Vol. 27 Issue 2 Reviews
David Temperley: The Cognition of Basic Music Structures

The MIT Press, 2001, ISBN 0-262-20134-8, hardcover, 360 pages, illustrated, notes, references; available from The MIT Press, Five Cambridge Center, Cambridge, Massachusetts 02142-1493, USA; telephone (+1) 800-356-0343; electronic mail mitpress-order@mit.edu; World Wide Web mitpress.mit.edu/.

Reviewed by Ian Whalley
Hamilton, New Zealand

Cover notes for this book state that the author sets out to address a fundamental question about music cognition: “How do we extract basic kinds of musical information from music as we hear it?” David Temperley takes a computation approach and develops models for generating aspects of musical structure based on “preference rules.” These are “criteria for evaluating a possible structural analysis of a piece of music. A preference rule system evaluates many possible interpretations and chooses the one that best satisfies the rules.”

Chapters two to seven cover the six models applied to metrical structure, melodic phrase structure, contrapuntal structure, pitch spelling and the tonal-pitch-class representation, harmonic structure, and key structure. The second half of the book argues that “preference rule systems are not merely valuable as proposals for how musical structures are inferred, but also shed light on other aspects of music” (p. x). This begins with a discussion of revision, ambiguity, and expectation as aspects of musical experience. Chapters nine and ten explore the application of metrical, harmonic, and key models to aspects of rock music and the metrical and phrase structures of African music. Chapter eleven proposes a framework for describing musical styles and looks at how preference rules might apply to composition and performance issues. The final chapter considers issues like motivic structure and musical tension in terms of applying preference rules to higher levels of musical meaning and structure.

The text is clearly written and suitable for general music readers who are literate in Western music notation. Rather than an accompanying CD/CD-ROM, there is a Web site that presents computational examples and implementations given in the book (www.link.cs.cmu.edu/cbms/).
The questions a lay person interested in the topic might ask on reading the publicity notes are: How is this approach different from established methods? How does one test/validate the system against the process people use when hearing music? And how far can the system be generalized to musical styles other than the Western “classical” music on which the tests were based?
Mr. Temperley presents a substantial and detailed real-time computational approach to music knowledge, leaving the impression that a more apt title for the book might have been “Preference Rules and Basic Music Structures.” In contrast, although many significant issues are raised in applying these to music cognition, the experimental design largely results only in raising a number of worthy speculations based on the assumptions made. The problem, in the face of the task set out to be achieved, lies in combining a computational approach to music knowledge within an experimental approach based in music cognition.

The introductory chapter outlines and freely acknowledges the limitations of the experimental design. The psychological feedback for the models proposed is based largely on the author’s intuitions as to the correct analysis of musical pieces and the assumption that most literate people would reach similar conclusions. The psychological validity of the models then only “provides a promising hypothesis about the cognition of basic music structures which warrants further investigation and study” (p. 2). Moreover, the application of the preference rule system is put forward as a representation of aspects of music literate people perceive in real-time, but which at the same time applies the approach to music and musical styles (p. 292) as examples, a “sleight-of-hand” freely acknowledged by Mr. Temperley. The argument is extended to stating that the preference rule system outcomes are reflected in musical styles (p. 297), but the hypothesis is not tested.

Between computation and cognition then, Mr. Temperley attempts to model an experience without a guarantee of how it works in cognitive reality, requiring a leap of faith in agreeing with his intuitions of how music is received. Experience suggests, however, that music cognition is not uniform between people or occasions. “Hearing” might depend on why one is listening, the medium through which music is transmitted, the knowledge and experience brought to the situation, and one’s primary instrument. Besides, music reception is not simply a matter of cognition but also of physiological and emotional response.

Within a computational/formalist/music-knowledge approach, the text impresses with extensive qualitative testing and implementation of preference rules. The system is largely based on, and extends, Fred Lerdahl and Ray Jackendoff’s A Generative Theory of Tonal Music (Cambridge, Massachusetts: The MIT Press, 1983) (GTTM). This is implemented using dynamic programming techniques, although there is again no experimental evidence on this technique’s relationship to psychological reality (p. 19). It is put forward as an area worth exploring.

The input mechanism for many of the tests is based on the piano roll approach, dislocating timbre from performance in context. Mr. Temperley admits the problems of encoding timbre, but largely avoids the criticism in his experiment design. The solution presented is clean computationally but not so convincing from a reception perspective, as it removes musical engagement as a by-product of the expressive manipulation of timbre in performance.

The primary contribution of the first section of the book lies in the computational contribution in some areas covering new ground in extending GTTM. There are, however, limitations, as in the melodic phrase model that seems to require more extensive implementation in contrast to other sections. Furthermore, aspects of the output of the preference rule models are not compared with recent research that may have performed tasks as well or better. For example, work on beat tracking, some of which predates Mr. Temperley’s work, would have provided a more valid comparison to the results in the text.

The second section of the book, largely speculative in nature, is refreshing and stimulating, partly because the questions arising have been so clearly outlined, and it is such a brave leap into the unknown. Chapter eight argues that continuing work on the preference rule framework, with modifications, is worthy of pursuit as a hypothesis about music cognition (p. 205). Examples of expectation, ambiguity, and revision in real-time are provided and applied. The approach opens many possibilities for further work.

Chapter nine is a discussion, rather than a computer implementation, of how aspects of the modified rules could be applied to rock music meter, harmony, and key. Mr. Temperley also speculates that both key and harmony models could be improved in application here by removing the syncopation information in pieces, so that works represent more directly the underlying structure. The chapter is extended to discuss how key could be determined even though different modes are commonly used in rock music.

The author also argues that cognitive principles are found to operate across styles (p. 237) without claiming that the principles are universal or innate. The main problem with this approach is again the difference between cognitive experience and experimental method. Using piano roll representation and removing timbre and syncopation information from rock music removes it from cognitive experience as a primarily aural art: resulting scores being only approximations of aspects of musical knowledge.

A similar problem arises in the next chapter, discussing the application of models to meter and grouping in African music, a limitation that is acknowledged in the text (p. 290). The analysis method used forces a Western music notation approach on the source material, and the evidence is drawn mostly from transcriptions by ethnomusicologists. Any model of the cognitive experience would negate large parts of the semiotic experience in such a cultural context; the similarities between African and Western basic structures may then be in part a product of the experimental design.

Chapter eleven illustrates how preference rule models may reveal elements of composition, style, and performance. The idea is to rate musical works with various models on a continuum of best fit between disorder and perfection, suggesting composers could optimize works between these extremes to produce music within a selected style. The difference between style and idea remains to be explored here.

The text ends with a discussion of functions and infrastructure. This centers on how the author’s models of basic structures affect higher ones such as motivic structure (p. 326). The relationship is dealt with in some detail, covering aspects of the current literature in the field. The discussion ends with a number of hypotheses on the recognition of motivic relationships between segments. My hesitation here is again on the cognitive aspects of the reception and its individualistic nature. Linking patterns and relationships in music can differ greatly between individuals, and particularly between composers and performers.

Make no mistake: this is a detailed, solid, lively, and thought-provoking book that is at times epic in scale. For those interested in a computational approach to musical knowledge, it extends past work, outlines various problems that need to be solved, and provides a benchmark for others to better. For those interested in music cognition, it outlines many of the current problems in the field in a challenging and frank way. The application of the models proposed to other aspects of music and style covers new ground that will provide the basis for continuing discussion. The limitations are in connecting the computational approach to experimental cognitive evidence and the assumptions in doing this inherent in the research design. However, the speculative framework presented here provides an extensive map for future work.