Cope discusses the results of his own efforts toward building an intelligent program capable of composing music in different styles. The basic idea behind his work is the possibility of a top-down approach to musical style analysis, parsing music being similar to techniques used in language studies. Although he modestly denies being the first to suggest the possible use of augmented transition networks (ATNs), one of the most used tools in computational linguistics, in the re-creative process of musical composition, Cope claims that his system is the first implementation of this idea. The author focuses his attention more on the procedural aspects of building a musical pattern matcher than on the expert system component, which is supposed to encompass knowledge acquired during Cope’s “twenty-three years of teaching musical theory.” Some interesting hints are given on the use of the program not only to replicate musical styles but as a valuable tool to assist a human composer.

In the first chapter, Cope gives a brief history of automated music composition, starting with Pythagoras, who was the first to state the intimate relationship between mathematics and music. Cope relates his work to previous experiences in developing mathematical models and techniques, in designing computer programs to both analyze musical style and generate compositions. The emphasis is on recent research, including commercially available programs. An introduction to his concept of EMI, or “experiments in musical intelligence,” is provided.

The second chapter is devoted to musical style representations. Acknowledging that defining and quantifying musical style is difficult, Cope narrows the definition of musical style to a collection of “identifiable characteristics of a composer’s music which are recognizably similar from one work to another.” Among these characteristics, he includes pitch and duration, timbre, dynamics, and nuance; his work focuses on the first two. Several pages are devoted to SPEAC, an ATN system developed by Cope in 1985, which provides one of several layers of abstraction for describing musical motions, harmonies, and motifs. The author also introduces the definition of a signature, one of the key terms of his book, as “a set of contiguous intervals found in more than one work by the same composer.”

Since the system is implemented in LISP, a small chapter is intended to be a LISP primer. Cope continues by describing an elementary LISP composing program in some detail. It would be difficult for someone without knowledge of LISP, or at least some understanding of nonprocedural language fundamentals, to grasp LISP’s expressiveness by reading Cope’s notes. Readers without the proper background would have to refer to one of the many LISP teaching manuals mentioned in the rich bibliography, unless they wanted to get entangled in a plethora of LISP functions described in detail, which tend to cloud the main purpose of this chapter, namely to understand the mechanism that enables Cope’s system to build a fresh two-part invention by using signatures derived from two similar works by Bach.

In the next chapter, the author provides several more examples of EMI’s “sophistication.” By coding at least a  couple  of musical pieces, listing all the patterns found in those works, discovering signatures, and applying ATN techniques to superimpose individual style traits on general rules for music composition, Cope’s algorithm is capable of generating a sonata in Mozart’s or Prokofiev’s style, a Joplin rag, a Chopin mazurka, or even traditional Indonesian music. Readers are expected to have the musical background not only to understand the musical notation but to be able to compare EMI’s output to the original works. This requirement may seriously narrow the intended audience.

This remark does not mean that readers who are laypeople with regard to either artificial intelligence applications or classical music (but have some understanding of the other area) should not attempt to read this book, which is the interesting result of many years of work. On the other hand, readers should not get over-optimistic and think that programs are now able to compose works like Mozart’s masterpieces.