Virtual Music (review)

David Cope's newest book, Virtual Music, arose out of a weekend of papers, panels, concerts, and discussions devoted to Mr. Cope's extraordinary Experiments in Musical Intelligence (EMI) software. EMI is a program which, when given a set of compositions by a particular composer (or composers) as input, attempts to autonomously compose new pieces in the style of the source music. The weekend's events included a highly distinguished panel of presenters, including Mr. Cope, Douglas Hofstadter, Eleanor Selfridge-Field, Bernard Greenberg, Steve Larson, Jonathan Berger, and Daniel Dennett, each of whom also contributed at least one chapter to the book.

Virtual Music is divided into four parts. The first provides a context for EMI by giving a historical overview of algorithmic composition, and also includes an informal description of the mechanics of EMI (given by Mr. Hofstadter). The second part provides a detailed "case study," demonstrating the composition of an EMI work from beginning to end. The third part consists of scholarly evaluation and commentary on the program. The concluding section includes Mr. Cope's response to the criticisms offered by the other scholars, as well as his speculations on the directions the software might take in the future. The book's multiple appendices contain generous amounts of musical examples, and an audio CD of EMI compositions is included as well.

Mr. Hofstadter's segment in the first part of the book is one of the highlights, both for his excellent overview of EMI and for the humorous manner in which he is able to raise some of the troubling philosophical questions that the software provokes. Mr. Hofstadter reduces the essential operation of the program (when given a set of pieces to operate on) to two actions: 1) chop-up, then 2) reassemble. In the "chop-up" phase, the program dynamically segments the input music into meaningful units (on several levels of hierarchical structure). This is no easy task, as the musical material must be chopped up finely enough so that the end result doesn't overly resemble the original sources, but not so fine that the musical coherence becomes lost. In the "reassemble" phase, the program constructs a piece of music by recombining the fragments, attempting both to create a coherent flow on the local level and to ensure that the global patterning of fragments resembles that of the source music.

A number of underlying principles guide the program through these two main stages. The local flow is achieved in part through strong voice-leading rules. As Mr. Hofstadter describes, "Imagine that we have just inserted a fragment f1, and are considering whether to insert fragment f2 right after it, drawn from somewhere [else] in the input." EMI's voice-leading rules stipulate that "the initial note of the melodic line of fragment f2 should coincide with the next melodic note to which fragment f1 led in the original context. In other words, a given fragment's melodic line should link up smoothly with the melodic line of its successor fragment."

The succession of fragments is also guided by a framework of "tension-resolution," which EMI quantifies by attaching one of the letters S, P, E, A, or C to the fragment. The letters stand for Statement, Preparation, Extension, Antecedent, and Consequent. This framework attempts to capture where on the tension-resolution continuum the fragment is situated. EMI determines the appropriate label for a given fragment by examining such things as the level of dissonance in the sonority as well as the metrical placement of the fragment. The software attempts to determine the tension-resolution status of a fragment not only on the local level but also on multiple hierarchical levels (the level of the phrase, of the period, of the section, etc.). Mr. Hofstadter sums up the core local and global processes of EMI as follows: 1) "Sequential assembly of fragments that...