT Press.
Vecchione, Bernard (1992) La recherche musicologique aujourd'hui -- Questionnements, Intersciences, Métamusicologie. (Followed with commented abstract by O. Laske) Interface, 21, 3-4, pp.281-322.
Zarrilli, Phillip (1987) Re-membering Performance. UCLA Journal of Dance Ethnology, 11, pp.4-21.