Peter Desain and Henkjan Honing: Music, Mind and Machine: Studies in Computer Music, Music Cognition and Artificial Intelligence.Thesis Publishers, Amsterdam. 330 pages, ISBN 90-5170-149-7 softcover.

Reviewed by Jason D. Vantomme, Evanston, Illinois USA for Computer Music Journal [in press].

Music, Mind and Machine is divided into four major sections. The first part provides an overview of the text as a whole research body. The second and third represent Desain and Honing's work on the perception and representation of musical time and temporal structure. The last section provides a sampling of the methodologies used by the authors to express their research in terms of computer programs.

Unlike many other compilations or anthologies that exist in the research field of computer music, Desain and Honing's text provides a clearly crafted introduction that details where the two researchers stand on the important issues involved in their research.

In their overview, the researchers show that their work relies on three important domains: music, the human mind, and machines. The first of these, music, is clearly the driving force of their work; that is, methodology and research in the service of understanding music rather than music as a vehicle for proof of methodology. Desain and Honing approach their musical research as objectively as possible. In order to do this, they are forced to ignore issues such as "sociological factors, fashion, the listener's associations and a multitude of other factors studied in sociology, anthropology and ethnomusicology." (pg. 3) However, they do not immediately discount the possibility that their approach is incorrect, and even confront its merits in an objective manner. This inclusion of self-criticism is refreshing and found all too seldom in other musical research. (It should also be noted that Desain and Honing are by no means hesitant about harshly critiquing the work of others, including studies that are often regarded canonical.)

The second domain that is essential to Desain and Honing's research is that involving the study of the human mind. The two authors immediately begin summarizing their views on the field with an attack on the use of metaphorical concepts (based on intuition) as an explanation of the musical mind. One example of this "metaphorical stretching" is the exploitation of the similarity between musical motion and physical motion as an explanation of musical thought. More specifically, physical energy is often equated with the notion of musical energy; the more performer energy, the greater the musical energy (communication of emotion) experienced by the listener. Thus, the metaphor becomes more than an analogy.

For this particular case, supporting evidence to the contrary is based not only on the breakdown of the solution-by-metaphor approach, but also on their own work concerning quantization and tempo tracking. They reason that "if [their approach] can be shown to hold, it will be a much more attractive explanation than the physical motion theory because it explains properties of good music performance directly from the musical material and from the perceptual processes themselves." (pg. 5) It is precisely such rationalization that makes Desain and Honing's work palatable to a larger research community including those who do not ordinarily draw on cognition or artificial intelligence methods.

The last major domain to play an integral role in Desain and Honing's work is that of the machine. The dichotomous nature of artificial intelligence (AI) has always left somewhat of a gap between what is considered an accurate representation of a human thought process (model) and that which simply provides the same result given similar input (system). Desain and Honing adhere predictably to the former and as such, strive to construct models of the human processes responsible for the perception of musical time and temporal structure.

One of the methodologies employed by the two researchers to express their models in terms of computer programs is known as "microworlds." Rationale for this approach is based on the notion that "good" AI involves an "adequate programming style, a clear description of these issues and the publication of the program in the form of a micro version." (pg. 6) This approach has also been embraced by other prominent researchers as a solution to the extensive reverse engineering of large programs often undertaken to verify their behavior. Desain and Honing's overview of the emergence and advancement of the microworld methodology is both comprehensive and objective.

Providing a review and/or critique of an anthology of already published works from various sources has certain disadvantages. Reading a summarization and critique of each article and its included research would be tedious, to say the least. However, it is important that exceptional works in the collection receive due credit. In the case of Music, Mind and Machine, these works are also those that will provide a vast resource of information and knowledge not only to researchers in music cognition, but also to those developing real-time performance systems, analysis systems, etc. Note that the intention is not to detract from the other articles in the collection; in fact, these "other" works are, for the most part, the most detailed and in-depth.

The first article that deserves mention is "Tempo curves considered harmful," the story of M (the mathematician), P (the psychologist) and their friend the pianist. As Desain and Honing talk the reader through M and P's pursuit to impress the pianist with their computational modelling of rhythmic expression, they also provide the basis (and a vocabulary) on which research in musical rhythm and temporal expression is based. These key points and terms are interspersed throughout the narrative text as large glossary-like blocks. This friendly and informative article should be required reading for all those dealing with issues in musical time.

The second article to highlight is "The quantization problem: traditional and connectionist approaches." Like the previously mentioned article, this paper provides a concise overview of past approaches to the quantization of time intervals. Included in this survey are explanations of inter-onset interval (IOI) quantization, onset quantization, various tempo tracking methods, as well as their own connectionist quantization method. The article also provides a brief microworld example capable of performing IOI quantization as well as tempo tracking with and without confidence-based adjustments.

"A connectionist and a traditional AI quantizer, symbolic vs. sub-symbolic models of rhythm perception" by Peter Desain is an instructive article for those interested in learning about the differences between artificial intelligence solutions based in traditional symbolic methods and those based on sub-symbolic models (connectionist). The examples for comparison are the venerable Longuet-Higgins's "Musical Parser" (symbolic) and Desain and Honing's own connectionist quantizer (sub-symbolic). Important to the success of this comparison is a summary of differing terminology as well as a comparison of the two methods in terms that are common to each.

Henkjan Honing's paper "Issues in the representation of time and structure in music" provides a summary of its subject along a path from the general to the specific. Honing begins his article with a description of the several fields that were affected by the notion of musical representation. A discussion on the most common approaches to representation flows into the more detailed aspects of representing musical time and structure in general. The importance of this article is found in Honing's ability to present the problem of musical representation within the context of knowledge representation research in general.

The final paper mentioned here is Peter Desain's "LISP as a second language: functional aspects". As a LISP programmer who has read enough code to distinguish good LISP authoring from correct LISP authoring, I can appreciate the cleverness of Desain's examples and descriptions (and even be humbled into re-evaluating my own coding practices). Music researchers who find themselves turning to LISP as their language-of-choice would do well to study this article carefully. In fact, this article should likely be read before those in the text that use LISP code and microworld examples as a method of explanation. While some of the solutions presented in this article might easily be replaced by the now widespread Common LISP Object System (CLOS), the understanding of Desain's solutions are applicable and crucial.

Music, Mind and Machine: Studies in Computer Music, Music Cognition and Artificial Intelligence is a text of broad-reaching influence. Though its main concerns deal with the various issues associated with musical time, Peter Desain and Henkjan Honing have managed to show the application of their work to many domains of musical research. For those who are not familiar with the work of these two authors and wish to learn a great deal about music cognition, this text is an essential read. For those who are familiar with Desain and Honing's research, Music, Mind and Machine provides a coherent road map to their important research contributions.